|USFDC Home | USF Electronic Theses and Dissertations||| RSS|
This item is only available as the following downloads:
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 22 Ka 4500
controlfield tag 007 cr-bnu---uuuuu
008 s2010 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0003319
An analysis of the influence of sampling methods on estimation of drug use prevalence and patterns among arrestees in the united states :
b implications for research and policy
h [electronic resource] /
by Janine Kremling.
[Tampa, Fla] :
University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains X pages.
Dissertation (Ph.D.)--University of South Florida, 2010.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
ABSTRACT: Using data from the Drug Use Forecasting (DUF) and the Arrestee Drug Abuse Monitoring (ADAM) programs collected by the National Institute of Justice the question whether the drug estimates of DUF, using a non-probability sample, and the drug use estimates of ADAM, using a probability sample, yield substantially different results will be explored. The following main questions will be addressed using equivalence analysis: Are there substantial differences in the DUF and ADAM samples with regard to the drug use information obtained from arrestees at nine sites across the United States? The analysis suggests that the drug use information contained in DUF and ADAM is not substantially different for marijuana, cocaine, and opiates for all sites analyzed together. Additionally, there are no substantial differences for seven of the nine sites. The implications of these findings are discussed.
Advisor: Tom Mieczkowski, Ph.D.
Drug Use Estimates
t USF Electronic Theses and Dissertations.
An Analysis of the Influence of Sampling Methods on Estimation of Drug Use Preval ence and Patterns Am ong Arrestees in the United States: Implications for Research and Policy by Janine Kremling A thesis submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Criminology College of Arts and Sciences University of South Florida Major Professor: Tom Mieczkowski, Ph.D. John Cochran, Ph.D. Chris Sullivan, Ph.D. Kim Lersch, Ph.D. Sondra Fogel, Ph.D. Date of Approval: February 12, 2010 Keywords: Equivalence Testing, Sampling, Methodology, Drug Use Estimates, Comparative Analysis Copyright 2010, Janine Kremling
Dedication This dissertation is dedicated to my fa mily, friends, and Professors who have supported me all the way and without whom I could not have completed my degree.
i Table of Contents LIST OF TABLES ................................................................................................................. .v LIST OF FIGURES.............................................................................................................vii ABSTRACT....................................................................................................................... .viii CHAPTER ONE: INTRODU CTION AND OVERVIEW OF THE STUDY........................1 Statement of the Problem............................................................................................1 Main Purpose of DUF and ADAM.............................................................................6 Customizing Programs to the Community..................................................................7 Purpose of the Study...................................................................................................8 Significance of the Study............................................................................................8 Uniqueness of the Current Study................................................................................9 Brief Overview of Equivalence Testing...................................................................12 Possible Outcomes of the Analysis...........................................................................13 CHAPTER TWO: RESEARCH ON ILLICIT DRUG USE.................................................16 Why is it Important to Study Illicit Drug Use...........................................................16 Budget Spent on Combating Illicit Drug Use...............................................16 Policies and Programs Combating illicit Drug Use......................................17 Importance of Studying Drug Use for Researchers......................................18 Ongoing National Studies of Drug Use Prevalence and Patterns: Why ADAM Should Be Implemented Nationwide......................................18 The National Survey on Drug Use and Health (NSDUH)............................19 Monitoring the Future (MTF).......................................................................23 Drug Abuse Warning Network (DAWN).....................................................25 Advantages of DUF and ADAM Over Other National Surveys...................27 Validity of Self-Report Data.........................................................................29 Explanation for the Failure to Report Behaviors Truthfully.........................32 Implications of the Lack of Reporting for Researchers and Policy Makers...........................................................................................................35 Validation Methods for Self-Reported Drug Use.........................................37 Urine Testing................................................................................................37 Localized Data Collected Every Quarter Over 15 Years..............................39 Changes in Drug Use Prev alence and Pattern s between 1988 and 2002........................................................................................................44 Drug Court Movement..................................................................................48 Drug Treatment in Correctional Facilities....................................................49 Data for Local Police Agencies....................................................................50 Educational Attainment and Drug Use.........................................................51
ii Lack of a National Study Monitoring Drug Use Among Arrestees..............52 CHAPTER THREE: METHOD...........................................................................................55 Overview of the DUF Study.....................................................................................55 Underlying Assumptions of the DUF Study.................................................55 Data Collection in the DUF Study................................................................56 Number of Arrestees.....................................................................................57 Description of Variables...............................................................................58 Response Rate...............................................................................................59 Drug Testing.................................................................................................60 DUF Sampling Procedure.............................................................................60 GAO Report Criticisms.................................................................................61 (1) Criticism on the Sel ection of Booking Facilities........................62 (2) Criticism on the Subject Sampling Procedure.............................62 (3) Criticism of the Inclusion and Exclusion Criteria.......................63 Overview of the ADAM Study................................................................................64 Sampling Procedures 1998...........................................................................66 Sampling Procedures 1999...........................................................................66 Sampling Procedures 2000 to 2003..............................................................67 The Sites Sampling Design..........................................................................67 Weighting Procedure....................................................................................68 The Facility-Level Sampling Design............................................................70 Face-Sheets...................................................................................................71 Response Rate...............................................................................................71 Drug Testing.................................................................................................74 Sites...............................................................................................................74 Number of Arrestees.....................................................................................75 Description of Variables Redesi gned Data Collection Instrument.............76 Calendar Method...........................................................................................76 Drug Treatment Data....................................................................................78 Methodological Issues of DUF and ADAM.................................................78 Availability of Arrestees.........................................................................79 Determining the Specific Drugs Used Within A Certain Jurisdiction....80 Number of Interviews for Each Quarter.................................................81 Participants as Volunteers.......................................................................84 Limitations of Urine Testing...................................................................85 CHAPTER FOUR: DATA DESCRIPTION AND ANALYTICAL STRATEGY..............88 Major Research Question of the Current Study........................................................88 Data...............................................................................................................88 Which Years are Included in the Analysis....................................................89 Which Variables are Used in the Analysis...................................................91 Introduction to Equivalence Testing.........................................................................94 Traditional Null Hypothesis Testing.............................................................96 .Calculation of the Traditi onal Null Hypothesis Test...........................97
iii Equivalence Testing....................................................................................98 Equivalence Test..................................................................................99 Calculation of the Equivalence Test.............................................99 Possible Outcomes and Interpretation......................................................100 Defining the Equivalence Margin.............................................................102 Threshold Levels for the Major Drugs......................................................103 Marijuana...........................................................................................103 Cocaine..............................................................................................104 Opiates...............................................................................................105 Barbiturates, Amphetamines, and PCP..............................................105 Sensitivity Tests........................................................................................106 .Urine Analysis Test of Marijuana, Dallas..........................................106 Urine Analysis Test of Opiates, Dallas..............................................107 Self-Reported Drug Use of PC P within 7 Hours, Dallas...................107 Power Analysis.........................................................................................107 Analysis of Dallas, Texas Data...............................................................................108 Decision Criteria.......................................................................................112 CHAPTER FIVE: RESULTS.............................................................................................117 Descriptive Statistics...............................................................................................117 Demographic Profile...................................................................................117 Race.....................................................................................................118 Empl oyment........................................................................................120 Education............................................................................................120 Charge Distribution.............................................................................121 Age......................................................................................................121 Drug Use Frequencies.............................................................................................122 Overall Results of the Equivalence Analysis..........................................................127 Site-Specific Findings.............................................................................................137 Dallas..........................................................................................................137 Denver.........................................................................................................141 Indianapolis.................................................................................................144 Miami..........................................................................................................147 New Orleans................................................................................................150 Phoenix.......................................................................................................154 Portland.......................................................................................................157 San Antonio................................................................................................160 San Jose.......................................................................................................163 The Impact of Different Alpha Levels and the Inclusion/Exclusion of Indeterminate Values..............................................................................................166 CHAPTER SIX: DISCUSSION.........................................................................................174 Major Goal and Possible Outcomes of the Study...................................................174 Analytical Strategy of the Current Study................................................................175 Main Findings.........................................................................................................177
iv Overall Finding for All Sites.......................................................................177 Overall Findings by Drug...........................................................................178 Site Specific Findings.................................................................................180 Discussion of the Findings......................................................................................181 Consistency of Findings..........................................................................................182 Explaining the Results............................................................................................183 Implications of the Current Study...........................................................................185 Limitations of the Current Study............................................................................186 Final Remarks.........................................................................................................189 REFERENCES...................................................................................................................191 APPENDICES....................................................................................................................205 Appendix A: Demographic Profile by Site.............................................................206 Appendix B: Drug Use Freque ncies by Year and Site...........................................215 ABOUT THE AUTHOR..........................................................................................End Page
v LIST OF TABLES Table 3.1 Number of Male Arre stees and Sites by Year DUF 57 Table 3.2 Cut-off Levels and Detection Periods for 10 Drugs 74 Table 3.3 Number of Inte rviewed Arrestees and Numb er of Sites by Year ADAM 75 Table 3.4 Sample Size for Each Quarter by Site and Year 83 Table 3.5 Comparison DUF and ADAM 87 Table 4.1 Codings and Descriptio ns for Demographic Variables 91 Table 4.2 Codings and Descript ions for Drug Use Variables 92 Table 5.1 Drug Use Frequencies for the Total Sample Lowest, Highest, and Average Prevalence Rates 126 Table 5.2 Equivalence Test Outcomes by Site 130 Table 5.3 Classification of Drug Use Values Across the Outcome Categories for Each Site 131 Table 5.4 Summary Table for the Di stribution of Drug Use Values Across the Outcome Categories 131 Table 5.5 Equivalence Test DUF and ADAM in Dallas, TX 140 Table 5.6 Equivalence Test DUF and ADAM in Denver, CO 143 Table 5.7 Equivalence Test DUF and ADAM in Indianapolis, IN 146 Table 5.8 Equivalence Test DUF and ADAM in Miami, FL 149 Table 5.9 Equivalence Test DUF and ADAM in New Orleans, LA 153 Table 5.10 Equivalence Test DUF and ADAM in Phoenix, AZ 156 Table 5.11 Equivalence Test DUF and ADAM in Portland, OR 159 Table 5.12 Equivalence Test DUF and ADAM in San Antonio, TX 162 Table 5.13 Equivalence Test DUF and ADAM in San Jose, CA 165
vi Table 5.14 Differences in Outcomes by Using different Alpha Levels and Changes in the Inclusion and Exclusion Criteria 173
vii LIST OF FIGURES Figure 5.1 Distribution of Marijuana Values Across the Outcome Categories 133 Figure 5.2 Distribution of Cocaine Values Across the Outcome Categories 135 Figure 5.3 Distribution of Opiate Va lues Across the Outcome Categories 136 Figure 5.4 Distribution of Variables Across the Outcome Categories in Dallas 139 Figure 5.5 Distribution of Variables Across the Outcome Categories in Denver 142 Figure 5.6 Distribution of Variables Across the Outcome Categories in Indianapolis 145 Figure 5.7 Distribution of Variables Across the Outcome Categories in Miami 148 Figure 5.8 Distribution of Variables Across the Outcome Categories in New Orleans 152 Figure 5.9 Distribution of Variables Across the Outcome Categories in Phoenix 155 Figure 5.10 Distribution of Va riables Across the Outcome Categories in Portland 158 Figure 5.11 Distribution of Variables Across the Outcome Categories in San Antonio 161 Figure 5.12 Distribution of Va riables Across the Outcome Categories in San Jose 164
viii AN ANALYSIS OF THE INFLUENCE OF SAMPLING METHODS ON ESTIMATION OF DRUG USE PREV ALENCE AND PATTERNS AMONG ARRESTEES IN THE UNITED STATES: IMPLICATIONS FOR RESEARCH AND POLICY JANINE KREMLING ABSTRACT Using data from the Drug Use Forecasting (DUF) and the Arrestee Drug Abuse Monitoring (ADAM) programs collected by the National Institute of Justice the question whether the drug estimates of DUF, using a non-probability sample, and the drug use estimates of ADAM, using a probability sample, yield substantially di fferent results will be explored. The following main questions wi ll be addressed using equivalence analysis: Are there substantial differences in the DUF and ADAM samples with regard to the drug use information obtained from arrestees at nine sites across the United States? The analysis suggests that the drug use inform ation contained in DUF and ADAM is not substantially different for marijuana, cocaine, and opiates for all sites analyzed together. Additionally, there are no subs tantial differences for seve n of the nine sites. The implications of these findings are discussed.
1 CHAPTER ONE INTRODUCTION AND OVERVIEW OF THE STUDY Statemen t of the Problem In 1987, the National Institute of Justice (N IJ), in cooperation w ith the Bureau of Justice Assistance (BJA ), implemented a national study tracking drug use prevalence and drug use patterns among arrestees. The program, called the Drug Use Forecasting (DUF) program, started in 12 cities. The DUF pr ogram was unique for three reasons: (1) it collected a urine specimen as a validation technique of self-repor ted drug use; (2) it collected data on drug use prevalence and pa tterns from arrestees, a population group at high risk for drug use that is not studied consistently across the United States; and (3) it provided local data on drug use prevalence and patterns with the explicit goal of providing policy makers with the necessary information to develop programs that effectively reduce drug use (Nationa l Institute of Ju stice, 1998). In 1993, the General Accounting Office (GAO) published a report on the strengths and limitations of the three national drug use studies funded by the Federal Government. One of these three studies was th e DUF program. The report stated that the major shortcomings of DUF were the use of a non-probability sample, more specifically a judgment-based sample, and a lack of standardization across sites. According to the GAO (1993) these two major shortcomings made it impossible for researchers to generalize findings to the population of arrestees in that specific geographic ar ea. As a result, the DUF data was said to be useless for po licy makers, who had as a major goal the development of programs aiming to reduce drug use.
2 After the evaluation by the GAO, NIJ deci ded to implement some major changes to the study design. The most important modification was the decision to change the judgment-based sample to a probability sample. In 1998, the name of the study changed from Drug Use Forecasting (DUF) to A rrestee Drug Abuse Monitoring (ADAM) program and the data collection was now standard ized at all sites, bu t the study still used a judgment-based sample. The probability sample was not fully implemented until the latter half of 1999. Similar to DUF, the ADAM study collected data about drug use prevalence and patterns and used urine analysis as a validation techni que for self-reported drug use. In addition, ADAM also collected data on drug market activity, drug treatment, and other drug-related issues (Natio nal Institute of Justice, 2000). Due to the implementation of a probabil ity sample, the findings of the ADAM program were now said to be representative of the target population of booked arrestees, allowing researchers to genera lize the findings to the genera l population of arrestees for the geographic area at which the study was implemented. However, the ADAM program was only carried out for approximately three years. The federal government terminated the ADAM study in January 2003. The high costs of the program and significant budget cu ts by Congress have been cited as the major reasons for the termination of ADAM by the National Institute of Justice (Yacoubian, 2004). Originally, Congress ha d allocated $20 million per year in discretionary money for social science resear ch, but this research money was reduced to $6 million for the year 2004. Thus, the ADAM program became too expensive to continue. Whereas the DUF program costs a bout 1 million dollars per year, ADAM costs about 8.4 million dollars per year (National Institute of Just ice, 2004). The much greater
3 costs of ADAM were due mainly to the greater number of sites where data were collected, the greater amount of time interviewers spent at the facility su pervised by a correctional officer, continuous training of the interviewers, modifications to the questionnaire (more extensive and detailed than DUF), and changes in urine specimen processing. All of these changes in the pr ogram (from DUF to ADAM) will be described in detail in Chapter Three. Between 2003 and 2006 no drug use data was systematically collected for the population of arrestees. In 2007, the Office of National Drug Control Policy (ONDCP) revived the ADAM program with its proba bility sample. However, the ADAM II program was only implemented at 10 sites th at were previously part of the ADAM program. These 10 sites repres ent individual counties in 10 separate states. This is significantly less than the origin al ADAM program (35 sites) and it is also significantly less than the National Household Survey ( NHSDUH) and Monitoring the Future (MTF), both of which are national studies. These di fferences are important for two reasons: (1) drug use varies by geographic location and (2) a rrestees have significantly higher rates of drug use as compared to the general populat ion, which is being studied by the NSDUH and the MTF. Research has consistently shown that drug use prevalence and patterns vary significantly across geog raphic locations, and as a result studying only 10 counties in the entire country does not provide sufficient da ta for drug using beha viors among arrestees (Feucht and Kyle, 1996; Peters, Yacoubia n, Baumler, Ross, and Johnson, 2002; Riley, 1997; Yacoubian, 2002). For example, the ADAM data itself shows that the prevalence of certain illicit drugs depends on the geogra phic location of the si te. In 2001, drug test
4 results demonstrated that the percent of a rrestees who tested positive for cocaine use ranged from a low of 11.0% in Des Moines, Iowa, to a high of 48.8% in New York. Similarly, the percent of arre stees who tested positive fo r marijuana ranged from 28.5% in Laredo, Texas, to 54.2% in Minneapolis, Minnesota, and for opiates from 2.0% in Omaha, Nebraska, to 27% in Chicago, Illinois (NIJ, 2001). Additionally, illicit drug use al so varies considerably with in states. The percent of arrestees who tested positive for cocaine in Texas varied between 20.4% in San Antonio and 45.0% in Laredo. Also, marijuana use in Texas ranged from 28.5% in Laredo to 40.7% in San Antonio (NIJ, 2001). This result demonstrated that Laredo had the lowest number of cocaine users, but the highest num ber of marijuana users. These within-state differences did not just apply for Texas, but were also ap parent for other states including California, Washington, and Florida. These geographic differences in drug use prevalence and patterns demonstrate the impor tance of collecting da ta from arrestees nationwide. Collecting data for 10 sites (c ounties) is not sufficient to provide a comprehensive overview of drug use prevalence and patterns in the United States because these 10 sites are not representative of drug use for other cities and states. As a result, important changes in drug using behaviors may not be discovered at all or they may be discovered only once they have become epidemic. Second, research has consistently shown th at arrestees have substantially higher rates of illicit drug use th an the general population (NSD UH) and school children (MTF) (BJS, 2004). Brecht, et al. (2003) estimated that about 65% of arrest ees use illic it drugs. By comparison, approximately 8.3% of the general population (as determined by the NSDUH) and about 9.5% of school childre n use illicit drugs (SAMHSA, 2003; 2008).
5 Thus, the NSDUH and the MTF track illicit drug use nationwide for population groups that use illicit drugs at much lower rates than arrestees. It might be more useful to put a greater focus on population groups who use drugs regularly because regular use of illicit drugs results in great costs for society. Illicit drug use is very costly for society for several reasons. First, drug use is related to criminal behavior in three ways : (a) users must obtain money for drugs, (b) drugs may have a detrimental effect on the in dividuals behavior, and (c) as part of the lifestyle and business methods of drug dealer s (NIDA, 1990). For instance, studies have found that the rise in crack cocaine was associ ated with a significan t increase in the urban crime rate, especially violent crimes (Grogger and Willis, 2000, Inciardi, 1990). Additionally, violent behavior has been f ound to be related to the use of other psychoactive substances, such as amphetamines, cocaine, LSD, and PCP (Roth, 1994). Research has also demonstrated that drug us ers engage in criminal activities, such as burglary and drug sales to obtain monies for drugs (Dembo, Williams, Wish, Berry, Getreu, Washburn, and Schmeidler, 1990). Finall y, the lifestyle and business practices of drug dealers include violence as part of the interaction between drug dealers, rival gangs dealing drugs, drug runners, and informants (Goldstein, 1987). Second, drug use contributes significantly to the costs of health care. Specifically, researchers have estimated that almost 50% of all health car e costs are related to alcohol and drug use. French and Martin (1996) provi de an overview of the cost of illicit drug use for society. According to the authors, these co sts can be divided into nine categories. The categories are: medical services costs; pren atal costs; drug abuse treatment costs; drugassociated disease costs; cost of alco hol, illicit drug and mental health (ADM)
6 comorbidity; crime-related costs; foster ca re payments; special education and early intervention costs; and costs of Aid to Families with Dependent Children (AFDC) including food stamps (French and Mart in, 1996, p. 454). These costs can only be reduced if drug use decreases. Arrestees ar e a population group who use drugs at high rates. Therefore it is important to stu dy their drug-using behaviors and implement programs that help decrease their drug use. Third, drug use has been shown to be asso ciated with loss of employment, loss of housing, and family disintegration (NIDA, 1990). Each of these conse quences constitutes and contributes to the costs of illicit drug use for society. For example, a loss of employment can lead to crimin al behavior because individuals who use illicit drugs have to find another way to get money for drugs. Th ey might engage in drug selling or other illegal activities that help them get more drugs. Thus, it is crucial to track drug-using behaviors among arrestees across the nation to be able to implement effective programs. In fact, the National Institute of Drug Abuse (NIDA) suggests that in order to reduce drug use, it is imperative to implement commun ity-based prevention and treatment programs" (Roth, 1994). If the goal is to decrease the costs of drug use for society and implement local prevention and treatment programs, as proposed by NIDA, then a reasonable approach would include arrestees because they have high rates of illicit drug use (65%), and who then, as a result, contribute greatly to the costs of drug use for society. Main Purpose of DUF and ADAM As stated previously, the main purpose of the DUF and ADAM program was and still is (with ADAM II) to guide policy and program implementation at the local level (GAO, 1993). In order to fulfill that purpos e it is necessary to know which drugs are
7 being used, how they are being used, and what distinguishes drug users, that is, arrestees who use illicit drugs, at the di fferent geographic areas within the United States. This information can then be used to develop pr ograms and services targeting specific drug users, and programs can be designed and implem ented that are tailored towards the needs of a certain community or geographic area. Tailoring the programs to the needs of the community is important in order to be effective in reducing illicit drug use. Customizing Programs to the Community Research on drug abuse treatment ha s shown that programs specifically customized for a certain individual are more effective in reducing future drug use and related issues, such as recidivism and infectious diseases (Hammett, Harmon, and Rhodes, 2002; Murphy, Collins, and Rush, 2007) Similarly, programs aiming to prevent the spread of infectious diseases, such as HIV, are also most effective when they meet the needs of the community (Kelly, et al., 1992) Accordingly, it could be expected that programs aiming to reduce illicit drug use among arrestees would be more successful if they would meet the specific requirements of drug users in a certain community. These customized programs might also be more cost effective because they dont waste resources on combating drugs that are not wi dely used. For example, there is no great need to implement programs targeting meth amphetamine users in Florida, because methamphetamine is used rarely in that area. There is, however, a need for such programs in San Diego and other west coast citie s where up to 22% of arrestees use methamphetamine. Collecting data locally but expanding the colle ction sites to be representative of the nation will provide the necessary information to implement customized programs, as a result reducing illi cit drug use, recidivism and the spread of
8 infectious diseases. The major obstacle, it seems, are the costs associated with such a program. Considering that the total economic costs of illicit drug use are estimated to be approximately $143.4 billion each year (Office of National Drug Control Policy, 2001) and that $50 million is spent by the govern ment via NSDUH to survey a population group that rarely uses drugs (8.2%), it would be reasonable to also implement a nationwide program monitoring illicit drug use among arrest ees, 65% of whom use illicit drugs. It is, however, also important to r ecognize current budget cuts and the economic situation. Thus, this study will examine wh ether it is possible to implement a study similar to ADAM that is equally effective but less expensive. Purpose of the Study As described above, the main reason for the change from a non-probability sample (DUF) to a probability sample (ADAM ), and eventually the termination of the study altogether, was the critique of DUF by the Genera l Accounting Office (GAO). A detailed overview of the DUF program and the criticisms of the GAO will be discussed in Chapter Three, but the main conclusion of the GAO was that the sampling procedure of DUF did not allow for the generalization of re sults to arrestees in general because it is unclear whether the information contained in DUF is valid with regard to drug-using behaviors among arrestees w ithin the geographic area s studied (GAO, 1993). To date, no one has systematically a ssessed this question. The purpose of the current study is to examine this issue by comparing the results of the probability sampling used in ADAM to the non-probability sampling of DUF. Specifically, the main research question is whether the non-probability sample of DUF contains drug use information
9 that could be said to be equivalent to the drug use information contained in the probability sample of ADAM. Significance of the Study The proposed research question is important for two main reasons. First, if the analysis indicates that the DUF and ADAM data do not prov ide substantially different information, this might help implement a drug abuse monitoring pr ogram that observes drug use among arrestees nationwide and provide s local data on drug use prevalence and patterns among arrestees. As a result, the research can guide policy development and program implementation aiming to redu ce drug use at the local level. Second, the current analysis is important because the DUF data (data between 1987 and 1999) has only been used by a few res earchers. The results of the current study might enable researchers to publish research findings from the DUF data and from both DUF and ADAM over time. This might be especially important for researchers monitoring (a) the relationship between drug use and crime over time; (b) the popularity of certain drugs over time; (c) the introducti on of new drugs and how they spread around the country; and (d) the rela tionship between newly implem ented drug laws, policies and drug use prevalence and patterns. Uniqueness of the Current Study The current study is also uni que. To date, there is little research comparing the data from two different samples, using diffe rent sampling strategies, with the purpose of exploring whether these two samples contain substantially different information. Although there is no study assessing diffe rences across the DUF and ADAM data systematically, there is some evidence that the information contained in DUF might not
10 be substantially different from ADAM. The National Institute of Justice (NIJ) conducted a study comparing drug use outcomes for one site for the DUF and ADAM data (NIJ, 1990). The study used data from the Uniform Crime Report (UCR) as a base rate to assess whether differences in the charge distribution of arrestees lead to biased drug estimat es. The results demonstrated that drug use estimates did not appear to be biased. Although the demographic characteristics of the sample were differe nt, the drug use information was similar. A second study was conducted in Anchorage, AK with the goal of determining whether the male and female arrestees interviewed were representative of the arrestee population at that site (Myr stol and Langworthy, 2005). For this purpose the authors compared the demographic information of the arrestee sample to the demographic information collected via face sheets from all arrestees, including nonrespondents. The authors found that although female arrestee s were not sampled in accordance with the probability sampling plan for males, they were more representative of the population of booked female arrestees than the male sample was of the population of booked male arrestees. This is notable b ecause, similar to DUF, female arrestees were selected via a convenience sample. Thus, the results of the study from Anchorage suggest that a nonprobability sample might include a similar subs et of arrestees as the probability sample. Both studies only examined one site, however. Additionally, some research has been done examining the equivalence between Internet-based and paper-and-pencil data collection. This re search suggests an overall equivalence between these two methods despit e differences in the demographic profiles of the two samples (Epstein, Klinkenberg, Wiley, and McKinley, 2001; Krantz, Ballard,
11 and Scher, 1997; Pasveer and Ellard, 1998). Th is supports the finings of the NIJ (1990) study which also found that demographic differences in the sample did not necessarily result in substantially different drug use information. Based on these studies, it is possible that the drug use information contained in the DUF sample is not substantially different from the ADAM sample. Thus, the current study attempts to assess the question wh ether the non-probability sample of DUF provides information about drug use prevalence and patterns that is comparable to that of ADAM in a more systematic fashion by examining all sites that have the same catchment area for DUF and ADAM. This study is possible because both DUF and ADAM examined drug use prevalence among arrestee s. Although ADAM had more sites and a more comprehensive interview instrument, th ere are nine sites that have the same catchment area and contain 14 variables for sel f-reported drug use and urine test results for the major drugs (i.e., marijuana, cocaine and opiates), thus providing the necessary information for this assessment. This study is also unique because it uses equivalen ce testing to examine the proposed research question. Equivalence means that there are not substantial differences (Rogers, 1993). Equivalence testing is an an alysis strategy widely used among clinical researchers to assess whether two differ ent drugs/treatments produce a comparable outcome. Equivalence testing assumes that two different drugs/treatments will always result in some differences, but these differences might not be of prac tical and/or clinical importance. Similarly, it can be assumed that two different samples will result in different outcomes. The crucial question, and the significant research question of the current study, is whether or not the sampling design and procedures used by ADAM resulting in an
12 approximation of a true probability sample produced results that are substantially different from DUF. Equivalen ce analysis is applied to asse ss this research question. Brief Overview of Equivalence Testing The method employed in this study is the confidence interval method first described by Westlake (1981). Rogers et al. (1993) introduced the method to the field of psychology. The main idea of this method is to calculate confidence intervals for the proportions for the DUF and ADAM drug estimates and conduct a traditional hypothesis test and an equivalence test simultaneousl y. The outcome of these two tests will show whether the DUF and ADAM data are substa ntially different or whether they are sufficiently similar to be considered equiva lent. Substantially diffe rent means that drug use estimates contained in DUF and ADAM are statistically different in the traditional null hypothesis test and not equi valent in the equivalence test. Overall, substantially different is defined as a difference of 20% or more between the drug use estimates of DUF as compared to ADAM. Equivalence is present if the drug use estimates are statistically significant in the equivalence test and not significant in the traditional null hypothesis test. Equivalence does not mean exactly the same rather it means the absence of a meaningful difference (Eur opean Medicines Agen cies, 2000; Rogers, Howard, Vessey, 1993; Allen and Seaman, 2006 ; Tryon and Lewis, 2009). Thus, overall, equivalence is said to exist if the difference between the drug use estimates in DUF and ADAM is less than 20%. The exact method is described and demonstrated on an example in Chapter Four.
13 Possible Outcomes of the Analysis Four possibilities exist with regard to the outcome of the current analysis: 1) The drug use estimates of the DUF and ADAM samples are equivalent (Eq). 2) The drug use estimates of the DUF and ADAM samples are different (D). 3) The drug use estimates of the DUF and ADAM samples are different and equivalent (D&Eq). 4) The drug use estimates of the DUF and ADAM samples are not different and not equivalent. They are stat istically indeterminate (ND&NEq). Possibilities one and two are fairly stra ightforward. First, the drug use estimates of the DUF and ADAM samples can be said to be equivalent if the drug use proportions are statistically equivalent and not statistically different Second, the drug use estimates of the DUF and ADAM samples can be said to be different if the drug use proportions are statistically different and not statistically e quivalent. Third, the drug use estimates of the DUF and ADAM samples can be said to be equivalent and different if the drug use proportions are statistically di fferent but statistically equi valent. In this case the researchers suggest that the data might not be substantially different. Rather the differences can be said to be trivial (Allen and Seaman, 2006; Rogers et al. 1993). Fourth, if the drug use estimates of the DUF and ADAM samples are not statistically different and not statistically equivalent no conclusions can be drawn with regard to the question whether there exist substantial differences. The results would be ruled indeterminate (Rogers, et al. 1993; Tryon and Lewis, 2009). For an easier overview Figure 1.1 demonstrates the four possibilities.
14 Figure 1.1. Possible Outcomes of the Analysis Equivalent No Yes Yes 2 (D) 3 (D & Eq) Different No 4 (ND & NEq) 1 (E) Before the question about the equivalence of the DUF and ADAM data can be assessed, it is critical to examine why it is im portant to study illicit drug use in general and why ADAM and DUF were crucial for re searchers, communities, and policy makers in their assessment and efforts to reduce drug use. Thus, Chapter Two discusses the importance of studying illicit drug use in ge neral, followed by a description of the importance of the DUF and ADAM program s for communities, policy makers and researchers, and why the original terminati on and subsequent revival of the study with only ten sites in 2006 constitutes a major loss an d an obstacle to the goal of reducing drug use. Chapter Three lays out the methodology of DUF and ADAM and discusses the major criticisms of the GAO (1990) on the DUF program and the specific changes that were made by the National Institute of Ju stice to improve the program and develop a dataset that could be generali zed to the greater population of booked arrestees within the geographic area studied. Chapter Four explains the analytical plan for the research question and the statistical analysis empl oyed and a specification of the variables included in the current study. Chapter Four also presents the descrip tive statistics for the demographic characteristics of the DUF a nd ADAM samples and drug use information. Chapter Five presents results for the equiva lence analysis determining the comparability
15 of the DUF and ADAM data with regard to drug use prevalence and patterns. Finally, Chapter Six concludes by summarizing and di scussing the results of the analysis and providing implications of the results for future research.
16 CHAPTER TWO RESEARCH ON ILLICIT DRUG USE Why is it Important to Study Illicit Drug Use? Budget Spent on Combating Drug Use Illicit drug use is a problem of high prior ity in the United States (Reuter, 2006). One indicator of the importance of reducing il licit drug use is the amount of monies spent by the government on decreasing illicit drug use as compared to other expenditures. The Office of National Drug Policy (ONDCP) estimat es the costs of drug expenditures of the federal government. Until 2002, the ONDCP combin ed drug-targeted (e.g., domestic and international enforcement) and drug-related (e.g., prevention and treatment) expenditures in their drug expenditure estimation. The drug-related e xpenditures also included substance abuse and rehabili tation research (ONDCP, 2002). Using this comprehensive approach, the ONDCP estimated that the government would spe nd $19.2 billion on the national drug control budget for the fiscal year 2002 ( ONDCP, 2002). This estimate decreased to $12.9 billion when expenditures as sociated with the consequences of drug use (e.g., cost of incarceration) were excluded (ONDCP, 2002). For the year 2010, the federal government has provided a budget of $15.1 billion (ONDCP, 2009). This budget includes funding for treatment, prevention, domestic law enforcement, interdiction, a nd international counterdrug supp ort. In comparison, the U.S. Department of Justice is allocated a to tal budget of $26.5 billion, and the budget for the U.S. Department of Education is $46.7 bill ion. Also, the budget allocated to combating illicit drug use is greater th an the total budget for the U.S. Department of Commerce,
17 which receives only $12 billion. In sum, th e federal government allocates a significant amount of money to re duce illicit drug use. In addition to the fact that the gover nment spends a considerable amount of money on combating illicit drug use as compared to other expenditures, these expenditures have significantly increased with in the past four decades. Specifically, the budget of the Drug Enforcement Agency (D EA) increased from $65.2 million in 1972 to $15.1 billion for 2010. At the same time, the number of total employees increased from 2,775 in 1972 to 10,891 in 2006 (Drug Enforcement Agency, 2007a). State and local agencies also spend a sizable amount of their budgets on drug law enforcement. For example, New York and California each spend about $1 billion per year on law enforcement efforts related to the prohibit ion of marijuana use alone (Drug Reform Coordination Network, 2005). Overall, drug la w enforcement receives a sizable amount of funding at the federal, state, and local levels. Policies and Programs Combating Drug Use The importance of combating illicit drug use and abuse is also illustrated by the fact that the government has implemented a considerable number of policies and programs targeting illicit drug use. The policies and programs are aimed either at drug supply reduction or drug demand reduction. An example of a program targeting drug supply would be the Organized Crime Drug Enforcement Task Force, which combines the expertise and resources of all federal agencies involved in drug law enforcement (including the FBI, the Bureau of Immigrati on and Customs Enforcement, the Bureau of Alcohol, Tobacco, Firearms and Explosives, and the U.S. Marshals Se rvice) with the goal of combating major drug trafficking a nd money laundering (DEA, 2007b). An example
18 of a drug demand reduction program would be the Drug-Free Schools and Communities Act administered by the U.S. Department of Ed ucation. The proposed goal of the act is to educate school children of th e dangers of alcohol and illicit drug use and to prevent such illicit drug use (Office of Safe and Drug Free Schools, 2006). Importance of Studying Drug Use for Researchers Illicit drug use is of great importance not only for law enforcement but also for researchers. There is a large amount of research examining drug use prevalence and patterns across different population groups, pred ictors of drug use, evaluation of drug prevention programs (schooland community programs) and drug treatment programs, the cost-effectiveness of treatment programs, and the effectiveness of drug courts. This section reviews the research relevant to the current study drug use prevalence and patterns as shown in the major national drug studies. This is important because it will demonstrate how important DUF and ADAM we re for drug researchers and why it is imperative for the advancement of drug resear chers and the implementation of effective drug reduction programs to systematically monitor drug-abusing behaviors among arrestees across the United States. Ongoing National Studies of Drug Us e Prevalence and Patterns: Why ADAM Should Be Implemented Nationwide The government has collected drug use data via self-report surveys administered to nationally representative samples of households (National Survey on Drug Use and Health) and youths (Monitoring the Future) fo r more than thirty years. Although both studies survey a nationally repr esentative sample of individuals, they rely on data sources with low credibility (self-report) and they survey population groups who are not heavy
19 substance users. Additionally, the Substan ce Abuse and Mental Health Administration (SAMHSA) sponsors the Drug Abuse Warn ing Network (DAWN) program, which monitors drug related emergency room vis its. Next, a brief overview of the NSDUH, MTF, and DAWN will be provided for a better understanding of their drug estimates and why these surveys have not played much of a role for drug researchers and policy makers. The National Survey on Drug Use and Health (NSDUH) The National Survey on Drug Use and Hea lth (NSDUH), formerly known as the National Household Survey on Drug Abuse (NHSDA), was implemented by the Federal Government as an annual survey in 1971. Th e NSDUH targets a representative sample of the civilian, non-institutionalized population aged 12 years and older via face-to-face interviews in all 50 states and the District of Columbia (Department of Health and Human Services, 2006). Thus, the survey in tentionally excludes persons who are institutionalized, such as persons in jail, prison, or mental hospita ls. The survey also excludes persons who have no fixed address (i .e., homeless and tran sient persons), and military personnel. The NSDUH inquires about alcohol and dr ug use for the following drug classes: alcohol, marijuana, cocaine, heroin, halluci nogens, inhalants, psychotherapeutics, and tobacco (SAMHSA, 2005). The que stionnaire first asks participants whether they have ever used any of these drugs If the participants report drug use for any of these drug classes, the interviewer continues with mo re detailed questions for each drug used, including the last time used a nd age of first use. The surv ey also includes a number of questions regarding demographi c characteristics, such as age, gender, pregnant women,
20 education, and employment. Additionally, participants provide information about previous drug treatment, need for drug treatment, needle sharing, experienced consequences of drug use, and thei r criminal history (SAMHSA, 2005). Over time, the NSDUH underwent a numb er of methodological changes, which included changes to the sampling method, the interview method, and the questionnaire (Gfroerer, Eyerman, and Chromy, 2002). The latest change in the sampling procedure was implemented in 2005. The NSDUH now employs a multi-stage probability sampling design consisting of three phase s: (1) Stratification of St ates into 900 State sampling regions; (2) Selection of 48 cen sus tracts per State sampling region, (3) Selection of area segments (census blocks). From these ar ea segments, four samples are drawnone for each quarter of the calendar yearallowing fo r continuous data collection. For each area segment, a listing of the a ddresses are obtained and sampling units selected. Finally, the interviewer randomly selects the sample person via a computerized procedure (SAMHSA, 2007). Changes to the NSDUH also include modifications to the interview method. Until 1998, the survey was conducted via paper and pencil method. In 1999, the paper and pencil method was changed to a computer-assis ted method to increase the response rate. Computer-assisted methods have been shown to result in higher reports of drug use, typically attributed to the hi gher degree of confidentiality. Respondents stated that they were more truthful because the interv iewer would not know about their drug use (SAMHSA, 2000). Several other methodological changes we re also implement ed (SAMHSA, 2007). The name of the program was changed from National Household Survey to National
21 Survey on Drug Use and Health (NSDUH). More importantly, survey participants now receive a payment of $30, which has subs tantially increased the response rate. Additionally, the data quality control procedure (including tr aining sessions for staff, higher degree of supervision of the interviews, evaluation of interviewers) was improved during 2001 and 2002. The sampling weighting procedures of the 2002 survey are based on the 2000 decennial data, whereas the samp ling weighting procedures for previous years are based on the 1990 decennial data (SAMHSA, 2007). In 2007, SAMHSA conducted a study examining the impact of the methodological changes in the NSDUH. Results showed that the response rate had improved significantly, which was proba bly due to the $30 incentive and the implementation of the comput er-assisted interv iewing method. Also, the reported drug prevalence rates were significantly higher, whic h was attributed to the increased quality control procedures and the computer-assi sted interviewing method (SAMHSA). The results of the analysis suggest that the 2002 data should not be compared to the results of previous years because the methodological ch anges had a significant impact on the study outcomes. This also implies that researchers interested in changes in drug prevalence would not be able to use all data between 1971 and 2002 (SAMHSA, 2007). Although the NSDUH is representative of the general population of the United States, it has some major limitations. First, the NSDUH excludes population groups that have been shown to be at high risk of drug use, specifically, instit utionalized persons and homeless persons. As described previously, ther e is considerable ev idence that persons who are institutionalized in jails, prisons, and mental hospitals have much higher rates of illicit drug use than the civilian, non-inst itutionalized population. These civilian, non-
22 institutionalized persons are likely only o ccasional users who mostly use marijuana, and increasingly prescrip tion medication, but who do not provide information about recent changes in drug paraphernalia and the intr oduction of new or modified drugs (i.e., methamphetamine, crack cocaine). It is, however the regular use of hard drugs (such as cocaine, heroin, methamphetamine) and the in troduction of new and/or modified drugs and changes in drug prevalence and patterns th at are the most crucial for researchers, policy makers, local law enforcement, and communities in their attempts to develop programs that effectively reduce drug use. Th e occasional user of marijuana is not the major problem. Second, the NSDUH relies on se lf-report data. As will be described in more detail in the next section, self-re port data has been shown to greatly underestimate drug use. This limitation is a major one, as it may lead to conclusions about illicit drug use that are not correct. The use of incorrect information about drug use by law enforcement and policy makers can result in the implementati on of policies that ar e determined to be ineffective when evaluated. The implementati on of ineffective drug combating strategies is a waste of resources that could be used more effectively elsewhere. Third, the NSDUH assesses geographic differences in drug use for the four major areas: the West, Midwest, Northeast, and Sout h. The results suggest that overall drug use is quite similar across the United States. In 2007, the West had an overall drug use rate of 9.3%, the Midwest had an overall drug use rate of 7.9%, the Northeast had an overall drug use rate of 7.8%, and the South had an overall drug use rate of 7.4%. The NSDUH does not track changes and differences for i ndividual drugs for these geographic regions. This is important because monitoring local areas can be of great help for local law
23 enforcement and communities because they are ab le to target hot spot s of drug sellers and users. As mentioned earlier, it is also importa nt for the implementation of local programs providing drug abuse treatment and for the reducti on of the spread of infectious diseases. This type of information cannot be provi ded by the NSDUH because the civilian, noninstitutionalized population very likely does not hang out at such hot spots and can therefore not provide information about it. Yet, it is these hot s pots and the persons who are dealing and using drugs that cause the major problems for communities. Monitoring the Future (MTF) Another national survey on drug use and abuse is Monitoring the Future (MTF). The MTF is a longitudinal survey implemente d in1975 with the intent to collect data about attitudes towards drugs and drug usi ng behaviors among high school children in the United States (Bachman, Johnston, OMalle y, and Schulenberg, 2006). The goal is to collect data from a nationally representative sample of high school students within the 48 contiguous states. The MTF surveys approximately 50,000 high sc hool children in the spring of each year. The survey includes approximately 110-120 public high schools and 15-20 private high schools. The data is collec ted via a stratified, multi-stage sampling procedure: (1) selection of th e geographic area, (2) selecti on of schools in the selected geographic area, and (3) selection of particip ants in each school. The survey is then administered during regular class periods. The survey asks extensively about the use of licit and illic it drugs including marijuana, sedatives, tranquilizers, hallu cinogens, amphetamines, cocaine, heroin, inhalants, steroids, alcohol, tobacco, stimulants, diet aids, and creatine. The survey also
24 asks the participants about attitudes towards drugs, consequences of their drug use, and whether they believe that they could st op or reduce their drug abusing behaviors (Bachman, et al., 2006). Additionally, the MT F asks a series of questions about demographic characteristics, including race, gender, parental education, and questions about school performance and satisfaction with school (Bachman, et al., 2006). The surveys greatest value lies in its ability to track changes in drug use over time, because the survey is administered repeatedly to the same segments of the population in private and public high schools (8th, 10th, 12th graders, college students, and young adults). As a result, the MTF allows researchers to assess changes in drug use prevalence and patterns in four areas: (1) Peri od effects (changes across all age groups for a certain year), (2) Age effect s (changes in drug use for all panels), (3) Cohort effect (differences among cohorts thr oughout the life cycle) and (4) Differences attributable to differences in the environment (e.g., high school employment) and changes in life (e.g., marriage, military, parenthood) (Johnston, O Malley, Bachman, and Schulenberg, 2007). The MTF, however, also suffers from a numb er of limitations. First, similar to the NSDUH, the survey only collects data via se lf-report surveys. Second, school drop-outs are not captured in the data. This is problema tic because research has demonstrated that drug users are overrepresented among high sc hool drop-outs (ADAM, 2000). Third, some schools may decline participation. These sc hools might differ from schools that are participating in ways that could bias th e sample. Fourth, the MTF does not provide representative data for local areas (only at the national level). Combating drug abuse, however, is very important for local law en forcement and communities. The data from the MTF is not especially helpful for such purposes.
25 As stated earlier, both the NSDUH and the MTF are quite expensive surveys that accomplish very little with regard to pr oviding useful information about drug use, especially hard drugs such as powder co caine, crack, heroin, and methamphetamine. They are also not helpful in determining drug use hot spots and the emergence of new drugs. It might be time to examine whether part of the funding that goes to these two surveys can be redistributed to a national su rvey that studies dr ug use among arrestees. Approximately 70% of jail inmates (BJS, 2002) and 83% of state prison inmates (BJS, 2004) reported drug dependency issues, but only about 9.5% of school children, 19.7% of young adults between 18 and 25, and 5.8% of persons 26 years of age or older have used illicit drugs (SAMHSA, 2008). Considering the fact that jail inmates and prisoners are using illicit drugs at much higher rates th an the general population, it would be more useful to examine arrestees than school ch ildren and non-instituti onalized adults with regard to illicit drug use, dr ug market activity, and criminal behavior associated with drug use. Drug Abuse Warning Network (DAWN) A third national data collecti on tool that provides inform ation about the extent of illicit drug use is the Drug Abuse Warni ng Network (DAWN). DAWN, also sponsored by SAMHSA, was first implemented in 1988 as a program that monitors drug-related emergency room visits and drug-relate d deaths (SAMHSA, 2008). In 2003, a new methodology was implemented with the goal to improve the quality and utility of the monitoring system. Changes were made to the following areas: sample, target population, geographic boundaries, definition and method of finding dawn cases, data content, and supervision of data quality. Due to the changes in the methodology, SAMHSA data from
26 1988 until 2002 cannot be compared to the data collected in 2003 and later years (SAMHSA, 2008). The new DAWN collects data from a sample of hospitals representative of 50 states and the District of Columbia. Eligible hospitals are short term hospitals, general hospitals, and non-federal hospitals operat ing 24-hour emergency departments (EDs). The new DAWN includes all types of drug-relate d ED visits regardless of the patients intent or age. Medical charts of ED visits are retrospectively re viewed, and cases that meet the criteria are selected. Demographi c information and self-reported drug use information obtained from the patient is r ecorded. DAWN collects information about all types of drugs, including illegal drugs, alc ohol, alcohol in combin ation with other drugs, dietary supplements, prescription and ove r-the-counter drugs, and non-pharmaceutical inhalants. The new DAWN program also in cludes toxicology results as a validation technique for the self-reported drug use. Befo re 2003, toxicology results were not used as a confirmation method (SAMHSA, 2008). The greatest strength of the new DAWN pr ogram is that it uses toxicology reports to confirm self-reported drug use and that it is a representative sample of the complete United States. Despite the changes in the methodology, there are two major limitations to the DAWN program. First, the data provided by DAWN is an incidence measure of the consequences of drug use (which could be fi rst time or long-term drug use) and does not provide information about the prevalence of drug use. Second, the data provides information about emergency room incidents, not the number of patients, because one patient could be treated in the emergency room several times and it would be recorded as a separate case each time. Thus, it is not cl ear how many people use drugs, how often
27 they use drugs, where they buy their drugs, a nd other information crucial for researchers and policy makers in comb ating illicit drug use. Advantages of DUF and ADAM Over Other National Surveys The DUF and ADAM program not only complemented the NSDUH, MTF, and DAWN, but also had a number of advantages over the other three surveys that greatly benefited researchers, policy makers, local law enforcement agencies, and communities in ways that none of the other three surv eys could. The exact methodology of DUF and ADAM will be explained in Chapter 3. Th is section will focus on examining the advantages of DUF and ADAM as compared to the NSDUH, MTF, and DAWN and why the DUF and ADAM program were so important for researchers, law enforcement, and policy makers. The three major advantages of the DUF and ADAM program were: (1) They assessed drug use among arrestees a populati on group shown to be at great risk of drug use, (2) they used bioassays to validate the self-report data and (3) they provided local data for every quarter of the year for a period of 15 years. As discussed in the previous chapters, DUF and ADAM focus on a very different population group than the other three national surveys. Specifically, they focus on arrestees, a subpopulation with a significantly higher rate of drug use than the general population or school children. For instance, th e Bureau of Justice Statistics (2004b) has found that 53% of state and 45% of federal pr isoners have substance abuse issues. These rates are significantly higher than in the ge neral population. The results from the National Survey on Drugs and Health (NSDUH) show that only about 8% of people 12 years and older have substance abuse issues (SAMHSA, 2007). This means that prisoners are about
28 5 times more likely to use illegal subs tances than the ge neral population. Another important difference is that psychotherapeutic drugs are among the preferred drugs used in the general popul ation. Psychotherapeutic drugs include antianxiety drugs (i.e., Xanax, Valium), antidepress ant drugs (i.e., Zoloft, Proxil), antimatic drugs (i.e., Eskalith), and an tipsychotic drugs (i.e., Thorazi ne). These psychotherapeutic drugs are not the preferred drugs of prisoners and arrestees. In sum, drug use behaviors among arrestees, prisoners, and the genera l population vary significantly. Studying drug use among the general population and not among arrestees and other population groups at high-risk of illegal substance use will resu lt in misleading conclusions about drug use in the United States. Overall, researchers estimate that the national surveys miss the vast majority of the drugs that are being used. For instance, Kleiman (2004) states that the NSDUH accounts for only about 30 metric tons out of about 300 me tric tons of the cocaine consumption. Thus, when policy make rs implement strategies aiming to reduce cocaine demand based on the household survey, they are relying on information from the wrong population group. Until 2003, the DUF and ADAM programs were the only national drug study that allowed researchers to assess the validity of self-reported drug use. This is important because the main criticism of self-reported be haviors pertains to the question of how valid the results are. Validity for this purpos e refers to the question whether the data recorded by the researcher accurately re flect the phenomenon under investigation (Harrell, 1985). Stated differently, do people answer questions in self-report surveys truthfully? Since 2003, the DAWN program also uses bioassays to confirm self-reported drug
29 use. The NSDUH and the MTF still rely on self -report only. This is a major limitation. Thus, not only do these two expensive na tional surveys assess drug use among the general population, which only accounts for a minimal part of the drug use activity, but these surveys also do not validat e the self-report data via obj ective validation techniques. Validity of Self-Report Data Self-report surveys have a long history a nd the validity of the obtained data has been suspect to many researchers because the collected data might be erroneous (Stone, Turkkan, Bachrach, Jobe, Kurtsam, and Ca in, 2000). As early as the late 1940s, researchers found systematic biases in self-re ported behaviors, probably due to incorrect information provided by the study pa rticipants (Stone, et al., 2000). Although early research on the validity of illicit drug use suggested that respondents reported drug use fairly accurate ly, more recent improvements in technology have allowed researchers to employ more sophisticated validation techniques which have led to a different conclusion (Harrison, 1995). The results of these more recent studies suggest that self-reported drug use is not reported as accurately as thought. Rather, respondents are likely to underre port drug use. For example, Fendrich, et al. (1999) found that respondents from a high-risk community sample heavily underreported cocaine and heroin use. They found that only 20% of cocaine-positive respondents also admitted to current use of cocaine. Similarly, Appel et al. (2001) found th at only 26% of the respondents from a sample of homeless and tr ansient persons in New York who tested positive for cocaine use also reported their drug use. There is, however, some research su ggesting that the unde rreporting of selfreported drug use might not be as large as believed. A recent SAMHSA study suggests
30 that there is a high degree of agreement between self-repo rted drug use and urine test results among the general population. Specifi cally, for marijuana use there was 89.9% agreement and for cocaine use there was 98.5% agreement (Harrison, Martin, Enev, and Harrington, 2007). Other studies show that the agreement between self-reported delinquency (as measured by self-reported arrest s) and official data (arrest records) is between 50% and 83% (Hindelang, et al., 1981; Hindelang & Krohn, 2000). Hindelang and Krohn (2000) conclude that these agreemen t rates are reasonably high. It is, however, impossible for researchers to know whether 50 % or 80%of their study participants told the truth unless the data is verified via objectiv e measures, such as official arrest records, criminal records, or urine analysis. It app ears that it would be of great importance to know if only 50% of the subjects gave tr ue answers. Thus, although some research suggests that self-report data can be a valid measure of delinquency and drug use, there is a substantial amount of research suggesting otherwise. A number of researchers have also de monstrated that the accuracy of selfreported drug use depends on the type of drug and the population subgroup. The more stigmatized the drug, the less likely res pondents report its use (Harrison, 1992, 1995; Mieczkowski, et al., 1991). Using data fr om the DUF program, Harrison (1992) found that arrestees most accurately report use of opiates (60%), followed by marijuana (55% concordance). Arrestees were le ast likely to report the use of cocaine (50% concordance) and amphetamines (40%). This also holds true for other population subgroups. A study by Hser, Maglione, and Boyle (1999) s uggests that self-reported drug use among emergency room patients (ER) and patients with sexually transmitted diseases (STD) was more accurate for marijuana as compared to any other drug. Hser, et al., (1999) study
31 also demonstrates that different population subgroups are more or less likely to accurately report drug use. ER patients and STD patients, as compared to arrestees, were less likely to report drug use overall, but especially th e use of hard drugs. The reason may be that ER patients and STD patients perceived the soci al stigma associated with illicit drug use to be greater than arrestees, who were alr eady stigmatized due to the arrest itself and admitting using drugs may be a minor issue comp ared to their arrest (Hser et al., 1999). More recent research by Golub, et al. (2005) has found that disclosure rates also vary across geographic locations. Epidemiological research suggests that pe ople even fail to disc lose illicit drug use in situations where a failure to do so might potentially harm them. For instance, Tassiopoulos, et al. (2004) found that 34.2% of out-of-treatment heroin users did not admit that they were also using cocaine. In an emergency situation, failure to disclose the use of an illicit drug may have a negative im pact on medical care. It appears that the stigma associated with drug use is perceived to be so strong that it is kept secret even in life-threatening situations. If i ndividuals do not disclose drug use in situations where their health is at stake, why would they disc lose such behaviors to an interviewer? Though there is considerable evidence th at drug use is generally underreported, there are exceptions. Persons who have either recently been admitted into a drug treatment program or who have recently fi nished a drug treatment program report drug use fairly accurately. A study by Hindin, et al. (1994) using da ta from persons entering a residential drug treatment program found that 89% percent of the re spondents who tested positive for cocaine use also reported such use. The concordance rate for heroin was even higher, with 96% of the individuals who tested positive for heroin use also reporting such
32 use. In some instances, over-reporting can o ccur. For instance, arrestees may overreport drug use if they believe that it may in crease their chances to enter a drug treatment program instead of going to jail (Hser, 1999). Arrestees may also over-report drug use if they have been arrested for a violent crim e and anticipate a long prison sentence. Drug use in this case may serve as a mitigating circumstance and make them appear less culpable (Nurco, 1985). Other population subg roups may also over-report drug use. For example, adolescents and college students might over-report drug use to make their behavior patterns fit the behavior that is accepted among their peers. Johnston and OMalley (1997) found that adoles cent males as well as male and female college students reported fewer incidents of drug use in a follo w up interview. The researchers concluded: the revised may well be the more accurate number, and the answers given at earlier ages . may be inflated (p. 78). Both unde r-reporting and over-r eporting can lead to inaccurate conclusions about drug use. DUF and ADAM we re unique because they validated self-reported drug use with drug te sts, therefore providing a more objective measure of drug use among arrestees. Without the validation of self-reported drug use via bioassays, researchers and policy makers are left guessing about how accurate the collected data is. Explanations for the Failure to Report Behaviors Truthfully Several explanations have been advanced to explain the under-reporting of drug use. Threats to the validity of se lf-report data are typically said to stem from the failure to remember events accurately and the failure to report behavior s truthfully (Harrell, 1985). For researchers studying the prevalence and patterns of illicit drug use and other
33 stigmatized behaviors, threats to validity stem mostly from the second categorythe failure to report behaviors trut hfully. To further explore th is question, researchers have proposed a number of theoretical explana tions of why respondents may not answer questions in self-report interviews truthf ully even though they are assured that the answers are kept fully conf idential or that the answer s are anonymous (Sloan III, Bodapati, and Tucker, 2004). One of these explanations is the Social Desirability Theory. The purpose of the self-report surveys is typically very obviou s to the respondent, and as a result it is rather easy for the respondents to manipulate the answers (Cook and Selltitz, 1964). The underlying assumption of the social desirability theory about human nature is that humans are social beings a nd that their behavior is oriented on the behaviors of others (Weber, 1968) Accordingly, social desira bility theory advances the thesis that persons will respond to questions in a way that is consistent with social norms, expectations, and socially desirable traits (Zerbe and Paulhus, 1987, p. 250). In other words, people will respond to questions in a way that places them in a good light. Researchers have found support for the social de sirability theory in a variety of topics. The social desirability thes is first received empirical support from a study conducted in 1953 by Edwards. Edwards (1953) found that ther e is a relationship between what survey respondents believed to be the socially desi rable answer and the answer given by these respondents. Additionally, Willis and Schechte r (1997) examined how study participants felt about what the interviewer might think a bout them and react towards them if they disclosed their drug abusing be haviors. Their study demonstr ated that individuals were very concerned about interviewer reactions a nd preferred not to talk about their drug
34 using behaviors because they felt that they were being judged. There is considerable evidence that self -reported behavior is biased towards normative behavior. For instance, the Magazine Audience Group by Crossley Incorporated (1941) examined the validity of self-reported educational attainment. The results of this study showed that individuals exaggerate their educational attainment regarding graduation from the various edu cational institu tions (e.g., grade school, high school, and college) (Parry and Crossley, 1950). In a study about redeeming war bonds, Hyman (1944) found that individuals were lik ely to deny that they had redeemed war bonds. The denial of redeeming war bonds was greater for individuals with a higher income status. Presser (1990) found that pe ople over-reported voting behavior and the attendance of religious services. To control for a social desirability bias, researchers have included measures in their questionnaires meant to provide information about whether the participant may have altered his/her answers because he wanted to give the correct answer. The most commonly used measure for this purpose is the Marlowe-Crowne Social Desirability Scale (Crowne and Marlow, 1964). Unfortunately, the purpose of this scale is quite transparent to the respondents and respondent s may answer the questions of the scale honestly but not the questions a bout drug use. As a result, re searchers might interpret the self-reported behavior as accurate when it is really not (Richter and Johnson, 2001). Threats to the validity of se lf-reported drug use also aris e from the fact that people may not recall events accurately. Even when the respondent is motivated to provide accurate information about self-reported drug us e or other behavior, they may simply not remember which drug they used at what point in time or how often they used it. The drug
35 use itself may distort their re call of events (Catania, 1993). Harrell (1985) suggests that general behaviors are easier to recall than specific behaviors. For example, drug users might recall accurately having used a certain drug in the last few days, but may not remember how often they have used the dr ug or how much of the drug they used. Additionally, more recent events are reca lled more accurately than events that occurred a while ago (Harrell, 1985). This is also referred to as recall decay (Johnson, et al. 1997). A study by Roberts, et al. (2005) suggests that individuals underestimate violent behaviors within a 1-3 year time period using a li fe-events calendar This type of calendar was also used by the ADAM study to assess drug use prevalence and patterns within the past 12 months. Recall decay is not only a problem for researchers studying drug using or criminal behaviors but also for researchers in other fields including medicine. Studies have found that individuals will significantly undere stimate injury rates if asked more than 2 months after the inci dent (Jenkins, Earle-Ri chardson, Slingerman, and May, 2002). Considering that many people cannot accurately remember which injuries they suffered or violent behaviors, r ecall decay is a serious threat to the validity of drug use data. Additionally, individuals may believe that a certain ev ent occurred more recently than it actually did. This is called forward te lescoping. Individuals might fairly accurately recall whether they ever used a certain drug, but they might not be accurate in recalling in which year or during what time period th e drug was used (Johnson and Schultz, 2005). Telescoping is especially a problem when an event occurs regularly and is not salient in the sense that it would be remembered easily (Magnusson and Bergman, 1990). Many drug users use drugs regularly, some several times a day. It is unlikely that these regular
36 drug using behaviors are particul arly salient, and as a resu lt it cannot be expected that drug use is being recalled accurately. Implications of the Lack of Reporting for Researchers and Policy Makers The implications of the lack of comple te reporting are manifold. For instance, it has implications for researchers who are drawing conclusions about the relationship between drug use and crime, changes in drug use prevalence and patterns, and other issues. A lack of truthful re porting will necessarily lead to inaccurate conc lusions. If the goal of research is to advance the knowledge in a certain field publishing results based on incorrect information would be counterproductiv e. The most important implication of a lack of truthful reporting be haviors of drug use probably pertains to the implementation of policies meant to reduce drug use, provide treatment to drug us ers, and prevent the spread of infectious diseases, including HI V and Hepatitis (Des Jarlais, 1998). If programs are based on invalid data, then the effectiveness of thes e programs might be quite low. Additionally, a lack of truthful reporting (mostly under-reporting) might lead to the conclusion that drug abuse treatmen t programs are not needed in a certain community. As a result, drug users will not receive treatment and continue to use drugs, increasing the likelihood of th e spread of infectious dis eases and criminal behavior associated with drug use. Due to the great likelihood that self-reported data is inaccurate, especially for sensitive and stigmatized behaviors (such as il licit drug use), researcher s need to validate data whenever possible. The ever-present issu e is what measure could be used to assess the truthfulness of the provided information.
37 Validation Methods for Self-Reported Drug Use In the case of illicit drug use, researchers have a variety of validation techniques available, including the analysis of urine specimens, hair samples, sweat, or saliva (Cone, 1997). Each of these methods is unique and pr ovides different types of information. Each method also has its strengths and weaknesses. The usefulness of a certain drug testing method depends on a number of factors, whic h all relate to the accuracy with which a certain method is able to detect drugs in biological fluid or tissue (Cone, 1997). The factors influencing the usefulness of a testi ng method are (1) sensitivity, (2) specificity, and (3) accuracy. Sensitivity refers to the l east amount of detectable drug, meaning that the more sensitive the test, the lower the con centration of the drug that can be measured (Cone, 1997, p. 109). Specificity refers to how se lective the assay is for the drug or the ability of the test to distinguish between different drugs (Cone, 1997, p. 109). The higher the specificity of a test, the more accurately it can determine the presence of a certain drug. The most common method utilized in drug use research is urine testing. Both DUF and ADAM employed this method. Research has demonstrated that, at the current time, it is the most accurate drug screening method fo r recent drug use (2-4 days) as compared to hair, saliva, or sweat (Mi eczkowski and Newel, 1997). Urine Testing Urine can be used for drug screening because urine is produced by the kidneys, which reabsorb and eliminate substances (suc h as drugs) that are waste products for the body (Cone, 1997). As a result, the substances that are being eliminated from the body will show up in the urine specimen. Most il licit drugs will be eliminated from the body within 48 hours of administration. If drugs are taken consistently and over longer periods
38 of time, the detection time can be longer. Th e detection time also depends on the drug. For example, heroin, cocaine, and mariju ana have a detection time of 1-3 days. Barbiturates, amphetamines, methadone, and methamphetamine have a detection time of 2-4 days (Cone, 1997). Thus, urine testing is ve ry useful for the det ection of short term drug use, but it is not useful for assessing drug use that occurred more than a few days ago. This also means urine testing cannot be useful to examine drug use over long periods of time. Another problem with urine testing is th at the cut-off levels for detecting the drugs are considerably different depending on the drug, meaning that some drugs are detected at much lower doses than others. Fo r instance, the cut-off level for marijuana is as low as 20 ng/ml, whereas the cut-off level for methamphetamine is 1000 ng/ml. Additionally, delay in drug test ing (due to holding the specimen for a certain period of time before testing it) may also lead to a failu re to detect drug use because the dose of the drug might have fallen below the cut-off level. Despite these shortcomings, urine testing ha s been shown to be a valid measure of recent drug use. The advantages of urine test s have not only been demonstrated by social and behavioral research, but they have also long been recognized in the criminal justice system. For example, urine tests are used by courts to monitor abstinence and relapse, because offenders will very likely not report drug use to the Court (Bureau of Justice Assistance, 1999). These studies and practices by the criminal justice system demonstrate the importance of using bioassays as a validation technique for self-reported drug use. In sum, the decision made by NIJ to opt for urin e analysis as a validation method for recent self-reported drug use for the DUF and ADAM programs appears to be an adequate
39 decision and was the major strength of DU F and ADAM over other national drug studies (such as the NSDUH and the MTF). The use of bioassays as a valid ation technique of self-reported drug use was not the only advantage of DUF and ADAM over the NSDUH and the MTF. Another important advantage was the collection of localized data. Localized Data Collected Every Quarter Over 15 Years As described earlier, the NSDUH, MTF, and DAWN provide national data on drug use only. Although they provide data fo r the major geographic areas (sub-state areas) of the United States, ther e is little data at the county and city levels. This county and city level data, however, is very usef ul for local law enforcement agencies and communities in their efforts to target hot spots of drug use and drug sale. DUF and ADAM filled that gap by providing county and c ity level data for the major drugs used. This site-specific information is important because the results of DUF and ADAM have consistently demonstrated that there are si gnificant differences in drug use by geographic region and over time. For instance, the DUF program showed that methamphetamine use in the United States was modest, with about 6%, but it was becoming a great problem in the western part of the country (NIJ, 1996). Sp ecifically, the six si tes with the highest methamphetamine use rates were San Di ego (37.1%), Phoenix (21.9%), Portland (18.7%), San Jose (18.5%), Omaha (8.1%), and Los Angeles (7.5%) (NIJ, 1996). A study by the National Institute of Drug Abuse (NIDA) examining methamphetamine use in 21 areas across the United States supports the results found by DUF and ADAM (NIDA, 2006). Methamphetamine use is mostly a problem of the Western states of the United States, esp ecially Honolulu, San Diego, Seattle, San Francisco, and Los Angeles. Additionally, methamphetamine use increased over time,
40 which showed especially in the number of drug treatment admissions and emergency room visits. Specifically, emergency room vi sits for methamphetamine related problems (provided by the Drug Abuse Warning Netw ork) increased by 70% between 1999 and 2002 (Franco, 2007). Additionally, drug treatment admissions (provided by the Treatment Episode Data Set collected by SA MHSA) for methamphetamine use have also greatly increased, not only in the Western Un ited States, but across the country. In 1992, only 5 states reported a high number of dr ug treatment admissions. In 2002, 21 states reported high number of drug tr eatment admissions (NIDA, 2006). The results from the MTF also demons trate that methamphetamine was much more widespread in the western part of th e United States. Unfortunately, questions about methamphetamine use were not included in the survey until 19 99. Since 1999, the MTF suggests that methamphetamine use has been declining, from 8.2% in 1999 to 4.5% in 2005. The DUF data showed that methamphetami ne began to rise sharply in the early 1990s. For instance, Herz (2000) using data fr om the DUF program for Omaha, Nebraska showed that methamphetamine use increased from 1% in 1990 to 10% in 1999. Being able to notice early when a certain drug is on the rise is important to prevent it from becoming an epidemic. Research suggests that recently implemented compstat systems can be used to track drug ac tivity. These systems help local police departments determine areas with high crime and drug activity. For instance, the cities of Lowell and Newark were able to allocate police officers depending on the need of certain communities by using the compstat system (Willis, Mastro fski, and Weisburg, 2003). One problem with the local compstat systems is that they dont exist across the entire country. Also, the data collected by these systems is not send to a na tional data collection agency. As a result,
41 these data are not available to researchers and others who are looking at trends across different geographic locations. Early knowledge about a ri sing drug problem enables law enforcement and the government to implement supply reduction strate gies that reduce the availability of the drug on the streets and increase the prices for the drug. These federal and local prevention and intervention programs are only possible, however, if the problem is known not only to the local police department who uses a co mpstat system but also to departments in surrounding geographic locations. This is impor tant because of a possible displacement of drug use and drug crime to other areas as a result of the increased police presence in areas with high drug use. As stated before, the distribution of the data is a necessary prerequisite to combating illicit drug us e. DUF and ADAM were doing that by collecting data at the local level and making it avai lable to researchers and policy makers. This systematic data collection and dist ribution contributed to the implementation of local programs. In 1996, the government implemented the Comprehensive Methamphetamine Act, which aimed to reduce drug trafficking and reduce the availability of chemicals needed to produ ce methamphetamine. Researchers suggest that this Act was a direct response of the White House to the rising levels of methamphetamine use as documented by the DUF program (National Criminal Justice Association, 1999). Although the DUF program was not the only data that showed the regional increases of methamphetamine use, it provided important information that helped getting the attention of policy ma kers. For instance, Oregon reduced the availability of drugs (mainly pseudoephedrin) needed to produce methamphetamine (as provided by the Comprehensive Methamphetamine Act). Studies showed that during the
42 time of supply disruption, methamphetam ine use decreased substantially (Cunningham and Liu, 2003; 2005). The reducti on of methamphetamine after the implementation of the Act was also evident in California and ot her states ((National Criminal Justice Association, 1999).). Without DUF and other programs, the sharp rise and extent of the methamphetamine problem might have gone unno ticed. As a result, the rising levels of methamphetamine use in the western part of the country might have further increased. Similar to the MTF, NSDUH data also fa iled to show the substantial rise in methamphetamine use in the West. Specifically, the NSDUH show that methamphetamine use was less than 2% unt il 1994 and rose between 1994 and 2001 to 4%. Also, as described above, the MTF suggest s that methamphetamine use has declined since 1990. In contrast, data from the ADAM program demonstrate that for some sites methamphetamine use has increased. Specifically, in San Jose, San Diego, and Phoenix, methamphetamine use increased substantia lly between 2000 and 2003. These results are also supported by the DAWN data, which shows an increase in emergency room visits for methamphetamine use between 1995 and 2002 in Los Angeles, Minneapolis, St. Louis, Seattle, Atlanta, New Orleans, and New York. Thus, the national data provided by the NSDUH and the MTF provide the false im pression that methamphetamine is on the decline in general, when there are si gnificant regional differences. Whereas methamphetamine use is declining in some re gions, it is still increa sing in others (Hunt, et al., 2006). The revived ADAM II program, conducte d by the ONDCP, also supports the finding that there are large re gional differences in methamphetamine use and while some regions, such as Washington, DC and Portland, still show a decrease in
43 methamphetamine use between 2003 and 2007, other sites (including Atlanta, Minneapolis, Sacramento, Indianapolis, Charlo tte, and Chicago) have remained stable (Office of National Drug Control Policy, 2008). Some police departments collect data on drug use in their community. For instance, the police department in High Poin t, North Carolina colle cts data about drug markets with the goal of targeting drug d ealers more effectively (High Point Police Department). Although that is a viable approach if the goal is to target drug dealers and dismantle drug markets in a specific area, this does not necessarily advance research or help policy makers implement strategies fo r a greater region or nationwide because the data is not collected systematically. Instead, each local agency uses its own methods (e.g. surveys, GIS, crime reports) to determine hot spots of drug dealers and high drug market activity. This data is not comparable across agencies. Additionally, the local agencies do not submit their data to a national data ba nk where researchers could access such data. Systematic data collection at the national level ensures that researchers have data available to advance our knowle dge about drug use and related issues. It is especially important to implement a national study that co llects data systematically from arrestees because their drug using behaviors are substan tially different from those of the general population and school children. The NSDUH data not only failed to show the geographic differences for methamphetamine use but also greatly underestimated the extent of the methamphetamine problem. In fact, critics of the anti-methamphetamine drug policy argued that the NSDUH data does not s upport the degree of attention given to methamphetamine (Franco, 2007). This is problematic because research has shown that
44 methamphetamine has stronger and longer las ting toxic effects than amphetamines or cocaine. For instance, smoking methamphetamine can create a high for 8 to 24 hours. In comparison, cocaine creates a high for only about 20 to 30 minutes. Additionally, the amount of dopamine released to the brain is three times higher for methamphetamine than for cocaine. Also, it takes approximately 12 hours for half of the methamphetamine to be metabolized (half-life). The half-life of co caine is only one hour. Methamphetamine has also been shown to increase the likelihood of HIV and Hepatitis infections due to risky sexual behavior. These findings indicat e the great health consequences of methamphetamine use, and as a result the need to contain methamphetamine use when it first started to become more popular. In the 1990s, when neither the MTF nor the NSDUH provided valid information about methamphetamine use, DUF and ADAM were crucial to proving to critics that the impleme nted drug policies were very necessary to decrease the use of methamphetamine in the we stern part of the count ry and to hinder the further spread of methamphetamine use to other parts of the United States. After having described the major adva ntages of the DUF and ADAM programs over the other national drug use monitoring programs the following section will now expand on differences and similarities between drug use data from the DUF and ADAM programs as compared to the NSDUH, MTF, and DAWN programs because these differences will demonstrate the importance of implementing a national program that systematically tracks drug use among arrestees. Changes in Drug Use Prevalence and Patterns between 1988 and 2002 Research from the DUF/ADAM programs, the NSDUH, and the MTF
45 demonstrate that changes in drug use preval ence and patterns tend to show up among the criminal justice population first (DUF/ADAM data) and then spread to the general population (NSDUH and MTF data). With regard to marijuana use among arrestees, three waves (or generations) are visible be tween 1987 and 1998 (Golub and Johnson, 2001). First, marijuana use declined in the 1 980s until about 1992. Second, beginning in 1992, marijuana use started to increase and then stabilized until about 1996. The third wave is characterized by a significant increase in marijuana use in 1996. This third wave appeared to plateau in 1999. Although these th ree waves are also apparent in the NSDUH and MTF, the beginning of a new wave showed up in the arrestee da ta from the DUF and ADAM data first and only later in the general population (G olub and Johnson, 2001). Specifically, the increase in marijuana use became obvious first among youthful arrestees in the DUF data in 1991. This increase did not surface among the general population until about one or two years la ter (Golub and Johnson, 2001). The results further suggest that the rise in marijuana use was more pronounced among the criminal justice population, indicating that it spread more widely among this population as compared to the general population (G olub and Johnson, 2001). Additionally, the NSDUH shows a general decrease of marijuana use as individuals get older, the increase and decrease of marijuana use among a rrestees was more dependent on geographic location. In some areas, marijuana use increased among older (born before 1967) arrestees (i.e., Los Angeles, Dallas, Denver, Houston), in other areas, marijuana use decreased among older arrestees (i.e., Po rtland, San Diego, San Jose) (Golub and Johnson, 2001). There were also significant differences with regard to crack cocaine. Crack
46 cocaine first appeared in the United Stat es in the 1980s and quickly became popular (Golub and Johnson, 1997). Golub and Johnson (1997) examined the prevalence of crack cocaine between 1987 and 1996 using the DUF data. Their major findings demonstrate that crack cocaine became popular first among older, more experienced drug users, who then introduced crack to young and new users. Crack cocaine was cheap, easy to use (smoking), and widely available. Crack cocaine became an epidemic in the late 1980s and began to decline around 1996 mostly be cause younger drug users began to look down on crack users (also referred to as crac kheads) (Golub and Johnson, 1997). The rising disdain of young drug users for crack resulted in the decline of cr ack cocaine use. In contrast, the decrease in crack cocaine us e among older, more experienced users was much less dramatic. In sum, the decline in crack cocaine use was not the result of a similar decline in the use of crack among all birth cohorts, but was caused mainly by the fact that youthful users stopped using crack cocaine (Golub and Johnson, 1997). Whereas the DUF data showed that crack cocaine was being used widely between 1982 and 1996, the NSDUH did not distinguish between crack and powder cocaine until 1988. By that time, crack cocaine use was already decreasing, especially among young users. Thus, the NSDUH could not provide da ta about crack cocaine use in the general population at the time when crack was at its peak. Without the DUF and ADAM programs, researchers, law enforcement, and policy makers might not have known the true extent of the crack epidemic at the lo cal and national levels. As stated above, even though local agencies might have had that inform ation for their jurisdic tions, this data is not available in a national database to researchers and policy makers. The data also suggests that there was great geographic variation in the extent of
47 crack use. The crack cocaine epidemic di d not decline at the same rate across the country. Rather, the DUF data indicates that there was a significant decrease in crack cocaine use at 17 sites, still at its peak at five sites, and two sites did not have an epidemic. The popularity of crack cocaine varied even across sites that were geographically close. For instance, the declin e of crack cocaine use in San Diego started in 1992 and was quite dramatic. Specifically, crack cocaine use declined from 37% in 1991 to 13% in 1996 among youthful offenders (23% decline). The decline of crack cocaine use in Los Angeles began around 1989, but was very slow. In Los Angeles between 1988 and 1996, crack cocaine use declin ed from 60% to 46% (14% decline). As was the case for marijuana and methamphetamine, the NSDUH data gives the impression that the rise and decline of crack cocain e happens at the same time across geographic locations in the country. As shown by DUF and ADAM, as well as DAWN, that is not correct. Again, this information is detrimental to the implementation of effective policies. There is no need for policies if crack cocaine is not being used. Similarly, the assumption that crack cocaine use is dec lining equally at all sites might lead to an abandonment of specific law enforcement strategies that might still be needed. In sum, the findings of the DUF and ADAM programs demonstrate the differences in drug use prevalence and pattern s found in the general population versus the population of arrestees. It also shows how important it is to monitor drug use among arrestees because trends tend to show up in the criminal justice population before they spread to the general population. DUF and ADAM were not only cruc ial for researchers examining drug using behaviors and the validi ty of self-reported drug use, but also to policy makers and law enforcement working to reduce drug use.
48 Drug Court Movement The development of drug courts and diverting drug offenders into treatment services within the community has shown to be beneficial for the drug offenders and the community (ONDCP, 2003). Specifically, comp rehensive drug treatment costs about $2,500 per year. In comparison, the costs of incarceration range anywhere between $20,000 and $50,000 per year for each person. Thus, drug treatment instead of incarceration can save states a sizable amount of money. Drug abusing offenders who receive treatment also have a significantly lower recidivism rate than drug abusing offenders who do not participate in drug court programs. The Office of National Drug Control Policy (2003) suggests th at the recidivism rate of drug court graduates is less than 4%, compared to 66.7% of drug offe nders released from prison (BJS, 2002). DUF and ADAM provided information essential to the implementation of treatment and prevention programs for drug using offenders in communities by providing local data about drug using behaviors. As a re sult, it furthered efforts to lower recidivism rates among these drug abusing offenders. For instance, the ADAM data aided in the implementation of a new drug policy (Proposi tion 36) in Los Angeles County (Drug Use Alliance, 2006). Proposition 36 enables cour ts to place non-violent offenders with substance abuse issues in trea tment programs rather than sending them to jail or prison. The DUF and ADAM data were important because they helped change the attitudes of policy makers and the public towards the treatment approach. The result of Proposition 36 is that in 2006 over 140,000 non-violent drug offenders received treatment (Drug Use Alliance, 2006). This is important because the lifetime prevalence rates for substance abuse
49 disorders among prisoners are between 68% and 74% (Karberg and James, 2005). The vast majority of these prisoners eventually return to the community, where they continue their drug using behaviors. Illicit drug us e causes great costs for society because drug using prisoners have a high rate of recidivism and are very li kely to engage in criminal activity. Specifically, Langan and Levin (2002) estimate that approximately two-thirds of drug involved offenders are re-arrested within three years of rel ease from custody. These repeat offenders cause significant costs to the criminal justice system. The DUF and ADAM studies aided in the increase of f unding for drug treatment in correctional facilities and encouraged the drug court movement. The drug court movement diverts drug abusing offenders away from the crimin al justice system and provides them with drug treatment. There is considerable evidence that comprehensive drug treatment can effectively reduce drug use and recidivism. This is especially true when the drug treatment is followed by an aftercare progr am (Hora, Schma, and Rosenthal, 1998; Inciardi, Martin, & Butzin, 2004; Martin, Butzin, Saum, a nd Inciardi, 1999; Office of Justice Programs, 1998; Pre ndergast and Wexler, 2004). DUF and ADAM were important for the pr ogress made with regard to diverting drug offenders to drug courts for two reas ons: (1) DUF and ADAM caught the attention of policy makers by showing the great extent of the drug use problem within the criminal justice population; and (2) by demonstrati ng the association between drug use and criminal behavior. For instance, the DUF findi ngs indicate that incr eases in drug use and addiction lead to an increase in the crime rate (Fagan and Chin, 1990; Goldstein, 1990). Drug Treatment in Correctional Facilities Additionally, correctional facilities have also expanded their treatment services.
50 According to the Bureau of Justice Statistics (2004a), the percentage of jail inmates who received drug treatment increased from 39% in 1996 to 47% in 2002. Similarly, drug treatment programs have also increased fo r state and federal prisoners(BJS, 2002). DUF and ADAM provided crucial information to jail administrators implementing drug treatment programs. Data for Local Police Agencies Furthermore, DUF and ADAM provided information about drug use, drug markets, and hot spots for local police agen cies. By providing local data about drug prevalence and patterns, specifi cally information about wher e arrestees buy their drugs and where they use drugs, DUF and ADAM help ed local police agencies in their efforts to target drug sellers and users more eff ectively. Local police are a critical group in reducing the sale and use of illegal drugs. The ability to target specific areas known to be hot spots for drug sellers and users aids in the decrease of drug activity in those areas. This is also true for firearms. DUF addenda collected data on firearm use which helped local police agencies. For instance, in Kansas Ci ty, police were able to target hot spots of firearm violence with gun seizures and significantly reduce the gun crime in these areas as a result (Decker, Pennell, and Caldwell, 1997). DUF and ADAM were, however, important not only for law enforcement but also for policy makers and communities dealing wi th disease control. There is a strong connection between drug use and HIV and ot her infectious diseases. (i.e., hepatitis, tuberculosis) (Foundation of Drug Research, 2008). Among drug users, these infectious diseases are often spread vi a needle sharing and risky sexual behaviors (Zack, 2008). The Center for Disease Control and Prevention (CDC) found that 42% of AIDS cases stem
51 from behaviors associated with drug use, eith er via needle sharing or unprotected sex. Additionally, 81% of AIDS cases in children are associated with transmission from the mother infected via drug injection and/or unprotected sex with a drug injecting person (CDC, 2004). Research has shown that substance abuse treatment is effective in reducing drug use and risky sexual beha vior among offenders (World Health Organization, 2004). For instance, the city of San Diego used information provided by the ADAM data to implement the Clean Syringe Exchange Program, which was found to reduce needle sharing and the spread of infectious dis eases (Burke, 2004; City of San Diego, 2001). Similarly, research using the DUF data in combination with data from the Needle Exchange Program (NEP) has demonstrated that cities which implemented the NEP reduced needle sharing and drug injection, as a result decreasing th e spread of HIV and other infectious diseases (D eSimone, 2005). This, in turn, reduces the costs for medical treatment and other services. The implications of these findings are that communities would benefit from educating drug users abou t the risks of unprot ected sexual behaviors and needle sharing. Communities would also benefit from programs that proactively reduce risky behaviors, such as needle ex change (or similar) programs for drug users (Stephenson, et al., 2005). Educational Attainment and Drug Use Finally, data about drug use prevalence a nd patterns among arrestees is important because it demonstrates how cruc ial it is to keep kids in sc hool and further educational attainment. There is a strong connection between e ducational attainment and drug use. As a result, there is also a connection be tween drug use and the number of unemployed people and people with low paying jobs. There is considerable evidence that people with
52 low educational attainment are overrepresen ted among arrestees and especially among arrestees with substance abus e issues (Harder and Chilcoa t, 2007). For example, DUF and ADAMclearly demonstrates th at the population of arrestees consists in great part of individuals with low educational attainment (i.e., no high school diploma), low income, and a lack of health insurance. This hinders their ability to seek treatment and other services that would increase their chances to break the cycle of drug use and offending. Specifically, on average 30% of arrestees did not have a hi gh school diploma, about 34% were unemployed, and a substant ial proportion (15%) had no fixed address (NIJ, 2000). Research has also shown that a hi gher educational achieve ment is related to a better understanding of risky behaviors and the ability to find resources and treatment options that help address substance abuse i ssues and modify behavi ors (Link and Phelan, 1995). These findings have implications for communities as they pertain to the importance of education in reducing drug use. Communities would benefit from programs that keep kids in school. Educatio nal programs are certainly cheaper than the medical and criminal justice costs associat ed with drug offenders. These examples demonstrate the importance of systematically studying drug using behaviors among arrestees. Only if we understand these dr ug using behaviors can communities and law enforcement implement effective programs that aim to reduce drug use and provide services to individuals who are using drugs. Lack of a National Study Monitoring Drug Use Among Arrestees To reiterate, currently th ere is no national study examining drug use prevalence and patterns among arrestees because th e revived ADAM II program (conducted by the ONDCP) only collects data for 10 counties with in the United States. To reiterate, even
53 though local agencies might be collecting simila r drug use data, this is not a systematic data collection and therefore results in a num ber of problems associat ed with the use of such data. These problems are: (1) it makes it more difficult for researchers to obtain such data, (2) it is not publicized widely wher e such data exists, and (3) methodological differences between local data collection mi ght make it difficult to compare data across sites. Researchers argue that the money spen t on examining drug using behaviors is distributed improperly. The National Househol d Survey costs $50 million a year, but it examines a population group that is at low risk of using drugs. Kleiman (2004) stated that for cocaine, the NSDUH accounts for about 10% of the actual co caine consumption, leaving the other 90% unexplained. Additionally, the NSDUH also does not examine the revenues of illicit drug markets or the crimes associated with drug use. The amount of drugs consumed, illicit drug markets, and the relationship of crime with drug use is arguably more important than simply knowing the number of people who has used drugs in the last year (Kleiman, 2004). Drug markets and crime constitute a sign ificant problem for society and are one of the top priorities of the government. Resear ch has demonstrated that, in many cases, criminal behavior (e.g., property crime, prostitution, drug sales) results directly from attempts to support a drug habit. For instance, Corman and Mocan (2000), in their timeanalysis study in New York City, found that robberies, burglaries, a nd motor vehicle theft increased during the time peri od of increased drug use. Additionally, according to a survey by the Bureau of Justice Statistics ( 2004b) of federal and state prison inmates, an estimated 17% of state prisoners and 18% of federal prisoners reported committing
54 offenses in order to support their drug habit. The next chapter will intr oduce the DUF program, examin e the criticisms brought forth by the GAO and analyze the changes made to the new ADAM program.
55 CHAPTER THREE METHODOLOGY OF DUF AND ADAM Overview of the DUF Study Underlying Assumptions of the DUF Study The DUF study was developed under the assumption that drug users are likely to be among the population of arrestees (BJS 1998, Mallender, Roberts, and Seddon, 2002). A number of researchers have supported that assumption (Brook, Whiteman, Finch, and Cohen, 1996, Wish and Gropper, 1990), and seve ral explanations have been advanced regarding this finding. First, drug users are more likely to be involved in criminal behavior because they need a certain supply of money to pay for the drugs and have to use illegal means (such as burglary and robbe ry) to obtain such monies (Petersilia, Greenwood, and Lavin, 1978, Goldstein, 1987). S econd, drug users may be more likely to engage in violent behavi or due to emotional and/or mental reactions such as aggressiveness and irritability caused by certain drugs (Bickel and DeGrandpre, 1996). Third, drug users are more likely to deal drugs to support th eir drug habit and more likely to get into turf wars amongst drug dealer s and as a result get arrested (Goldstein, 1987). There is also evidence, however, that the relationship between violent crime and drug use may not be as strong as assumed and that higher arrest rates of drug users for violent crime may instead be due to dr ug enforcement policies and procedures (Resignato, 2000). Said differe ntly, the number of drug user s who commit violent crimes may not be significantly different from the number of violent crimes committed in the whole population, but because police resources focus on drug users/drug dealers, they are
56 overrepresented among arrestees. Regardless of the reasons, the high like lihood of finding a larger number of drug users among arrestees makes prison and jail pop ulations the most suitable sampling frame when studying drug use, changes in drug patt erns, and characteristics of drug users, because arrestees were seen as the leadi ng indicator as compared to other population groups. Stated differently, the sampling fram e was based on research showing that new drug use patterns and drug paraphernalia show up first among arrestees and then spread to other population groups. With this appr oach, the DUF study filled an important gap in the research on drug use prev alence and patterns because other drug studies (e.g., the National Household Survey and Monitoring the Future) studied populations such as households and school children, among whic h drug use is much less widespread. Additionally, no other national drug study besi des DUF used bioassays as a validation technique for the self-reported drug use, ma king DUF an incredibly valuable tool for researchers studying not only preval ence and patterns of drug us e but also the validity of self-reported drug use. Data Collection in the DUF Study The quarterly data collection was based on voluntary participati on of arrestees for bot the self-report interview and the urine test. The data collection began in 1987 in 12 cities across the United States. DUF increased the number of sites steadily over time from 12 sites in 1987 to 21 sites in 1988, 22 sites in 1989 and 24 sites for the years 1990 until 1997 (National Institute of Ju stice, 1998). The 24 sites were Atlanta, Birmingham, Chicago, Cleveland, Dallas, Denver, Detroit, Ft. Lauderdale, Houston, Indianapolis, Kansas City, Los Angeles, New York (Ma nhattan), Miami, New Orleans, Omaha,
57 Philadelphia, Phoenix, Portland, St. Louis, San Antonio, San Diego, San Jose, and Washington, DC. The DUF sites were not di stributed evenly across the United States, however. Rather, the majority of sites were concentrated in the Pacific region, the southern part of United States, the middle of the United States, and the East Coast. California (4), Texas (3), and Florida (2) had multiple sites (National Institute of Justice, 1998). Number of Arrestees The number of booked male arrestees vari ed from year to year. As can be expected, with the increasing number of site s, the number of male arrestees also increased. The increase was, however, not consistent. The number of arrestees peaked in 1991 with 22,335 male arrestees and then decl ined steadily to 19,736 male arrestees in 1997. For an easier overview, Table 3.1 shows the number of male arrestees and number of sites for each DUF year (Nati onal Institute of Justice, 1998). Table 3.1. Number of Male Arrestees and Sites by Year DUF Year Male Arrestees Number of Sites 1987 2,993 11 1988 10,548 20 1989 16,186 21 1990 20,556 23 1991 22,335 24 1992 22,265 24 1993 20,551 23 1994 19,987 23 1995 20,737 23 1996 19,835 23 1997 19,736 23 1998 20,715 35 The table shows that the number of interviewed arrestees increased
58 simultaneously to the increas ing number of sites between 1987 and 1991. Within that time period, the number of interviewed arrestees increased from 2,993 arrestees and 23 sites during the first year of the program to 22,556 arre stees and 24 sites in 1991 and 22,265 arrestees and 24 sites in 1992. In 1993, the sample size declined somewhat, probably due to the reduction of sites from 24 to 23. In the following years, the number of interviewed arrestees remained relatively stable and the number of sites was 23 until 1998. Even though 1998 is considered by NIJ to be the first ADAM year, the probability sampling procedure was not implemented until the second half of the year 1999. For the purpose of this research, the year 1998 w ill be considered a DUF year because the sampling method was the same as in the previous years. Description of Variables The collected data includes information for the following five topics: (1) drug use by type of drug and for each individual offender (self-report and urine analysis), (2) dependency on alcohol/drugs (s elf-report), (3) need for tr eatment (self-report), (4) relationship between drug use and crime (offenses), and (5) indicators of self reported drug use compared to indicators of drug use according to the urin e analysis (National Institute of Justice, 1997). Arrest records were used to obtain inform ation about birth year, race, and the top charge. The DUF questionnaire included ite ms regarding participant demographic characteristics (e. g., age, gender, race, marital status, educational attainment, employment status, and living circumstances ; National Institute of Justice, 1998) and lifetime and recent drug use (within the past three days) of 22 drugs for the years 1987 till 1995 and 15 drugs for the years 1996 and 1997.
59 For each drug, arrestees were questioned regarding: (1) age of first use (2) frequency of use during the past month, (3) re cent drug use (past three days), (4) route of administration, (5) perception of past depende ncy, (6) perception of current dependency, and (7) past drug treatment. Additionally, arrestees were questioned regarding arrests during the past 12 months and whether they we re under the influence of drugs at the time of the crime. Besides questions about de mographics, drug use, and arrests, the questionnaire contained items regarding how much money arrestees spent on drugs during an average week and whether they had been in the emergency room for drugrelated incidents (National Inst itute of Justice, 1998). Additionally, NIJ used several addenda to assess topics of interest that were not typically part of the survey. For exampl e, in 1995, the survey included a heroin addendum and in 1996 a heroin and gun adde ndum. Another addendum asked arrestees about their knowledge and consideration of AIDS when using intravenous drugs (National Institute of Justice, 1997). These addenda were not collected systematically; rather, they were only collected at certain sites for a limited period of time (National Institute of Justice, 1997). Response Rate The datasets provided by NIJ only include data on arrestees that agreed to the interview and completed both the self-report interview and the urine analysis. Approximately 90% of those asked to particip ate agreed to do so and of those, 80% provided a urine sample (National Institute of Justice, 1997). There is, however, no information about arrestees who were either not asked to participate or did not consent to an interview.
60 Drug Testing The urine analysis included screening for 10 drugs (marijuana, opiates, cocaine, barbiturates, amphetamines, PCP, methadone, benzodiazepines (Valium), methaqualone, and propoxphene (Darvon) via radioimmunoassay (National Institute of Justice, 1995). To ensure that positive urine tests for am phetamines were correct, gas chromatography was used. The urine tests for all DUF sites were done at a centra l location. The outcome of all urine tests was dichotomous: either pos itive or negative for the tested drug (NIJ, 1995). DUF Sampling Procedures The sampling design was a non-probability sample guided by target numbers of interviews (250 males and 100 females) and a priority charge system. The data was collected over a period of 10 days or until the target sample number had been reached (McBride and Swartz, 1990). The problems asso ciated with this type of sampling will be explained in greater detail in the next section when disc ussing the GAO report and the criticisms brought forth by the GAO. The process of obtaining participants invol ved a number of steps. First, arriving arrestees were brought into the booking area of the facility (holdi ng cell, central booking area, or other applicable ar ea); second, the site admini strator explained the study and asked for volunteers. The site administrator to ld the arrestees only about the self-report interviews. At that time, the arrestees did not know that they would also be asked to provide a urine sample. As an incentive to participate, most sites offered candies, cigarettes, or coffee. Third, the site admi nistrator recorded the top charges for each arrestee and recruited participants based on the priority charge system. The site
61 administrator attempted to recruit volunteers wi th non-drug felony charges, followed by non-drug misdemeanor charges, and finally offe nders with drug-related charges. Fourth, each participant received an ID that matched the survey and the urine test. Fifth, the participant completed the interview. After th e completion of the survey, the participant was asked to provide a urine sample. If the arrestee provided the urine sample, a staff member would then collect the sample and la bel it with the same ID as the questionnaire (Swartz, 1990). All self-report interviews and the urine test results were gathered after each quarter by NIJ. NIJ checked the accuracy of the data by looking at the consistency of answers and undocumented codes. The NIJ staff also standardized the missing data codes across all sites. The self-report data was then merged with the urine test data and the complete dataset was re-formatted to make it usable for researchers. Finally, the datasets were made available to researchers via the Inter-University Consortium for Political and Social Research ( www.icpsr.umich.edu) (National Inst itute of Justice, 1997). The following section will discu ss the criticism s brought forth by the GAO and how these criticisms led to a change from a non-pr obability sample to a probability sample implemented in 1998. GAO Report Criticisms Although the GAO report (1993) highlighted th e strengths and unique features of the DUF program, especially the collection of drug use data from arrestees in different cities across the country and th e validation of the self-report data via urine analysis, the report also heavily criticized the non-probabi lity sample used by DUF, stating that due to its non-probability sample and lack of standard ization across sites, th e sample of arrestees
62 in the DUF study may not provide accurate in formation about arrestees in general and thus results obtained from th e study about drug use prevalence and patterns may not be accurate (GAO, 1993). The GAO report criticized DUF for three major reasons, all of which pertained to the sampli ng design: (1) the selection of booking facilities included in the study, (2) the subject sampling procedure, an d (3) the inclusion a nd exclusion criteria in selecting arrestees within the book ing facilities for the interview. (1) Criticism on the Selection of Booking Facilities The DUF program collected self-report da ta and urine tests at central booking facilities in a number of citi es across the country. The GAO report criticized that the different booking facilities re presented very different geog raphic units including entire cities or even counties, parts of a city or a county, or a central city plus additional cities. The GAO report took issue with the fact th at booking facilities encompassed very different geographic areas, as a result maki ng it difficult to draw conclusions about drug use prevalence and patterns for the great er area. The GAO report stated that the limitations of the study due to the selec tion of booking facili ties hindered the development of drug policy and programs: The re is no evidence to support generalizing partial data to an entire city or county and Caution is warra nted in using these data to determine booked arrestee drug prevalence ra tes (GAO, 1993, p. 52). This statement leads us to the second major criticism regarding the sampling design: the subject sampling procedure. (2) Criticisms of the Subject Sampling Procedure The DUF program used a judgment-based sample of arrestees based on a target number of interviews (225 male and 100 fema le arrestees per quarter per facility).
63 Starting in 1990, the sampling of the subjects was based on the 20% rule, meaning that every fifth arrestee interviewed should have been charged with a drug offense (GAO, 1993). The GAO pointed out that the 20% rule l ead to unpredictable consequences. It can either lead to an underestimation of the dr ug use prevalence among arrestees if more than 20% of the arrestee population uses drugs, or it can lead to an overestimation of the drug use prevalence among arrestees if less than 20% of the arrestees use drugs (GAO, 1993). The criticisms pertaining to the target rate of offenders and the 20% rule are closely related to the inclusion and exclusion criteria of arre stees in the DUF study. (3) Criticism of the Inclusion and Exclusion Criteria The third major sampling issue highlight ed by the GAO report pertains to the criteria used to select arresteesthe inclus ion and exclusion criter ia. The criticisms by the GAO focused on the fact that there were no standardized procedures across sites. Each DUF site used its own criteria within some set standards. For example, the DUF program established a rank order of criminal charges that was used to select male arrestees. Each DUF site, however, made its own decisions about which types of offenders would be interviewed. For instan ce, the San Diego site eliminated all misdemeanor offenders, and the Miami site onl y interviewed male offenders arrested on felony charges. Additionally, in New York (Manhattan) the booking facility limited the number of misdemeanor offenders, as a re sult decreasing the num ber of misdemeanor offenders available for interviewing. In Om aha, Nebraska, all male arrestees were interviewed (disregarding the ch arge rank-order altogether) due to the small arrestee pool available. Six sites (Denver, Detroit, Fort Lauderdale, Kansas City, Houston, and Indianapolis) had access to arrestees who committe d criminal offenses in court, in jail, or
64 in custody. The other sites either eliminated this offender group or simply did not have access to them. For female arrestees, there were no rank orders or other guidelines besides the target interview number of 100 females per quarter per site (GAO, 1993). In sum, each DUF site worked differently, and the implications of these differences were heavily criticized by the GAO concluding: Evaluat ors of the data are thereby unable to determine whether decreas ing or increasing drug use scores represent statistically significant shifts in actual drug use. An indivi duals conclusion about drug use patterns and trends must therefore rely on intuitive reactions rather than being statistically based (GAO, 1993, p.57). The GAO report and the criticisms brought forth therein had great consequences for the study. In 1996, NIJ began restru cturing the methodology of the drug use forecasting study and it was re-named th e Arrestee Drug Abuse Monitoring Program (ADAM). The purpose of the restructuring was to enable the researchers to use inferential statistics and obtain results that could be generalized to the la rger population. In 1998, NIJ implemented the ADAM study. With the implementation of the new study came a number of changes in the da ta collection process as well as the questionnaire itself. Overview of the ADAM Study The redesign of DUF/ADAM began in 1997 and consisted of three components: (1) expansion of the program, (2) implemen tation of a probability sample, and (3) implementation of a redesigne d data collection instrument (Yacoubian, 2004). The goal of these changes was to produce results that could be generalized to the target population of booked arrestees and to make the program mo re valuable and useful for practitioners and policy makers (NIJ, 1998). The first changes to the study were made in 1997 and
65 1998 by expanding the number of sites from 23 to 35 and by standardizing the sampling procedure across sites (Yacoubian, 2004). NIJ began implementing the redesigned study in the second half of 1999, but it was not fully implemented until the beginning of the year 2000 (Arrestee Drug Abuse Monitoring (ADAM)). Next, the new sampling procedure will be described, followed by a description of the expansion of the program and the new data collection instrument. Starting in 1999/2000, NIJ implemented comp lex data collection procedures that would ensure standardized outcomes. The st udy attempted to collect data on offenses, offenders, and drug use that would be a repr esentative mix of the larger population. To achieve this goal, NIJ applied the same defi nitions of catchment areas to all sites. Catchment areas are regions from which ar restees are drawn at the sites (NIJ 2000, p. 178). The definition of these catchment areas are as follows: (1) persons taken to a booking facility, and (2) the ca tchment area is a county. In addition, NIJ standardized the definiti ons of offenses because of the wide variety of definitions for certain offenses among counties and states. First, in order to ensure that all arrested persons had the sa me probability of being sampled, only booked arrestees were included in the study. This sp ecification was necessary to avoid a bias in the sample towards more serious offenses, because not all counties handle arrests the same way. Some counties arrest and book all pe rsons; others arrest but then release the persons if the offense committed was a misdemeanor. Second, NIJ standardized the definitions for terms such as misdemeanors and felonies vary. What is defined as a misdemeanor offense in some counties is cons idered a felony offense in other counties, depending on state law. Some counties distinguish misdem eanors from felonies in
66 regards to the length of jail time; other count ies distinguish misdemeanors from felonies based on the seriousness of the crime (NIJ 2000). NIJ also standardized training procedures for all sites to guarantee that the data collection procedures were not only the same at each site but also fully followed at each site. NIJ staff monitored the training as well as the data collection by implementing performance standards and feedback to each s ite after every data collection cycle. In addition, follow-up training sessions were conducted at regular time intervals to ensure that the interviewers followed the procedures strictly and consistently (NIJ 2000). Sampling Procedures 1998 In 1998, ADAM expanded the number of site s considerably from DUF (24 sites) to a total of 35 sites for adult male arrestees, 32 sites for adult female arrestees, 13 sites for juvenile male arrestees, and 8 sites for juvenile female arrestees where interviews were completed and urine specimen collected. Interviews were conducted with arrestees who had been at the facility for no more th an 48 hours. In 1998, NI J responded to one of the criticisms of the GAO by standardizing the catchment areas. The catchment area was now the same for all sites the county (NIJ, 1998). The target number of interviews plus urine specimens was 225 for adult male arrestees (all 35 sites), 100 for adult female arrestees (32 sites), 100 for juvenile male arrestees (13 sites), and 100 for juvenile female arrestees (8 sites) per quarter. However, the sampling was still conducted on a voluntar y basis as a judgment-based sample. Sampling Procedures 1999 In the third and fourth quarter of 1999, NIJ began for the first time to employ a probability sample. The sampling plan is very similar to the sampling plan for the years
67 2000 to 2003 (explained in detail below). The differences are: (1) in 1999, interview shifts lasted between 4 and 8 hours, whereas interview shifts lasted 8 hours starting in 2000 and (2) in 1999, the sampling, data collectio n, and training procedures were still in process of being fully implemented. Sampling Procedures 2000 to 2003 Beginning in 2000, NIJ fully implemented the changes to the design of DUF/ADAM. The sampling plan included a probability-based sampling design for all persons arrested and booked at each site. The sampling me thod was tailored to each county and its characteristics. In addition, the sampling strategy itself wa s tailored to each site and the sample was weighted. The goal behind this new sampling design was to represent with known probability the likelihood that a male arrestee was selected for an interview and to use that information to weight each sample case (NIJ, 1998, p. 7). Another change to the design was that now ev ery day of the week and every hour of the day were represented in the sample to en sure that no bias would occur due to the possibility that interviewers were not pres ent at the time the person was arrested and booked. Hence, all persons arrested within the last 48 hours of the beginning of the work shift were included in the popul ation from which the sample was drawn. In addition, the new design included booking facilities representing all types of facilities from small to large, urban and rural, quick re lease and slow release to avoid a bias and include all types of offenders (NIJ, 2000). The Sites Sampling Design NIJ employed one of four sampling m odels depending on the characteristics of the county. A total of four sampling models we re necessary to ensure that the sampled
68 persons would be representative of the populat ion of the whole county. First, the single jail sampling model was developed specifica lly for counties where data was collected only in one jail. Second, for counties in whic h there were between two and six jails, a stratified sampling model was employed. The stratified model allowed for data collection in all jails by determining a target number of arrestees to be sampled. In each jail the target number of a rrestees to be sampled was pr oportionate to the number of arrestees booked. Third, a strat ified cluster model was utilized in counties with more than six jails. Each jail was included in th e strata. From each stra tum one or two jails were sampled. Fourth, NIJ developed a feeder model from the stratified cluster sample for counties with booking facilities which tran sferred booked arrestees very quickly to a central holding facility. The f eeder model became necessary in those counties because the booked arrestees would have been inaccessi ble for interviews. In counties where the feeder model was utilized, booked arrestees we re interviewed at th e holding facilities and in the jails that transfe rred arrestees to those holding facilities (NIJ, 2000). Weighting Procedure In addition, the data was weighted to gua rantee that the sample was representative and generalizable to the county populati on. ADAM employed a non-traditional method to determine the weights, because at the time of the sampling it was unknown who would be arrested. Hence, the probability with which one were to be sampled for the interview was also unknown. The weighting of the data wa s based on the following assumptions: (1) Arrestees charged with more serious crimes sp end more time in jail facilities than those arrested on less serious charges, (2) Arrest ees booked at the same time of day are processed similarly; that is, they all spend approximately the same amount of time in the
69 jail before arraignment and/or transfer to an other holing facility, (3 ) The stock and flow model may mean more serious offenders will be over-represented in the stock population, while the flow sample should represent all charges. The assumption was that by using this procedure all arrestees should have a similar chance of bei ng selected into the sample, and (4) Arrestees who are booked on days where many arrestees enter the facility had a lower chance of being sel ected for the sample than if they were arrested on days when few arrestees enter the facility (NIJ, 2000, p. 181). A post-sampling stratification desi gn was developed to calculate the probability of inclusion in the sample of like groups of arrestees (NIJ, 2000, p. 181). NIJ collected census data for the total popul ation of arrested i ndividuals for each collection site. Each collection site also provided demographic, booking, and offense information about all persons arrested and booke d at the site where the interviews took place. The information obtained from each s ite was then compared to the sample collected at each site and match with th e county population. Furthermore, based on the information about the characteristics of arrestees and the county population, the population was divided and strata developed. Each stratum c ontained a certain number of arrestees and from this information the pr obability of each arrestee to be sampled compared to all other arrestees was calculated (NIJ, 2000). The weighing process can be affected by three problems: (1) Ineligibles, defined as persons that are not eligible to be part of the study, such as persons on extradition, court, and federal holds, (2) Duplicates, defined as persons that are entered into the system more than once because of aliases or mistakes of the staff, and (3) Inconsistent booking times because jails vary on their definition of booking times (intake v. time of
70 intake) and the time at which the booking time is logged by the jail (NIJ, 2000, p. 181). The Facility-Level Sampling Design The sample size for each facility was proportionate to the number of persons arrested and booked in each county. The ar rest and booking information was provided either by the site staff of the facility (if possible) or by the FBIs Uniform Crime Reports (if the information was not available from th e site staff). The number of booked arrestees to be sampled was proportionate to the numbe r of facilities in th e county and the number of persons booked at each facility. The targ et number of interv iews (per quarter) depended on the sampling design of each facility and the number of booked arrestees in each facility. Interviews were conducted ev ery day of the week for one-to-two weeks during the 8-hour-period in which the inta ke was at its highest. The number of interviewers remained constant throughout the work shift (NIJ, 2000, p. 181). As stated above, the population to be sa mpled must include all persons arrested and booked, at all days of the week and all hours of the day. To facilitate such a sampling design, each day was split into Stock and Flow periods. The stock period is the part of the 16 hours per day in which intervie wers are not present. In order to sample persons from the stock, the jail staff provi ded a list of all pers ons booked during those 16 hours of the day. From that list, wh ich is ordered by the booking time, booked arrestees were sampled depending on the target number in even intervals. The stock interviewers stopped working once they reached the target number. Once determined, the number of interviewers did not change at the sites. This wa s mostly done due to practical reasons to make the implementation of the new sampling procedures easier and to avoid problems with the weighting procedure (NIJ, 2000).
71 The Flow period represented the part of the day during which arrestees were interviewed, which was eight hours each day. The Flow hours were tailored to each site, with the goal to have interviewers pres ent during the 8 hours of the day in which the number of arrested and booked persons was the highest. Persons were sampled from each period, the Flow and the Stock. Flow in terviewers work continuously during these 8 hours regardless of whether the target number of interviews for the day has been reached. After each interview, the ADAM staff sampled the persons booked closest to the time of the completion of the interview (NIJ, 2000, p. 181). The sampling was carried out for one-to two-weeks at the time (NIJ, 2000). Face-Sheets The ADAM staff filled out a face-sheet (i nformation about booking and charges) for each arrestee sampled, regardless of whet her the arrestee agreed to interview and regardless of whether the arrestee was avai lable for an intervie w. This procedure decreases potential bias that could occur and enables the site staff to check whether the correct sampling procedure has been followed (NIJ, 2000). In facilities where arrestees are transferred to other facilities very qui ckly or released very quickly it sometimes became impossible for the ADAM staff to meet the quota for the stock interviews. In such cases, it was possible to have separate interview periods for stock and flow (NIJ, 2000). Response Rate The response rate for male arrestees in 2000 was 56.3%. Of the non-respondents, 13.6% declined to interview, 22.8% were not av ailable (either releas ed, in a holding cell, or in medical unit), and 7.3% were not asked to participate. Of those 56.3% that agreed to
72 interview, 89.9% also provided a urine sa mple. Similar to the year 2000, the overall response rate for male arrestees in 2001 was 55.1%. Of the arrestee s that did not respond, 12.2% refused to interview, 22.5% were not available (either releas ed, in a holding cell, or in medical unit), and 10.2% were not asked to participate for various reasons. Of those who interviewed, about 90.5% percent also pr ovided a urine sample. It is difficult to compare these response rates to that of DUF because DUF used a convenience sample and therefore contains no information a bout non-respondents. The DUF documentation also does not include information on how ex actly the response rate was calculated. Arrestees who were available for the in terview might differ in their drug using behaviors from arrestees who were available for the interview. For instance, arrestees who had already been released were more likely to have committed misdemeanor offenses and/or were less likely to have used drugs. Similarly, arrestees who were dangerous or violent might also differ in their drug using behaviors as compared to arrestees who were compliant with law enforcement. There exists no data on these unavailable arrestees that woul d allow for a comparison. Although this is speculative, it is possible that the probability sample of ADAM might have included ar restees very similar to the arrestees in the DUF program, which is arrestees available at the time of the interview and who were willing to participate. The ADAM data also suggests that fo r both years, 2000 and 2001, the response rate of arrestees who were asked to pa rticipate in the ADAM study varied greatly depending on the site. In 2000, there were 35 site s at which data was collected for at least one quarter. Half of these sites had a respons e rate of 81% or higher for arrestees who were available and asked to participate. Ov erall, the response rate varied between 5.9%
73 and 40.1%, with Fort Lauderdale being the lowe st and the Charlotte Metro area being the highest (NIJ, 2003). Simila rly, the percentage of interviewees who agreed to the urine test also varied considerably be tween 74.7% and 96.6%, with th e Albany/Capital Area having the lowest rate and Fort Lauderdale having the highest rate (NIJ, 2003). These response rates were lower as compared to the DUF data, which had a 90% response rate for interviews. This is about 9% higher than the averag e response rate for the ADAM program (U.S. Department of Justice, 1998). Again, for the DUF data there is no data that would allow for a comparison of the arre stees who refused to the arrestees who participated. The differences in the res ponse rates for DUF and ADAM can likely be attributed to the differences in the sa mpling method. During the DUF study, the local DUF interviewers had some discretion in whom to ask to volunteer. The sample was not a probability sample of all booked arrestees rather the intervie wers would ask for volunteers (based on certain cr iteria described earlier). A dditionally, many sites offered incentives to the arrestee to increase their wi llingness to participate. This might explain the higher response rates for DUF as compared to ADAM. Contrary to DUF, the ADAM program in cluded a face sheet for all arrestees selected for the sample. The face-sheet incl uded information about gender, race, age, residence, charge type, a nd charge seriousness. A st udy by Myrstol and Langworthy (2005) used this face-sheet data to compare the realized samp le of arrestees in Anchorage with the arrestees who were not included in the sample. The authors showed that despite a relatively high attri tion rate, both the male and female realized sample was representative of the population of booked arrestees.
74 Drug Testing Table 3.2 shows cut-off levels and detection period for the drugs examined in the ADAM study. Contrary to DUF, all ADAM sites sent the urine specimen to a central facility where it was tested for 10 drugs, as a result standardizing the drug testing. The table includes cut-off levels and detection periods for each drug. The cut-off level is defined as the amount of the drug in nanograms per milliliter below which the amount is considered undetectable and the result is negative (NIJ, 2003, p. 16). The detection period is defined as the number of days af ter ingestion during which the drug can be detected in the body (NIJ, 2003, p. 16). Table 3.2. Cut-off Levels and Detection Periods for 10 Drugs Drug Cut-off Level Detection Period Cocaine 300ng/ml 2-3 days Marijuana 50ng/ml 7 days (infrequent use) 30 days maximum (chronic use) Methamphetamine 300ng/ml 2-4 days Opiates 300ng/ml 2-3 days PCP 25ng/ml 3-8 days Amphetamines 1,000ng/ml 2-4 days Barbiturates 300ng/ml 3 days Benzodiazepines 300ng/ml 2 weeks maximum Methadone 300ng/ml 2-4 days Methaqualone 300ng/ml 10 days maximum Propoxyphene 300ng/ml 3-7 days Sites The ADAM study started with a considerable increase in the number of sites and states included in the study compared to DUF. ADAM started out with 35 sites in 1998, 12 sites more than the DUF study had in 1997. ADAM then increased the number of sites to 39 sites in 2001. The 39 sites were Alba ny/Capital Area (NY), Albuquerque (NM),
75 Anchorage (AK), Atlanta (GA), Birmingham (AL), Charlotte-Metro Area (NC), Chicago (IL), Cleveland (OH), Dallas (TX), Denver (C O), Des Moines (IA), Detroit (MI), Ft. Lauderdale (FL), Honolulu (HI), Houston (TX) Indianapolis (IN), Kansas City (MO), Laredo (TX), Las Vegas (NV), Los Angele s (CA), New York (NY), Miami (FL), Minneapolis (MN), New Orleans (LA), Okla homa City (OK), Omaha (NE), Philadelphia (PA), Phoenix (AZ), Portland (OR), Sacramento (CA), Salt Lake City (UT), San Antonio (TX), San Diego (CA), San Jose (CA), Seat tle (WA), Spokane (WA), St. Louis (MO), Tucson (AZ), and Washington, DC. Although NIJ added a considerable number of sites (14) to the ADAM study, the distribution patterns is similar to the distri bution of sites in the DUF study. The majority of sites were concentrated in the Pacific re gion, the southern part of United States, the middle of the United States, and the East Coas t. California (4), Texa s (4), Florida (2), Missouri (2), and Arizona (2) had multiple sites. The ADAM sites were distributed across 26 states. Number of Arrestees The number of arrestees increased steadily between 1998 and 2001. Table 3.3 represents the number of male arrestees for each year congruent with the increasing number of sites where data was coll ected. Between 1999 and 2001, the number of interviewed arrestees steadily increased without an increase in the number of sites. By 2001, the sample size was nearly twice that of the last DUF years. Table 3.3. Number of Interviewed Arrestees and Number of Sites by Year Year Male Arrestees Number of Sites 1999 31,210 35 2000 35,784 35 2001 39,406 39
76 Description of Variables Redesigned Data Collection Instrument Similar to DUF, arrest records were us ed to obtain information about race, gender, birth year, and top charge (NIJ, 2003). Also similar to DUF, ADAM collected information on demographic characteristics, recent and long term drug use, and drug use patterns (NIJ, 2003). There were, however, a number of modi fications to the ADAM questionnaire. The ADAM questionnaire was expanded to include screening questions for drug dependence and need of treatment, informati on about drug use within the previous year via the calendar method, drug market activity, and drug treatment by drug within the past year. The drug dependence and need of treat ment screening consisted of questions regarding the frequency of drug use, thoughts about drug use, reasons of drug use, intentions to reduce or stop dr ug use, and objections from friends or family to the drug use (NIJ, 2003). Calendar Method The implementation of the calendar me thod was probably the most important change in the questionnaire (NIJ, 2003). The main purpose of the calendaring method was to collect drug use information within th e past year and to improve the validity of self-reported drug use among arrestees. The basi c idea was to examine annual patterns of drug use and related behaviors over time because asking arrestees about drug use only during the last 30 days does not capture the complexity of drug use patterns (NIJ, 2003). The method of calendaring was utilized to help the respondents remember drug using behaviors over such a long period of time. To help arrestees remember what drugs they
77 used and how frequently they used them, in terviewers first asked arrestees about major events during the last year (e.g., birthday, holid ays, family events, and life events) with the purpose of conceptually dividing the ye ar into manageable time periods for the respondents. The major events reported by the arrestee serve as an anchor helping them remember their drug using behavior at that time. Using this method, data on drug use patterns was collected for each drug month by month (NIJ, 2003). As stated above, one of the main goals of the calendaring method was to improve the validity of self-reported drug use. Whet her this goal was accomplished has not been extensively assessed. A study by Yacubian (2003b) concluded that the calendaring method did not influence the validity of self-re ported drug use. The results demonstrated that the concordance rate be tween self-reported drug use and the urine analysis tests for the ADAM years during which the calendaring method was used (2000) were similar to the concordance rate for those years during which the calendaring method was not used (1999). Specifically, for the year 1999 (no cal endaring) the concordance rate was 64% for marijuana, 48% for cocaine, and 8% for heroin. Similarly, in 2000 (calendaring was used) the concordance rate was 59% for mar ijuana, 47% for cocaine, and 6% for heroin. Thus, the calendaring method did not improve the validity of self-re ported drug use. The modified ADAM questionnaire also included a section on drug market activity including questions about drug purch ase patterns, place and neighborhood of purchase, and difficulties purchasing drugs (N IJ, 2003). More specifically, arrestees were asked a battery of questions about how of ten they bought drugs, how much they bought, who they bought drugs from, how they contac ted the drug dealer, the relationship of buyers and sellers, how much they paid for the drugs, the payment method, the
78 neighborhood in which they bought the drugs, how they located drugs, difficulties in locating and buying drugs, and why drug pur chases failed. This information was collected for each drug (NIJ, 2003). How accurate this information is unknown. It could be expected, however, that recall decay and fo rward telescoping might factor into this issue. Drug Treatment Data Finally, the modified ADAM questionnair e included a section on drug treatment within the past year (NIJ, 2003). Arrestees we re asked about drug treatment in inpatient and outpatient treatment facilities. This information was obtained for each drug and each month for the last 12 months (NIJ, 2003). Although the data obtained from the ADAM sample are now said to be representative of the population of booked arrestees within the catchment areas, problems and limitations wi th sampling arrestees and other populations have likely led to limitations on the represen tativeness of the data, a number of which are similar to limitations of the DUF data. Thes e limitations that a pply to both DUF and ADAM and their implications will now be discussed. Methodological Issues of DUF and ADAM The DUF and ADAM programs had similar goals. The redesign of the DUF program and the implementation of ADAM we re meant to improve the program and increase the quality of the data. Although the program made every effort to meet these goals, the problems associated with interviewing arrestees acr oss the United States posed serious limitations. The major goal of ADAM was the identification of the drug use prevalence and patterns among arrestees. The goal was to produce results that were representative of the arrestees within the catchment area. This may not be the case,
79 however. Even though ADAM used a sophistic ated, probability-based sampling method to determine the sites within counties and the persons to be interviewed within sites, the total sample was determined by the cost stru cture of each site (Yang, 2004). Thus, similar to DUF, sampling in the ADAM program was not uniform for all jurisdictions. Some sites excluded booking facilities with a low volume (e.g., Cleveland). Other sites interviewed arrestees in only a few of th eir several booking facilities (Birmingham). Additionally, some sites only interviewed felony offenders (e.g., Chicago) and males (e.g., Sacramento) (NIJ, 2003). For both DUF and ADAM, it is likely that the exclusion of booking facilities with certain characteristics influenced the results across sites because different booking facilities might ha ve different arrestees with regard to demographic characteristics and drug use (Yacoubian, 2000). Availability of Arrestees. Additionally, both DUF and ADAM were only able to sample arrestees who were held long enough in the facility to be interviewed. This is typically true for more serious offenders and indigent offenders who do not have the financial means to make bail. Research sugge sts that felony offenders use different drugs (crack, cocaine, heroin) than misdemeanor o ffenders (marijuana, ecstasy). For instance, Webb and Delone (1996) found that felony arrestees were significantly more likely to test positive for cocaine use. There is also a substantial am ount of research showing that indigent individuals use differe nt drugs as compared to indi viduals with greater financial means. A study by Peters, et al. (2002) usi ng the ADAM data suggested that arrestees living under the poverty line were more like ly to use opiates and benzodiazepines as compared to arrestees liv ing above the poverty line. Additionally, crack cocaine is much more widespread among poor black drug
80 users than any other populat ion group. For instance, Ha rtley, Maddan, and Spohn (2007) found that among black persons facing drug ch arges, 85.3% used crack cocaine. Among white offenders, only 5.8% faced charges for th e possession and/or sale of crack cocaine. White offenders were more likely to use powder cocaine. Other re search supports the notion that crack users are disproportionate ly more often minorities and poor as compared to users of other drugs (Beckett, Nyrop, and Pfingst, 2006). Although the ADAM weighting procedure atte mpted to ensure that the different types of offender populations were represen ted equally in the sample, the weighting procedure was not able to solve the probl em associated not having access to certain offender groups because they are released immediately or who dont make it to the facility in the first place because the arresting officer decides to issue a citation instead of an arrest (Yacoubian, 2000). These problems associated with collecting data from arrestees were present for both DUF a nd ADAM. Thus, both samples should contain information that is representative of more serious and indigent offenders. Myrstol and Langworthy (2005) found that for the ADAM data from Anchorage, the realized male arrestee sample was indeed slightly biased towards felony offenders. This might also be similar for the DUF data. Determining the Specific Drugs Used Within a Certain Jurisdiction. Another goal of ADAM was to determine the specific drugs used within a certain jurisdiction (NIJ, 2003). Yacoubian (2004) sugge sts that this goal might not have been met due to the short time period during which interviewers collected data. At most facilities, data collection took place 56 days in one calendar ye ar (14 days per quarter). Similarly, DUF also collected data for each quarter of th e calendar year. The difference was that DUF
81 used a predetermined number of interviews (f or male offenders it was 225). Once the 225 interviews were completed, data collection st opped. When ADAM was redesigned, the assumption of the data collection by NIJ was that if these 56 days were spread out evenly across the calendar year, all type s of drugs used would be detected, and the magnitude of a certain drugs representation could be computed. This assumption is somewhat problematic. Although it is possibl e that the different types of drugs used in a certain jurisdiction might be detected within the inte rviewing period, the rela tively short period of data collection per year might not have allowed ADAM to accurately determine the magnitude of the drugs repr esentation in that jurisd iction (Yacoubian, 2004). For instance, a certain drug could have started to become more popular during the end of the quarterly data collection period and then leve led out before the new data collection period started. Regardless of whether that is the cas e or not, it is a problem that applies to DUF as well as to ADAM, which might have resulted in data that is not substantially different. Table 3.4 presents the sample size for each qua rter by site and year. It appears that the DUF sample for each quarter and each site is very consistent Thus, both studies collected data for a certain period of ti me and a limited number of arrestees. Number of Interviews for Each Quarter. Some other patterns emerge from the data. In general, the number of intervie ws conducted during each quarter in the DUF sample is substantially greater than in the ADAM sample. The exception is Phoenix, where the sample size is somewhat greater for the ADAM data. The ADAM program did not collect data for 2001 in Miami and only coll ected data for the first quarter of 2001 for Dallas. Additionally, the ADAM sample for the first quarter of the San Antonio sample has only 65 cases, which is less than half as many as for the other three quarters.
82 The DUF data also has some sites that did not collect data for each quarter and year. Portland did not collect data for the fourth quarter in 1997. Also, the DUF sample for three sites (Dalla s, Indianapolis, and Phoenix) ha s no data for one quarter for 1998 and two sites (Miami and San Jose) are mi ssing data for two quarters of 1998. The DUF reports do not include an explanation why some sites did not collect data for a certain quarter. The sample sizes for DUF, however, were somewhat greater than for ADAM (with the exception of Phoenix) which means th at the sample size is not a major problem for DUF. The sample sizes for DUF are probab ly greater because of the sample target number. Each site attempted to collect 225 in terviews per quarter for male arrestees. During ADAM, the sites collected interviews within a certain time period. Additionally, due to the fact that the ADAM questionnaire was much longer than the DUF questionnaire not as many interviews could be completed during ADAM. Table 3.4 presents the sample sizes for each quarter for DUF and ADAM. These sample sizes were aggregated to represent one year. Finally, the analysis includes two years for DUF and for ADAM. The data for th e two years for DUF were aggregated and the data for the two years for ADAM were aggr egated, resulting in a fairly large sample for DUF and ADAM. Specifically, the sample sizes for DUF range from 1,272 to 1,959 with the lowest sample size for Miami and the highest sample size for New Orleans. The sample sizes for ADAM ranged from 535 to 2,850. For ADAM Miami had the fewest cases and Phoenix had the most cases. Overa ll, both samples have a large enough sample size for the analysis (as will be described in the following chapter).
83 Table 3.4. Sample Size for Each Quarter by Site and Year DUF ADAM Quarter 1997 1998 2000 2001 Dallas Q1 247 199 182 178 Q2 240 0 266 0 Q3 245 193 0 0 Q4 248 175 76 0 Year 980 567 524 178 Total 1,547 702 Denver Q1 227 248 125 175 Q2 241 247 150 171 Q3 244 215 166 173 Q4 240 246 177 182 Year 952 956 618 701 Total 1,908 1,319 Indianapolis Q1 224 231 135 191 Q2 247 239 146 180 Q3 241 0 173 175 Q4 225 138 176 186 Year 937 608 630 732 Total 1,545 1,362 Miami Q1 222 219 156 0 Q2 196 199 182 0 Q3 219 0 197 0 Q4 217 0 0 0 Year 854 418 535 0 Total 1,272 535 New Orleans Q1 246 246 147 158 Q2 249 247 156 154 Q3 249 230 150 157 Q4 250 242 151 163 Year 994 965 604 632 Total 1,959 1,236 Phoenix Q1 247 195 251 411 Q2 250 195 331 381 Q3 248 0 339 389 Q4 238 238 345 403 Year 983 628 1,266 1,584 Total 1,611 2,850
84 DUF ADAM Quarter 1997 1998 2000 2001 Portland Q1 243 215 101 190 Q2 253 212 155 166 Q3 149 178 165 181 Q4 0 151 180 157 Year 645 756 601 694 Total 1,401 1,295 San Antonio Q1 230 228 65 124 Q2 234 227 124 163 Q3 230 228 153 137 Q4 237 226 178 178 Year 931 909 520 602 Total 1,804 1,122 San Jose Q1 215 225 124 183 Q2 221 209 127 166 Q3 214 0 141 186 Q4 235 0 134 187 Year 885 434 526 722 Total 1,319 1,248 Participants are Volunteers Most importantly, similar to DUF, the ADAM sample also consists of volunteers because se lf-selection bias applies to ADAM as well. Even though a probability sample is drawn fr om the persons arrested, arrestees still had the choice whether to agree to the interview a nd the urine sample or not. This means that even a probability sample is made up of volunteers, which may differ in their characteristics and drug use from persons who denied participa tion in the study. For ADAM, refusal rates varied considerably ac ross sites, with a low of 5.9% in Fort Lauderdale to a high of 40.1% in the Charlott e-Metro area. This data was not available for DUF. However, the annual reports from the DUF years 1997 and 1998 estimate the refusal rate to be around 10% for the interviews It can be expected that the refusal rate
85 for DUF also varied depending on the site. It is possible that the refusal rates of DUF were similar to ADAM because both studies collected data on a sensitive topic in a comparable setting (jail) from booked arrest ees. Unfortunately, this question cannot be assessed in depth due to a lack of data for the DUF study. It is likely, however, that selfselection bias might have influenced the data collected in both studies Also, a number of arrestees did not provi de urine specimens. NIJ estimated that approximately 20% of all interviewed arrest ees in the DUF program did not provide a urine sample (U.S. Department of Justice, 1998). For the ADAM sample, refusal of urine analysis ranged from a high of 25.3% in Albany/Capital Area, New York to a low of 2.1% in Oklahoma City, Oklahoma. The average refusal rate was 10.8%. Research suggests that arrestees who refused to provide a urine specimen are more likely to be drug users compared to those who agreed to the urine sample (Chen, Stephens, Cochran, and Huff, 1997). This causes an underestimation of drug use among offenders and influences the results and thereby the policy implications drawn from such results. To reiterate, this shortcoming is similar for DUF and ADAM, and for ot her studies that use bioassays. Limitations of Urine Testing Also applicable to both DUF and ADAM are the limitations of urine testing. First, drug testing via urine analysis can only assess drug use for the last 2-4 days, depending on the drug. Seco nd, some drugs are harder to detect than others, which means there is an unknown amount of false negatives. ADAM improved the quality of the urine testi ng procedure by having all urin e samples shipped daily to a central lab. This allowed for an immediat e processing of the urine specimen and the testing was completed at the same lab, as a result providing consistent testing. In
86 comparison, during the DUF program, urine samples were shipped to the lab weekly or, at some sites, the urine samples were processed in the lab on site. Riley, Lu, and Taylor (2000) examined the impact of the differen tial shipment procedures and found that the drug testing results were very consistent. Specifically, th e concordance rate was 99%. Thus, it can reasonably be believed that the de lay in shipping did not significantly change the drug testing outcomes. For an easie r overview, Table 3.5 summarizes the methodology, procedures, and response rates for the DUF and ADAM data. In sum, there are several limitations th at are applicable to both DUF and ADAM that can potentially have a substantial influence on the drug use information obtained from the arrestees. The crucial question is whet her these limitations result in similar data in the sense that the drug use information co ntained in both datasets is not substantially different. Drawing on the two studies c onducted by NIJ (1990) and Myrstol and Langworthy (2005) this possibility cannot be ruled out. Based on the limitations faced by both studies and previous research it is hypothesized that the drug use estimates contained in DUF and ADAM are not substant ially different despite differences in the sampling method. The following chapter will describe the data and research strategy used to assess the current research question. Du e to the fact that the anal ysis strategy employed in the current study is rarely used in social scienc es, the analytic strate gy will be explained by using data from Dallas, Texas, as an example. The results of each step and how they lead to the next step will be discussed. Then, the results for all nine sites will be presented and discussed.
87 Table 3.5: Comparison of DUF and ADAM Variable DUF 1997/1998 1998/Q1,Q2 1999 ADAM 2000/2001 Annual Cost appr. $1 million appr. $8.4 million Sampling Method Non-probability sample Non-probability sample Probability sample (Availability-Judgmental sample) (Availability-Judgmental sample) Standardization of Sites No Yes Yes Number of Sites Beginning Number 12 36 39 Final Number 23 36 40 Representative for U.S. No No No Selection of Booking Facilities Representative for U.S. No No No Consistent Catchment Area No No Yes Facility-level sampling Sample Size Target number of interviews Target number of in terviews Propor tionate to persons arrested Inclusion/Exclusion Criteria Pr iority charge system Priority charge system Random sample Twenty percent rule Twenty percent rule from booking lists Urine Analysis across sites Central Lab and at Sites Ce ntral Lab Central Lab Shipment Procedures Every two weeks Daily Daily Number of Drugs Between 3 and 10 Between 3 and 10 10 Response Rate Interview 90% 90% 56% Urine Specimen 80% 80% 91% (of those who interviewed)
88 CHAPTER FOUR DATA DESCRIPTION AND ANALYTICAL STRATEGY Major Research Question of the Current Study To reiterate, the m ajor research ques tion of the current study is: Are the drug estimates of the Drug Use Forecasting Program (DUF), using a non-probability sample, and drug estimates of the Arrestee Drug A buse Monitoring Program (ADAM), using a probability sample, substantially different or ar e they similar enough that they can be said to be equivalent? This research question is based on the criticisms of the sampling procedure of the DUF study by the GAO, that is it is unknown whether the drug use data contained in the non-probability sample of DUF can be said to be representative of the drug using behaviors among arrestees. This que stion has not been wi dely explored. Yet, researchers have used both data sets, DUF and ADAM, to conduct cross-sectional and longitudinal research and make policy recommendations. Assuming th at the probability sample of ADAM resulted in a sample and drug use information that can be generalized to the population of arrestees in that specific geographic area (catchment area) the drug use information in the ADAM data will be used as the baseline data and compared to the DUF data. Data The datasets used in this study were obta ined from the Interu niversity Consortium for Political and Social Research (ICPSR) and the National Institute of Justice. For DUF, data was available from 1988 until 1998. For ADAM, data was available for the years 2000 and 2001. The DUF data includes data for males and females for the years 1988
89 1997 and for males, females, boys, and gi rls for the years 1994 till 1997. The ADAM data includes data for males and female s for 1998 and 1999 and for males only for 2000 and 2001. The current study will only use data on male arrestees because those data are available for all years for DUF and ADAM and because the ADAM female sample was not a probability sample (NIJ, 1998). Both DUF and ADAM utilized three data sources: (1) arrest records, (2) face-to-face interviews, and (3) urine specimen (NIJ, 1998). Which Years are Included in the Analysis? For the analysis, the ADAM data includes the years 2000 and 2001 and the DUF data includes the years 1997 and 1998. The most appropriate years would be 1998 and 1999 for DUF because those would be the closest to the ADAM years. It is important to use the most consecutive years for the analysis because changes in drug use preval ence and patterns over time could bias the analysis. Research from DUF and ADAM has shown that the popularity of a certain drug varies over time. For instance, Golub and Johnson (2001) demonstrated that marijuana use among arrestees changed substantially between 1991 and 1996. Specifically, in 1991 about 25% of youthful arrestees used mariju ana, whereas five y ears later 57% reported the use of marijuanaa significant increas e of 30%. Additionally, Golub and Johnson (1997) found that crack cocaine use declined in Fort Lauderdale from 50% in 1987 to 19% in 1990. This means that within three years crack use dropped by 31%. Using the most consecutive years is thus crucial fo r the current analysis. Similarly, a study by Yacoubian, et al. (2004) implied that meth amphetamine use dropped by more than 50% between 1995 and 1996 for several citie s included in the DUF program. Other drug use surveys also support this re sult. The NSDUH demonstrates that in 1997 approximately 32.9% of persons 12 years or older had used marijuana in their
90 lifetime and 5.1% had used it within the pa st month. In 2001, 34.2% of persons 12 years or older had used marijuana in their lifetim e and 4.8% had used it within the past month. This demonstrates that the percentage of persons who had ever used marijuana increased by 1.3%, but marijuana use in the past month decreased by .03% (NHSDA, 1997, 2001). These examples show that drug use preval ence is not stable even within a few years. This is of great importance for the current study because the goal is to determine whether the drug use information contained in DUF and ADAM is comparable. Natural fluctuations in the data can have a great imp act on the findings. Thus, it is crucial to use the most consecutive years in the current an alysis. The year 1999 cannot be included in the analysis, however, because it represents a hybrid year. Part of 1999 was a DUF year in which data was collected from a non-probabi lity sample, but during the latter part of 1999 data was collected with the newly impl emented probability sampling method. The year 1999 is divided into quarters 1 and 2 for DUF and quarters 3 and 4 for ADAM because the ADAM report from 1999 states that the new probability sampling procedure was established in 1999. The main problem is th at there appears to be no consistent date at which the probability sample was implemente d for each site. The reports states: in the third and fourth quarters of 1999, ADAM sites began implementing new sampling procedures (NIJ, 1999, p. 7). This statemen t implies that the realization of the probability sampling procedure did not occur at the beginning of the third quarter at all sites. Rather, the probability sample was established at different times across sites. Also, the 1999 data does not include face sheet data as is provided starting in 2000. The face sheet data shows the basic demographic data for all arrestees selected into the sample.
91 This allows for a comparison of the arrestees who refused or could not be located to those arrestees interviewed. Thus, the lack of f ace sheet data and the statement in the 1999 report imply that 1999 is not an appropriate year for the purpose of the current study because it is important that the comparis on is conducted between the non-probability sample and the probability sample to ensu re the validity of the analysis. Thus, the analysis will consist of 1997 and 1998 for DUF and 2000 and 2001 for ADAM. Which Variables are Used in the Analysis? As described in the previous chapter, the main goal of DUF and ADAM was to collect data about drug use prevalence and patterns among arrestees. Thus, the curr ent analysis will focus on the variables measuring drug use. The current study will also examine the demographic profile of DUF and ADAM to assess whether differences in the demographic profile also reflect in differences in drug estimates. Table 4.1 pres ents and defines the demographic variables use in the study. Table 4.2 shows the codings and descriptions of the drug use variables included in the current analysis. Table 4.1. Codings & Descriptions for Demographic Variables Variable Name Variable Description Coding Race Race of Arrestee Black = 1; White = 2; Hispanic = 3; Other = 4; Not obtained = 99 Employment Employment Status of Arrestee Full Time = 1; Part Time = 2; Unemployed = 3; Other = 4; Not obtained = 99 Highschool Graduate Arrestee has Graduated from Highschool Yes = 1; No = 0 Offense Category Arrestees Highest Charge Violent = 1; Property = 2; Drug = 3; Other = 4 Age Arrestee Age at the Time of the Offense >18
92 Table 4.2. Codings & Descriptions for Drug Use Variables Variable Name Variable Description Coding MJ Urine Test Result for Marijuana Yes = 1; No = 0 COC Urine Test Result for Cocaine Yes = 1; No = 0 OP Urine Test Result for Opiates Yes = 1; No = 0 MJ72 Self-Reported Marijuan a Use Within the Past 72 Hours Yes = 1; No = 0 COC72 Self-Reported Powder Cocaine Use Within the Past 72 Hours Yes = 1; No = 0 CRK72 Self-Reported Crack Cocaine Use Within the Past 72 Hours Yes = 1; No = 0 HER72 Self-Reported Heroin Use Within the Past 72 Hours Yes = 1; No = 0 PCP72 Self-Reported PCP Use Within the Past 72 Hours Yes = 1; No = 0 AMPH72 Self-Reported Amphetamine Use Within the Past 72 Hours Yes = 1; No = 0 BARB72 Self-Reported Barbiturate Use Within the Past 72 Hours Yes = 1; No = 0 EVERMJ Self-Reported Drug Use Ever Used Marijuana Yes = 1; No = 0 EVERCOC Self-Reported Drug Use Ever Used Powder Cocaine Yes = 1; No = 0 EVERCRK Self-Reported Drug Use Ever Used Crack Cocaine Yes = 1; No = 0 EVERHER Self-Reported Drug Use Ever Used Heroin Yes = 1; No = 0 MJALL Urine Test Result and Self-Reported Drug use for Marijuana (Combines MJ, MJ72, and EVERMJ) COCALL Urine Test Result and Self-Reported Drug use for Cocaine (Combines COC, COC72, CRK72, EVERCOC, and EVERCRK) OPALL Urine Test Result and Self-Reported Drug use for Opiates (Combines OP, HER72, and EVERHER) The differences and similarities of th e demographic profile and the drug use frequencies will be compared based on the percentage change between the DUF years 1997 and 1998. The percentage difference represents the difference as a percentage of the baseline value (ADAM data). To calculat e the percentage difference the difference between the two proportions is calculated and then divide d by the baseline value. The formula for calculating the percentage difference of two proportions is:
93 %Diff = (p1-p2)/p1 For example, 32.7% of the ADAM arrest ees and 43.3% of the DUF arrestees tested positive for marijuana. The calculation would be: %Diff = (.434 .327)/.327 %Diff = .327*100 %Diff = 33% The percentage difference of the dem ographic variables for the DUF and ADAM data can be considered a straightforward way to get a preliminary overview of the magnitude of differences between DUF and ADAM. To further assess the question of whether the drug use information contained in the DUF and ADAM samples are substantially different, the current study will conduct equivalence testing, an analysis technique ty pically used by clinical researchers who are examining whether two drugs or treatments produce outcomes that are not substantially different, meaning that this difference woul d be of clinical importance. Equivalence testing has been used for a long time by medical researchers and in cl inical trials. Since it was first introduced for the use in psychol ogical research by Rogers, et al. (1993) equivalence testing has also become more wide spread in other fields as well (Hersen and Gross, 2008). For instance, Epstein, et al. (2001) employed this technique to assess whether web surveys would produce results si milar to the traditional paper-and-pencil surveys. Leff, et al. (2005) employed equivalence analysis to compare the quality of services across different hea lth care providers. The following section will now in detail describe equivalence analysis, why it is appropriate for the current study, what the possible outcomes are, and how they will be interpreted. Additionally, since this type of
94 analysis is very rarely used by social scientists, an example will be presented for a better understanding of the statistical technique. Introduction to Equivalence Analysis As stated earlier, the main goal of this study is to assess whether the drug use information in DUF and ADAM are substantially different or whether they can be said to be equivalent. Research suggests that equivalence testing is an appropriate method when comparing outcome measures for two differe nt groups or samples (Hauck and Anderson, 1986; Rogers, et al. 1996; Stegner, Bo strom, and Greenfield, 1996; Tryon, 2001). Equivalence testing has been used extens ively in biomedical research where the main goal is to determine whether two drugs or treatments produce equivalent outcomes (Cleophas, Zwinderman, Cleopha s and Cleophas, 2009). For this purpose, the researcher compares the outcomes and side effects of the new drug/treatment to the established drug/treatment. Equivalence testing is ba sed on the underlying assumption that two treatments or drugs will alwa ys lead to some differences in the outcome. The important question is whether these differences in the outcome are of clinical and/or practical importance (Pocock, 2003). For instance, re searchers are introducing a new cancer treatment that has fewer side effects. The ne w treatment with the fewer side effects is only useful, however, if it is equally effective in treating th e illness as compared to the standard treatment. Thus, the goal is to asse ss the equivalence of these two drugs with regard to the treatment effect or outcome. Similarly, the current study seeks to de termine whether the drug use information in DUF and ADAM are equivalent despite differences in the sampling methods. For the purpose of the current study, the ADAM data will be considered to be the established
95 treatment and DUF as the new treatment. Whereas clinical studies aim to assess the effect a new treatment has as compared to the standard treatment, the current study examines whether the profile of percentage outcomes (the ef fect) of 14 drug use variables is equivalent for DUF and ADAM. Equiva lence does not mean exactly the same, rather it refers to the absence of a meani ngful difference (European Medicines Agency, 2000; Rogers, et al. 1993; Allen and Seam an, 2006; Tryon and Lewis, 2009). Stated differently, the question is whether the profile of percentage outcomes of the 14 drug use variables is comparable (H ersen and Gross, 2008). Hersen and Gross (2008) suggest that the determination of whether effects ar e comparable should be made based on real world outcomes (p. 216). Equivalence can be assessed by cons tructing the confidence interval for proportions, in the current st udy drug use proportions. Specifica lly, an equivalence limit or margin (to + ) is chosen defining how different the groups or samples can be before the difference is of practical im portance (European Medicines Agency, 2000; Tryon, 2001; Wiens, 2001). After defining th e equivalence margin, the two-sided confidence interval is calculated, represen ting the range of differences between two samples. If the two-sided confidence interval lies within the equivalence limit, then the two groups or samples can be said to be e quivalent or comparable (Allen and Seaman, 2006; European Medicines Agency, 2000; Hersen and Gross, 2008; Rogers, et al., 1993). The ADAM data will be used as the baseline value and to calculate the equivalence margin because it constitutes a representative probability sample. Equivalence analysis as de scribed by Rogers, et al. (1993) and Allen and Seaman (2006) consists of two parts: traditional null hypothesis test and equivalence test using the
96 confidence interval approach proposed by We stlake (1981). The researcher can then evaluate the results of these two tests and determine whether the two groups/proportions are substantially different or equivalent. This section will describe how these two tests are carried out and how the out comes can be interpreted. Traditional Null Hypothesis Testing Examining the differences between two groups has traditionally been based on null hypothesis testing. Null hypothesis testing might, however, not be the best strategy when comparing differences between two gr oups or samples. In the case of null hypothesis statistical testing (NHST) the researcher te sts the hypothesis of no difference, specifically, whether an observe d difference between two groups or samples is due to chance (Harris, 1997). Researcher s who fail to find significant differences sometimes conclude that the two groups are equivalent. This is not correct, however. A finding of no difference does not show that the two groups or samples are equivalent (Gladstein and Makuch, 1984). Anot her issue is that a statistically significant difference is not necessarily a difference substantial e nough to be of practical importance. However, researchers who do find significant differences often concl ude that the two groups are indeed different, regardless of how trivial thes e differences might be or whether they are of practical importance (Rogers, 1996, Tryon, 2001). Another issue with traditi onal null hypothesis testing is that it can be overly conservative, increasing the chances of maki ng a Type I Error of rejecting the hypothesis of no difference. In the current study th e z-score will be employed as the test of statistical significance. The a ssociated p-value represents the probability of a Type I Error. A Type I Error means that the null hypothe sis will be falsely rejected, that is, the
97 null hypothesis of no difference is rejected although it is in fact true and there really is no difference. For the purpose of the cu rrent study, the p-value represents the probability that the researcher concludes that there is a statistically significant difference between the DUF and ADAM data when in fact there is no difference. The current study uses an -level of .05, which is a standard leve l in criminology and other fields. An -level of .05 means that if th e test would be carried out 100 times, 5 of these tests would suggest significant differences, and as a result rejecting the null hypothesis of no difference when in fact the null hypothesis is true and there is no difference (Moses, 1992). Accordingly, the hypothesis of no differen ce is rejected if either of the twotailed tests is significant, that is, the zvalue with a 95% confidence interval and an level of .05 must be greater th an 1.96 or below -1.96 (Agres ti and Finlay, 2007). Z-values that lie in the middle of the normal distri bution (e.g. in betwee n -1.96 and +1.96) are indicative of the norm, whereas extreme valu es (values in the tails of the normal distribution) represent the une xpected, something out of the ordinary or significantly different (Agresti, 2007). For instance, the p-value for a z-score of 1.97 is .048, which is statistically significant at the .05 level. As the result, the null hypothesis of no difference would be rejected. If the z-scor e is .876, however, the associated p-value would be .38, which is not equal to or lower than .05, and the research er would not reject the null hypothesis. The equatio ns used for the traditiona l hypothesis test are shown below: Calculation of the Tradit ional Null Hypothesis Test The traditional z test is computed as:
98 z = (p1 p2)/SE p = corresponding to the calculated z value The standard error is calculated as: SE = The 95% Confidence Interval is: LCL (p1-p2) z /2(SE) LCL (p1-p2) + z /2(SE) The current study uses the tw o-tailed p-value with an level of 5% because that is the level used regularly in criminology. Accordingly, the p-va lue is significant if it is smaller than .025 in the table of the standard normal distribution. In this example, the zscore corresponds to a p-value of .044 in th e table of the standa rd normal distribution. This p-value of .044 constitutes the one-taile d p-value for the null hypothesis. Thus, the two-tailed p-value would be 2*.044 = .088, which is >.05 the predetermined level of 5%, indicating that the value is not signifi cant. The null hypothesis of no difference cannot be rejected. This result implies that the hypothesis of no difference between DUF and ADAM cannot be rejected. As stated previously, however, this cannot be interpreted as equivalence. To determine whether these two proportions can be said to be equivalent, an equivalence test is conducted simultaneou sly to the traditional null hypothesis test. Equivalence Testing In contrast to traditional null hypothesis te sting, equivalence testing reverses the null hypotheses. Thus, the null hypothesis is the hypothesis of a difference, specifically, the null hypothesis states the difference among group means is greater than some minimal difference representing pract ical equivalence (Allen and Seaman, 2006,
99 p.1). For the current study, the null hypothesi s will be examined with regard to group proportions, that is, proportions of booked arrestees who used a specific drug. Similar to traditional significance testing, the goal is to re ject the null hypothesis. This reversal of the null hypothesis allo ws the researcher to draw a conclusion of whether there is equivalence between the two groups or propor tions (Hauck & Anderson, 1986; Rogers, et al. 1993). Thus, equivalence testing can be seen as a complementary test that allows the researcher to better assess the magnitude of differences (Allen and Seaman, 2006; Hauck and Anderson, 1986; Rogers, 1996). The equations employed for the equivalence test are as follows: Equivalence Test Calculation of the Equivalence Test The equivalence z test is computed as: z1 = (p1 p2) 1/SE p1 = corresponding to th e calculated z-value z2 = (p1 p2) 2/SE p2 = corresponding to th e calculated z-value The equivalence margin is computed as % of the baseline value (explained below) or stated as an equation: (equ als the computation of the percentage difference described above) 1 = 20% p1 2 = +20% p1 The equivalence confidence interval is calculated as: LCI: (p1 p2) z(SE) UCI: (p1 p2) + z(SE) The standard error is calculated as:
100 SE = For calculation purposes, only the lower z-value with the higher corresponding pvalue needs to be computed (indicating that equivalence does not exist) because it has a greater likelihood of being non-significant th an the higher z-valu e with the lower corresponding p-value. For dem onstration purposes, this calculation is shown at the end of this chapter using Dallas as an example. Possible Outcomes and Interpretation Four main outcomes are observed: (1) The drug use information in DUF and ADAM is substantially different (D), (2) Th e drug use information in DUF and ADAM is equivalent (Eq), (3) The drug use information in DUF and ADAM is different and equivalent (D&Eq), and (4) The results are statistically indeterminate (ND&NEq) (Allen and Seaman, 2006; Rogers, et al., 1993; Tryon and Lewis, 2009). First, the results are substantially different if the traditional test is statistically significant at the .05 level and if the equivalence test is not statistically significant at the .05 level. Second, the results indicate equivalence if the traditional test is not statistically significant at the .05 level, and the equivalence test is statistically si gnificant at the .05 level. Third, the analysis show s that the two proportions are st atistically different but also equivalent. According to some researchers, in this case the differen ce can be said to be trivial (Allen and Seaman, 2006; Rogers, et al., 1993). The fourth possible outcome is referred to as statistically indeterminate be cause there is no clear evidence for either statistical difference or e quivalence (Allen and Seaman, 2006; Rogers, et al., 1993; Tryon and Lewis, 2009). The results will be said to be statistically indeterm inate if the results indicate that the drug use information is neit her statistically differe nt nor equivalent. For
101 a better understanding, Figure 1 demonstrates how the confiden ce interval can be used to determine whether there is equivalen ce between the DUF and ADAM samples. Figure 1: Confidence-Interval Approach fo r Equivalence Testing (Allen and Seaman, 2006; Rogers, et al., 1993) Equivalent (E) Different (D) Different but Equivalent (DE) Not Different and Not Equivalent (NDNE) 0 + Equivalent As evident in Figure 1, statistical equivalen ce (E) exists if the confidence interval includes 0 and lies within the equivalence margin (to + ). For instance, the equivalence margin is -24 to +24 and the twosided confidence interval is -10 and +10. A statistical difference (D) is observed if the confidence interval falls outside the equivalence margin and does not include 0. Th is would be the case if the equivalence margin is -24 to +24 and the two-sided confidence interval is + 20 and + 46. The results are said to be different but equivalent (DE) if the confidence interval lies within the equivalence margin but does not include 0. Fi nally, the results are indeterminate (not different and not equivalent) if the confidence interval is not contained within the
102 equivalence margin but does include 0. As shown in Figure 1, this could be the case if the two-sided confidence interval is partially insi de and partially outside of the equivalence margin. For instance, the equivalence margin is -24 to +24 and the two-sided confidence interval is 20 and + 46. Defining the Equivalence Margin One of the major issues is how to defi ne an appropriate equivalence margin (European Medicines Agency, 2000; Wiens, 2001). Equivalence testing is mostly used in the field of biostatistics and in clinical trials to compare whether two treatments or drugs are equivalent with regard to their eff ectiveness (Hauck and Anderson, 1986; Wiens, 2002). In the field of bioequivalence studies, th e equivalence limit is typically defined as 20% (European Medicines Agency, 2000). Addi tionally, the European Medicines Agency (2000) states that a 90% conf idence interval is an acceptabl e equivalence interval to evaluate whether the average va lues of the outcome data are sufficiently close (European Medicines Agency, 2000). Rogers, et al. (1993) suggests that an equivalence interval of 20% is appropriate (p. 557). Equivalence testi ng has only rarely been used in the social sciences field. For instance, Epstein, et al. (2001) compared the equivalence of internet versus paper-and-pencil assessments. They al so used a 90% confidence interval and a 20% equivalence margin. To further assess th is issue of an appr opriate equivalence interval, drug use research was examined. To date, there are no set standards in the field of drug use research for the question of what constitutes a substantial diffe rence with regard to changes in the drug use prevalence over time or across different popul ation groups. The term substantial is, however, used regularly by researchers to desc ribe differences in the drug use prevalence
103 and patterns. What is substantial differs depending on the drug and the population group being examined. The current study looks at drug use among arrestees, and therefore the drug use literature examining th is population group wa s investigated to determine what is typically considered a substantial change for the drugs included in the current analysis: marijuana, powder cocaine, crack cocaine, hero in, amphetamines, barbiturates, and PCP. Threshold Levels for the Major Drugs To reiterate, there exists no clear standard for the question of what constitutes not just a statistically significant ch ange for but also a change that is of practical importance. The current study examines data that include s changes during a time period of five years (1997/1998 compared to 2000/2001). Thus, studies that include trend analysis over several years will be examined as well as re ports that assess changes within the last 12 months. Marijuana. For marijuana, NIJ (1997) reporte d that for the time period 1996 to 1997 Marijuana positive rates for juvenile males showed moderate increases (2-8 percentage points) in the majority of site s (p. 8). Additionally, the DUF report by NIJ (1995) states that there were two sites with sizable increases between 1994 and 1995: New Orleans (up 9 points to 16%) and Washington, D.C. (up 8 points to 18%) (p. 10). This statement implies that an increase of 7% or less is not substa ntial for a one-year time period. Golub and Johnson (2001) suggest that a percentage change of less than 5% within one year is probably due to rando m variation. Furthermore, a study by the Australian government examining the im pact of legalizing marijuana use on the prevalence rate implies that a percentage of 4% is not substantial (National Drug and Alcohol Research Centre, 1996).
104 Overall, the examined studies suggest th at an increase or decrease in marijuana use of less than 8% is not considered substantial. An 8% increase for marijuana use at a marijuana use rate of 35% corresponds to a percentage difference of 23% (calculated as described above). The annual report published by NIJ ( 1999) defines a substantial increase to be at least 10% (marijuana positiv e rates) within two years. The median rate of marijuana positive drug tests in 1999 was 39%. Thus, hypothetically, if there is an increase from 29% to 39% (10% increase), that co rresponds to a percen tage difference of 34%. Thus, with regard to marijuana there s eems to be consistency in the finding that a percentage difference of more than 20% is not necessarily considered substantial. Thus, the current study will use a margin (percentage change) of 20% of the baseline value. Cocaine. The threshold value for cocaine seems to be similar to that of marijuana. This is not nece ssarily surprising because coca ine is a popular drug that is used at high rates among arrest ees. NIJ (1994) reported that there were a number of substantial increases in cocaine use. In thei r study, substantial was defined as percentage changes of more than 5% within one year Golub and Johnson (1997) supported a larger threshold level of 10% within one year by argui ng that: A substantial decline of at least 10 percent in the overall rate of detected cocaine/crack use was observed in Cleveland, Dallas, Detroit, Houston, Los Angeles, New Orleans, Philadelphia, San Diego, San Jose, and Washington, D.C (p. 11). For a thre e-year period, however, Golub and Johnson (1997) state that a change of 9% is not substa ntial. Similar to marijuana, a threshold level of 8% would be a middle ground. Again, an 8% in crease or decrease at a median use rate of 30% constitutes a percentage difference of 27%. According to these findings, an equivalence margin of 20% does not seem excessively liberal.
105 Opiates. The annual DUF report from 1995 only highlighted changes in cities with 10% or more (NIJ, 1995), s uggesting that percentage cha nges of less than 10% were not substantial enough to draw special atte ntion. Furthermore, the ADAM report from 1998 noted: The most substantial declines fo r females were recorded in Washington, D.C. and Cleveland (24 percentage points in each) followed by Detroit (17.8), San Jose (17.6), Dallas (17.3) and San Diego (16.6). While it is not possible to know the standard error of these figures, va riations of this size suggest substantial changes (p. 10). Again, a threshold level of 10% for a five-year time pe riod appears to be r easonable. The annual DUF report from 1999, however, defines substan tial as an increase or decrease of 5%. A 5% increase at a median use rate of about 10% would be a percentage difference of 50%. Thus, a percentage difference of 50% would be considered substantial. Additionally, the authors state that there was no substantia l change between 1998 and 1999. The median drug use positive rate in 1999 was 8%, up 1% from 7% in 1998. This 1% increase results in a percentage difference of 12%. Similar to marijuana and cocaine, an equivalence margin of 20% seems to be reasonable for opiates. Barbiturates, Amphetamines, and PCP. These three types of drugs are used much more rarely than marijuana, cocaine, or opiates. For instance, amphetamines were used by less than 3% of arre stees across sites. Similarly, barbiturates and PCP were typically used by only 1% of the arrestees. As a result, very small increases or decreases will result in a large percentage difference, and as a result smaller changes are considered substantial. For instance, an increase by 0.6% from 0.2% to 0.8% constitutes a percentage change of 65%. NIJ (1999) implies that for methamphetamine, a change of 3% can be considered
106 substantial. NIJ (1995) states that a 5% change from 3% to 8% represents a sharp increase for methamphetamine. Methamphetami ne is, however, used at higher rates than other amphetamines, barbiturates, or PCP. Thus, the threshold for these three drugs would be lower. A percentage difference of 20% (as proposed for marijuana, cocaine, and opiates) would translate into a change of a bout 0.2% at a use rate of about 1.2%, which appears to be average for the current data. In order to get a better understanding of what a 20% equivalence margin means in absolute numbers, a series of examples will be considered. Sensitivity Tests Urine Analysis Test of Marijuana, Dallas The example used for this demonstration is drug positive tests of marijuana use in Dallas, Texas. The baseline proportion for the ur ine analysis test for marijuana use in the ADAM data is .327 or 32.7%, with a sample size of 802 arrestees. In actual numbers this means that 262 arrestees tested positive for marijuana. The following three equivalence margins will be considered: 5%, 10%, and 20%. First, an equivalence margin of 5% woul d result in an increase or decrease of 1.6% (5%*.327), which is 4 arrestees (1.6%*262)/100. According to the drug use and epidemiological literature, a change of 1.6%is not a substantial change for marijuana use. Second, an equivalence margin of 10% would result in an increase or decrease of 3.3% (10%*.327), which equals 9 arrestees (3.3% *262)/100. A change by 3.3% for marijuana is also not considered substa ntial. Third, an equivalence margin of 20% corresponds to a change of about 6.5%, or 17 arrestees. Accordi ng to the literature, a change of about 7% to 8% is considered substantial. Thus, the 20% equivalence ma rgin appears to be a rather
107 conservative measure. Urine Analysis Test of Opiates, Dallas Next, these sensitivity tests were repeated for opiates, which have a lower use rate than marijuana and cocaine. Specifically, only 4.2% (34) of the arrestees in Dallas tested positive for opiates in 2000/2001 (ADAM). A 5% equivalence margin is equal to a change of 0.2% or 1 arrestees (exact: .07) A 10% equivalence margin represents a change of .04 or 1 arrestee (exact: 1.4 arrest ees). A 20% margin shows a change of 0.8% or 3 arrestees (exact: 2.7 arrestees). Simila r to marijuana, a 5% and 10% equivalence margin do not correspond to a change in drug use prevalence that is considered substantial and important according to the literature. Self-Reported Drug Use of PCP within 72 Hours, Dallas PCP is used very rarely. In the current data, only 0.6% of arre stees (or 5 arrestees) reported having used this drug within the past 72 hours. In that case, a 5% equivalence margin means that there would be a ch ange of .03% or .0015 arrestees. A 10% equivalence margin relates to a change of .06% or 0.3 arrestees. Accordingly, a 20% equivalence margin equals a change of .12% or 0.6 arrestees. Again, the 5% and 10% equivalence margin do not constitute a substa ntial difference. Thus, the 20% margin will be used. Power Analysis Both, the traditional test and equivalence test are algebraically similar to the independent samples t-test, which tests wh ether the means or proportions of two groups are statistically different from each other. As a result, the statistical power is also similar (Tryon, 2001). According to Cohens (1992) Power Primer the recommended sample
108 size for an independent samples t-test in or der to find a small effect for differences in proportions is N = 392 for an -level of .05 and N = 584 for an -level of .01. The sample sizes in the current study are larger than these recommended sample sizes, including Miami which only has data for one ADAM year. Thus, there is sufficient power to conduct the traditi onal null hypothesis and the equivalence test. After having discussed why equivalence tes ting is an appropriate approach for the current study, that there is enough power to conduct the analysis, and how the equivalence margin will be defined, the following part shows the calculation of the equivalence test using the data from Dallas. Analysis of Dallas, Texas Data For an easy overview, a summary of the key analysis steps is shown below: Step 1: Two simultaneous one-sided te sts (using the equations provided above) a) Traditional Hypothesis Test The goal is to reject the null hypothe sis of no difference or stated differently: reject the null hypothesi s asserting that the difference between two means (or proportions) is less than or equal to the smaller delta (1) (Rogers, et al., 1993, p. 554). b) Equivalence Test The goal is to reject the null hypothe sis of a difference or stated differently: reject the null hypothesis asserting that the difference is greater than or equal to the larger delta ( 2) (Rogers, et al., 1993, p. 554). Step 2: Evaluation of the Outcome Four Possibilities:
109 (a) DUF and ADAM are equivalent (E) (b) DUF and ADAM are different (D) (c) DUF and ADAM are different but equivalent (ED) (d) DUF and ADAM are not different and not equiva lent (NDNE) Analysis Example: Self-reported marijuana use in Dallas, TX (Ever Used Marijuana) Step1: Two simultaneous one-sided test s (using the equations provided above) a) Traditional Hypothesis Test p1 = 68.5% or .685, n = 802 (ADAM) p2 = 78.3% or .783, n = 1547 (DUF) Standard Error: SE = [(.685(1-.685))/802 + (.783(1-.783))/1547]1/2 SE = .019 Traditional 95% confidence interval LCL: (.685 .783) 1.96(.019) = -.136 UCL: (.685 .783) + 1.96(.019) = -.060 Z-test z = (.685 .783)/.019 z = -5.035 p = .000 (one-tailed) The z-score when using a 95% confidence interval is below 1.96 and thus the pvalue is below the .025 level, implying a sta tistically significant difference for this specific drug variable. Thus, the null hypothesis of no difference can be rejected. As described previously, this is not sufficien t to conclude that the two proportions are
110 substantially different. In order to draw that conclusion it has to be shown that the two proportions are statistically di fferent and not statistically equivalent. Thus, next the equivalence test is computed. c) Equivalence Test Standard Error: SE = .019 (as calculated previously) Equivalence margin 20%: (equals the co mputation of the percentage difference described earlier) 1 = 20% .685 = -.1370 2 = + 20% *.685 = .1370 = .1370 The equivalence interval would be -.1370 to +.1370 or 13.7%. Next, the equivalence confidence interval will be calculated to see whether it is contained within this equivalence margin. Example 90% Equivalence Confidence Interval: LCL: (.685 .783) (1.645)(.019) = -.130 UCL: (.685 .783) (1.645)(.019) = -.066 The confidence interval does not include 0 but it falls inside the equivalence margin of .1370. Figure 2 shows the equivalence marg in and the confidence interval.
111 Figure 2: Example Equivalence Test -.130 -.066 -.1370 Equivalent +.1370 To further assess whether substantial differences exists, the equivalence z will be computed. If both tests have a p-value with a level of .05 or less the null hypothesis of a difference can be rejected and the two proportions can be said to be equivalent. As described above, only the lower z-score with the higher p-value needs to be computed because it is more likely to be not statistically significant than the higher z-score with the lower p-value. Example Equivalence z: z1 = ((.685 .783) (-.685))/.019 = 2.004 p = .000 This finding indicates that the z-test with the higher p-value is significant for the one-tailed test at the .05 level because the z-score when using the 90% confidence interval is greater than 1.645. Accordingly, the null hypothesis of a difference can be rejected and equivalence can be conclude d. This example implies that the DUF and ADAM data are statistically different but stat istically equivalent. The following section will now present the results of the current study. Decision Rules
112 At this point it is important to lay out the criteria for the interpretation of the results and the hypotheses. As described at th e beginning of this Chapter, there are four possible outcomes: (1) DUF and ADAM can be said to be equivalent (Eq) (2) DUF and ADAM are different (D) (3) DUF and ADAM are different and equivalent (D&Eq) (4) DUF and ADAM are not different a nd not equivalent (indeterminate) (ND&NEq) To reiterate, the main question is whether the drug use estimates of DUF and ADAM are substantially different. Two of the four possibl e outcomes are straightforward with regard to their interpretation. If th e variable is statistically di fferent and not statistically equivalent (D), then the variable is classified as substantially differe nt. If the variable is statistically equivalent and not statistically (Eq) different, it can be classified as not being substantially different. The other two outco mes are somewhat more ambiguous. If the variable is statistically different and statisti cally equivalent (D&Eq), it can be classified as not substantially different because even though the traditional null hypothesis test showed a statistically signifi cant difference this difference was marginal in the sense that the confidence interval of the proportion sti ll falls within the equivalence margin. Finally, if the variable is not statistically different and not stat istically equivalent (ND&NEq), it cannot be classified as either substantially different or not substantially different. These variables are st atistically indeterminate. It is to be expected that each site wi ll have several of these outcomes and thus there is the issue of how these multiple findi ngs are to be interpreted. For example, for
113 Dallas 27% (3) of the variables fell into the category equivalent 27% (3) were statistically different but statistically equivalent (D&Eq), and 45% (5) of the variables were not statistically different and not statistically equivalent (ND&NEq) and cannot be judged because they were statistically indeterm inate. The following decision criteria are proposed to judge these results. First, for the purpose of the current study the findings of Equivalent (Eq) and Different but Equivalent (D&Eq) will be combined and in terpreted as not substantially different. Researchers su ggest that a finding of Different but Equivalent (D&Eq) means that although there is a significant difference using the trad itional null hypothesis test these differences might be trivial and not of practical importance. Specifically, Rogers, et al. (1993) states that the diffe rence was larger than the standard null value (usually zero) but smaller than a difference that would make the groups nonequivalent (p.561). This interpretation is also supported by Allen a nd Seaman (2006) who stat e that there is a difference, but it is trivial (p. 78). They sugge st that this finding might occur because the study was overpowered. This term refers to th e possibility that with a large sample size it is possible to detect a significant difference wh en in fact the difference is scientifically insignificant (Frank and Althoen, 1994). This type of error (Type I Error) is also one of the main criticisms on traditional null hypothesi s testing, namely that with large sample sizes researchers will often fi nd a statistically significant difference (Batanero and Diaz, 2006). Finding a statistically significant difference does not necessarily mean that the difference is of practical importance (B atanero and Diaz, 2006; Levin, 1998; VachaHaase, 2001). Conducting the trad itional null hypothe sis test and the equivalence test simultaneously provides a better understan ding of the magnitude of the actual
114 differences. In the current study, the percen tage outcomes for certain drugs may be statistically different but at th e same time statistically equivale nt in the sense that the 90% equivalence confidence interval falls with the equivalence margin. Thus, these outcomes will be interpreted as not substa ntially different or similar. Second, sites will be classified as unable to be assessed if more than one third of the variables fall into the category Not Statistically Differe nt and Not Statistically Equivalent. In actual numbers, it means that sites will be said to be unable to be assessed if more than three va riables are statistically indeterminate. Rogers, et al. (1993) suggest that such findings might be due to excessive within group variation and as a result the equivalence analysis is inconclusi ve. This is believed to be a conservative approach. Third, sites will be classified as substantia lly different if 20% or more of the drug use values show substantial di fferences. In actual numbers, that means that sites will be determined to be substantially different if three variables or more drug use values are classified as Different. As described in Chapter 2, dr ug use prevalence and patterns are not constant over time and a certain amount of changes can be contributed to these naturally occurring differences as long as the di fferences reflected in the current study are consistent with changes evident from other drug use studies. For example, urine test results for cocaine in Denver are substantiall y different. The DUF and ADAM data show that cocaine use declined c onsistently from 40.2% in 1997 to 32.8% in 2002 (see Table B.2., Appendix B). This decrease in cocaine use is supported by the data of the major national drug use surveys. Similarly, data from the DAWN study show s that emergency room visits due to
115 heroin use have increased by 33% between 1995 and 2001. This is a substantial increase and it can be expected that such increase wi ll also reflect in the DUF/ADAM data. This study includes three measures of heroin use (MJ, MJ72, and EV ERMJ), two of which refer to recent use (MJ and MJ72). These two variables measuring recent heroin use would likely be affected by the actual incr ease of heroin use as shown by the DAWN data. Data from the Treatment Episode Data Set (TEDS) also shows a decrease in emergency room admissions for cocaine use for Texas. The NSDUH also demonstrates a decline in cocaine use in Te xas, albeit a smaller dec line (SAMHSA, 2001, Appendix A). Additionally, the MTF suggests that cocaine use among school children declined from about 5.9% in 1997 to 4.8% in 2001. Considering the consistency of these findings with regard to a decline in cocaine use, the subs tantial difference found fo r the urine test result for cocaine in Denver might be attributable to an actual change in drug using behaviors rather than the change in the sampling me thod. Thus, it appears reasonable to expect some variables to be substantially different. Limiting the number of variables than can be substantially to two could be s een as a conservative approach considering the fact that the data covers a five year time period and that there are two m easures of recent drug use for each drug, that is, the urine analysis result fo r marijuana, cocaine, and opiates, and drug use within the past 72 hours for marijuana, cocaine, and heroin. In summary, the following decision criteria are proposed: 1) The findings of Equivalent (Eq) and Different but Equivalent (D&Eq) will be interpreted as not substantially different or similar. 2) Sites will be classified as unable to be assessed if more than one third of the
116 drug use values fall into the category Not Statistically Different and Not Statistically Equivalent (ND&NEq). 3) Third, sites will be classified as substantia lly different if 20% or more of the drug use values show substa ntial differences (D).
117 CHAPTER FIVE RESULTS This Chapte r presents the results of the equivalence analysis. This section will begin by examining the demographic profile for DUF (1997/1998) and ADAM (2000/2001) to get a general overview of the similarities and differences in the two datasets. After the assessment of the demographic profile, the current analysis will take a look at the drug use frequencies fo r DUF (1997/1998) and ADAM (2000/2001). The purpose of examining the drug use frequencies is twofold: (1) to provide an overview of the prevalence of each drug among arrestees a nd (2) to decide whic h drug use variables will be excluded from the equivalence analys is. Finally, the results of the equivalence analysis will be presented to assess whet her the drug use estima tes of DUF and ADAM are substantially different. The results of the equivalence analysis will be presented in two parts: (1) overall results fo r the equivalence tests, and (2) site specific results. The decision criteria described in Chapter Four will be used to interpret the findings. Descriptive Statistics Demographic Profile Tables A.1. to A.9. in Appendix A display the demographic profile of the sample for the nine sites included in the current study. The data shows that the demographic profile of DUF (1997/1998) and ADAM (2000/ 2001) has similarities and differences. The results are presented by demographic category. Race. The racial distribution is describe d for each racial category, including
118 Black, White, Hispanic, and Other. The percen tage of Black arrestees appeared to be similar for Indianapolis, New Orleans, Phoenix, San Antonio, and San Jose. These sites had a difference of less than 5% between DUF and ADAM. Specifically, in Indianapolis, 57.2% of arrestees were Black in the DUF sample, and 56.5% were Black in the ADAM sample. In New Orleans, 87% of arrestee s were Black in the DUF sample, and 86.6% were Black in the ADAM sample. In Phoenix, 13.7% of arrestees were Black in DUF and 11.9% in ADAM. In San Antonio, 11.3% of arrest ees were Black in the DUF sample, and 12.7% were Black in the ADAM sample. Finall y, in San Jose, 11.1% were Black in the DUF sample, and 11.9% were Black in the ADAM sample. Four sites had a difference of 6% or more. These sites are Miami, Portland, Dallas, and Denver. The largest difference was apparent in Dallas, where 58.4% of arrestees in the DUF sample were Black, but only 49.1% of arrestees in the ADAM sample were Black. Additionally, the differe nce in Miami was 6.7%, in Portland 5.4%, and in Denver 8.2%. There was no consistent pattern with regard to the direction of the difference; that is, in Portland, Dallas, and Denver, the DUF sample had a greater number of Black arrestees as compared to the ADAM sample, whereas in Miami the DUF sample had fewer Black arrestees than the ADAM sample. The findings for White offenders demons trate that Portland, Indianapolis, New Orleans, Denver, San Antonio, and San Jose have differences of less than 5% between DUF and ADAM. In Portland, the DUF sample included 61.7% White arrestees; the ADAM sample included 64.6% White arrestees. Indianapolis had 37.9% White arrestees in the DUF sample, and the ADAM sample had 42.4%. The DUF sample in New Orleans consisted of 11% White arrestees, whereas the ADAM sample consisted of 12.9%. The
119 Denver site had 28.6% White arrestees in the DUF sample and 27.6% in the ADAM sample. San Antonio included 32.9% White ar restees in the DUF sample and 35.3% in the ADAM sample. Finally, the DUF sample in San Jose consisted of 32.1% White arrestees and the ADAM sample of 30.4%. Phoenix, Miami, and Dallas showed differe nces that were larger than 5%. The largest discrepancies were found in Miami, where about 15.7% of arrestees were White in the DUF sample, but 43.2% of arrestees were white in the ADAM sample. The remaining two sites demonstrated a differe nce of 6.6% (Phoenix) and 6.5% (Dallas). Again, no consistent pattern for the direction of the discrepancies emerges. Phoenix and Miami had fewer White arrestees in the DUF sample and more in the ADAM sample. Contrary, in Dallas the DUF sample included a greater number of White arrestees than the ADAM sample. The assessment of differences for Hispanic offenders is more complicated because the percentage of Hispanics in the sample varies greatly. Three sites had less than 7% Hispanic offenders in the sample overall. These sites were Portland, Indianapolis, and New Orleans. With regard to the differences between DUF and ADAM, the data shows that in Portland the DUF samp le had 5.6% Hispanic arrestees, whereas the ADAM sample had 7.3% Hispanic offenders. Indianapolis had 4.1% Hispanic arrestees in the DUF sample but only 0.8% in the ADAM sample. In New Orleans, the DUF sample consisted of 0.8% Hispanic offe nders and the ADAM sample of 0.2%. The vast majority of sites show substan tial differences. The greatest differences were found in Miami, where the DUF sample included 37.7% Hispanic arrestees, but the ADAM sample only had 3.9% Hispanic offe nders. Also, in Denver 33.8% of the DUF
120 sample were Hispanic arrestees, but 41.5% of the ADAM sample were Hispanic offenders. In Dallas, 9.8% of arrestees were Hispanic in the DUF sample and 14.8% in the ADAM sample. In Phoenix, 32.3% of arrestees in the DUF sample were Hispanic, and 25.6% of arrestees in the ADAM sample were Hispanic. In San Antonio and San Jose, the percentage of Hispanics in the sample appeared to be similar, with differences of less than 3%. Of the three racial categorie s, Hispanic arrestees showed the greatest discrepancies between DUF and ADAM. Th e important question is whether these differences in the racial dist ribution also lead to differen ces in the drug use estimates between DUF and ADAM. Employment. For the percentage of arrestees employed full time, only one site had a difference of more than 5%. That si te was Miami, with a difference of 6%. Conversely, the category part-time employment was substantially different for all sites with the exception of Phoenix, where the difference was only 2%. The remaining sites demonstrated differences of more than 6%. Similarly, the percenta ges of arrestees who stated they were unemployed were also subs tantially different at all sites. The DUF sample had much fewer arrestees who were unemployed at the time as compared to the ADAM sample. Specifically, at most sites mo re than twice as many arrestees in the ADAM sample fell into the unemployed category as compared to the DUF sample. Education. The frequency distributions also su ggest substantial differences with regard to the high school graduation rate of arrestees between the DUF and ADAM samples. The greatest discrepancies were f ound in San Jose and Miami. In San Jose, the difference was 27.8% because only 49.8% of arrestees in the DUF sample graduated from high school but 77.6% of arrestees in the ADAM sample graduated from high
121 school. Similarly, the difference in Miami was 26.8%; 40.3% of arrestees in the DUF sample graduated from high school as compared to 67.1% in the ADAM sample. The remaining sites had differences of 10% or mo re. There is a consistent pattern for this categorythe ADAM sample has a higher percen tage of high school graduates for all sites. Thus, there appears to be a systematic bias in the data for high school graduation rates. Charge distribution. The charge distribution for DUF and ADAM were expected to be different because of the differences in the arrestee selection procedure. As described in Chapter Four, DUF used a priority charge system, whereas ADAM used the UCR in their determination of the sample. Despite these differences in the arrestee selection procedure, four sites appeared to have a similar charge distribution. Specifically, in Denver, Phoenix, Indianapolis and Portland, the charge distribution for DUF arrestees was within 5% for all th ree categories (viole nt, property, and drug offenses) for that of ADAM. The other five sites demonstrated substantial differences, with discrepancies of 10% and more for each charge category (violent, property, and drug offenses). Again, these differences were to be expected because DUF and ADAM used different methods to determine how many arre stees should be included for each charge category. Age. Contrary to the charge distribution, th e average age of the arrestees in the DUF and ADAM samples was almost exactly the same for all sites. Specifically, for eight sites the average age was the same for DUF and ADAM. The sole exception is San Jose, where the average age of the arrestees in the ADAM sample was 32 as compared to 31 for the DUF sample.
122 In sum, from the frequency distributi ons there appear to be a number of similarities between DUF and ADAM but also differences for the demographic profile of the two samples. This is not an une xpected finding because ADAM employed a probability sample and a weighting procedure to ensure that the arrestee sample was representative of arrestees booked in the catchment area, whereas DUF did not use these procedures. The main question is whether these differences influence the drug use estimates produced by DUF and ADAM in a way that makes them substantially different. Drug Use Frequencies Table 5.1. shows the prevalence rates for drug use by presenting the lowest, highest, and average prevalence rates for all sites combin ed. This data was compiled from Tables B.1. B.9. in Appendix B, where the prevalence ra tes of drug use are displayed for each site for each year for DUF (1997 and 1998) and ADAM (2000 and 2001). The drug use data for both DUF and ADAM appears to be similar with regard to the prevalence of the drugs examined. The most-often used drug is marijuana (MJ). The high prevalence of marijuana shows in both the urine test results and th e self-reported data for the DUF and ADAM data. Second, cocaine (COC) has the next hi ghest prevalence rates for the urine test results and the self-report data for DUF a nd ADAM. The self-report data further shows that the average prevalence ra te for crack cocaine (CRK) us ed within the past 72 hours was higher than for powder cocaine. Contrar y, for lifetime use (eve r used drug) powder cocaine had higher average prevalence rates th an crack cocaine. Again, these patterns are consistent for both DUF and ADAM. Third, in both datasets, opiates (OP), including heroin (HER), had much lower prevalence ra tes than marijuana and cocaine, but were
123 more popular than PCP (PCP), amphetamin es (AMPH), and barbiturates (BARB). Fourth, PCP, amphetamines, and barbiturates we re used very rarely by arrestees in the DUF and ADAM samples. These overall findings will be examined in more detail now. First, the prevalence rates shown in Ta ble 5.1. indicate that in the ADAM sample between 60.2% and 84.1% of arrestees stated that they have used marijuana at some point in their life (EVERMJ). On average, 74.2% of arrestees have previously used marijuana (EVERMJ). Similarly, the DUF data shows that between 63.3% and 87.4% of arrestees have used marijuana (EVERMJ). On average, 76.0% of arrestees admitted to having used marijuana (EVERMJ). With regard to recent use, the ADAM data implies that between 22.1% and 35.0% of arrestees reported having used marijuana within the past 72 hours (MJ72), with the average being 27.0% (MJ72). The DUF data suggests that between 18.9% and 34% have used marijuana in the past 72 hours (MJ72) with the average be ing 26.9%. Additionally, the ADAM data demonstrates that on averag e 38.4% of arrestees tested positive for marijuana (MJ) use. Similarly, in the DUF sa mple, 36.9% of arrestees tested positive for marijuana (MJ). Second, cocaine is also used regularly by arrestees. Specifically, the ADAM data demonstrates that between 3.0% and 9.7% of arrestees used powder cocaine within the past 72 hours (COC72) and between 3.3% and 11 .2% used crack cocain e within the past 72 hours (CRK72). On average, 5.8% of arrestee s had used powder cocaine in the past 72 hours (COC72) and 9.8% had used crack cocaine in the past 72 hours (CRK72). In comparison, the DUF data shows that between 3.2% and 11.3% of arrestees used powder cocaine within the past 72 hours (COC72) and between 2.8% and 17.2% used crack
124 cocaine within the past 72 hours (CRK72). For the DUF data, the average use rates within the past 72 hours were 7.1% for pow der cocaine (COC72) and 11.8%% for crack cocaine (CRK72). The ADAM data also shows that betw een 27.9% and 50.8% of arrestees have used powder cocaine (EVERCOC) and betw een 14.8% and 41.1% have used crack cocaine (EVERCRK) at some point in the life For the DUF data, the lifetime prevalence rates range from 32.3% to 54.1% for powde r cocaine (EVERCOC) and 12.2% and 44.5% for crack cocaine (EVERCRK). Additionally, between 12.0% and 44.9% of arrestees in the ADAM tested positive for cocaine (COC), with the average being 28.9%. For the DUF data, the percentage of arrestees who tested positive for cocaine (COC) ranged from 11.8% to 53.9%, with an average of 33.9%. To reiterate, the urine an alysis cannot distinguish between crack and powder. Thus, only overall cocaine use is reported. Third, opiate use (including heroin use) is rarer among arrestees as compared to marijuana and cocaine. On average, 7.1% of a rrestees tested positive for opiates (OP) in the ADAM data and 6.8% in the DUF data. The urine positive tests ranged from 3.5% to 15.1% for ADAM and 2.2% to 14.8% for DUF. With regard to lifetime heroin use (EVERHER), between 6.6% and 24.2% of arrestees in the ADAM data and between 7.6% and 31% of arrestees in the DUF data reported having used heroin. On average, 14.2% of ADAM arrestees 15.8% of DUF arrestees had used heroin at some point in their life (EVERHER). The data for recent heroin use (HER72) de monstrates that, on average, 4.4% of arrestees in the ADAM data and 4.5% of arrest ees in DUF used heroin within the past 72
125 hours. Heroin use within the past 72 hours (HER72) ranged from 1.3% to 11.2% for ADAM and 0.3% to 11.0% for DUF. The descriptive analyses also show that the variables PCP72, AMPH72, and BARB72 have very low prevalence rates. Spec ifically, the prevalence rates for recent PCP use (PCP72) varied between 0.0% and 1.6% for ADAM and 0.1% to 0.8% for DUF. These percentages are equivalent to a range of 0 to 20 arrestees. On average, PCP72 was used by .04% of arrestees in the ADAM data and 0.3% of arrestees in the DUF data. The prevalence rate for recent use of amphetamines (AMPH72) ranged from 0.2% to 1.4% for ADAM and 0.0% to 3.4% for DUF. On average, 0.7% of arrestees used amphetamines within the past 72 hours (AMP H72) in the ADAM data and 1.6% in the DUF data. Similarly, the prevalence rates for barbitu rates (BARB72) were also very low, varying from 0.0% to 0.5% for ADAM and 0.4% to 2.0% for DUF. On average, 0.2% of arrestees used barbiturates in the ADAM data and 0.8% in the DUF data. These percentages equal a total numb er of arrestees between 0 and 16. Due to these very low sample sizes, the variables PCP72, AMPH72, and BARB72 were excluded from the equivalence analysis.
126 Table 5.1. Drug Use Frequencies for Total Sample Lowest, Highest, and Average Prevalence Rates ADAM 2000/2001 DUF 1997/1998 Variable Lowest HighestAverage N Average LowestHighest Average N Average Urine Test MJ 32.7 48.7 38.4 1,308 27.5 44.3 36.9 1,600 COC 12.0 44.9 28.9 1,308 11.8 53.9 33.9 1,600 OP 3.5 15.1 7.1 1,308 2.2 14.8 6.8 1,600 Self-Report Drug Use Within 72 Hours MJ72 22.1 35.0 27.0 1,308 18.9 34.0 26.9 1,600 COC72 3.0 9.7 5.8 1,308 3.2 11.3 7.1 1,600 CRK72 3.3 12.6 9.8 1,308 2.8 17.3 11.8 1,600 HER72 1.3 11.2 4.4 1,308 0.3 11.0 4.5 1,600 PCP72 0.0 1.6 0.4 1,308 0.1 0.8 0.3 1,600 AMPH72 0.2 1.4 0.7 1,308 0.0 3.4 1.6 1,600 BARB72 0.0 0.5 0.2 1,308 0.4 2.0 0.8 1,600 Ever Used Drug EVERMJ 60.2 84.1 74.2 1,308 63.3 87.4 76.0 1,600 EVERCOC 27.9 50.8 38.9 1,308 32.3 54.1 40.3 1,600 EVERCRK 14.8 41.1 35.2 1,308 12.2 44.5 30.4 1,600 EVERHER 6.6 24.2 14.2 1,308 7.6 31.0 15.8 1,600
127 So far, the descriptive anal ysis has demonstrated that there are differences in the demographic characteristics of the DUF and ADAM samples. The descriptive statistics, however, also show that the overall prevalen ce of the different drugs included in the analysis appears to be similar. These fi ndings support the findings of the NIJ (1993) study which found that even though the demographic characteristics and charge distribution of the DUF and ADAM samples were substantia lly different, the drug use estimates were similar. The following section will present the results for the equivalence analysis to determine whether these findings also hold up when a more rigorous statistical test is applied. Overall Results for the Equivalence Analysis To reiterate, the research question is whether the drug use estimates for selected drugs are substantially diffe rent between DUF and ADAM. Th e main hypothesis is that the drug use estimates contained in DUF a nd ADAM are not substantially different despite differences in the sampling method. In order to assess the hypothesis, the DUF and ADAM data for the nine sites was compared to determine how many drugs and sites, if any, appear to be substantially different and how many, if any, appear to be similar. First, the drug use estimates for the 11 vari ables are similar and different for DUF and ADAM for all sites combined will be di scussed. Tables 5.2. through 5.4. present this information. Table 5.2. shows how many va riables are subs tantially different, how many are not substantially different or similar, and how many are statis tically indeterminate for all sites combined. Each site had a total of 11 drug use estimates one for each drug. The sites with the greatest number of substantially different dr ug use values are Phoenix with
128 36% (4 values) and Portland with 27% (3 va lues). In Denver and New Orleans, 18% (2) of the drug use values were classified as substantially different. For the three sites Indianapolis, Miami, and San Jose, 9% (1) of the drug use values were substantially different. There are two sites where none of the drug use values were substantially different. These sites are Dallas and San Antoni o. Altogether, out of a total of 99 drug use values (11 variables 9 sites), 14 values were found to be substantially different. As explained above, the categories Equivalent and Different but Equivalent will be counted as not substantially different or similar. Thus, these two categories will be evaluated together. The highest number of values that were classified as not substantially different was found in New Orleans with 81% (9 values). Specifically, four drug use values fell into the category Equivalent and five values were clas sified as Different but Equivalent. Several sites had 54% (8) of the drug use values that fell into either the Equivalent or Different but Equivalent category. These sites were Denver, Portland, San Antonio, and San Jose. In Denver, four drug use values were Equivalent and four values were Different but Equivalent. In Portland and San Jose, three drug use estimates showed Equivalence and five va lues were Different but Equivalent. Finally, in San Antonio, six drug use values were classified as Equivalent and two values were classified as Different but Equivalent. In Indianapolis and Miami, 45% (7) of the values showed no substantial differences. For Indianapolis, four values were Equivalent and three values were Different but Equivalent. In Miami, five values proved to be Equivalent and two fell into the category Different but Equivalent. Additionally, for two sites, a to tal of 36% (6) of the drug use values fell into either
129 of these two categories. In Phoenix and Dallas, three drug use values were classified as Equivalent and three values fell into the cate gory Different but Equivalent. No site had less than 36% (6 values) classified as either Equivalent or Different but Equivalent. Finally, there were also a number of sites with values that fell into the category Not Different and Not Equiva lent. These values are statistically indeterminate and cannot be judged as either substantially different or not. The greatest number of statistically indeterminate values had Dallas, where 45% (5) of the values fell into that category. Three sites had 27% (3) values that cannot be judged: Indianapolis, Miami, and San Antonio. In San Jose, 18% (2) values we re indeterminate and in Denver and Phoenix, 9% (1) value was indeterminate. At two side s, none of the values was statistically indeterminate, that is, all drug use values were categorized as either substantially different, equivalent, or different but equi valent. These sites were New Orleans and Portland.
130 Table 5.2. Equivalence Test Outcomes by Site Number of Drug Use Variables Site Different Equivalent Different but Equivalent Not Different and Not Equivalent Percent Total Percent Total Percent Total Percent Total Dallas 0% 0 27% 3 27% 3 45% 5 Denver 18% 2 36% 4 36% 4 9% 1 Indianapolis 9% 1 36% 4 27% 3 27% 3 Miami 9% 1 45% 5 18% 2 27% 3 New Orleans 18% 2 36% 4 45% 5 0% 0 Phoenix 36% 4 27% 3 27% 3 9% 1 Portland 27% 3 27% 3 45% 5 0% 0 San Antonio 0% 0 55% 6 18% 2 27% 3 San Jose 9% 1 27% 3 45% 5 18% 2 Total Values 14 35 32 18 The results for each site as displayed in Table 5.2. gives a general overview of how many drug use values fall into each outcome category. This data does not provide information, however, of whether these outcome categories consist of the same variables across sites. For instance, Dallas, Phoenix, Portland, and San Jose have three drug use values that were categorized as Equivalent. The question is whether those three values are the same for the four sites or whether each site has different patterns with regard to the question how the drug use values are di stributed over the outcome categories. Thus, the next step in the analysis was to examin e the distribution of these drug use variables across the outcome categories to get a better understanding of possible patterns in the data. The results are presented in Tables 5.3. and 5.4.
131Table 5.3. Classification of Drug Use Values Across the Outcome Categories for Each Site Drug Use Variables Site MJ COC OP MJ72 COC72 CRK72 HE R72 EVERMJ EVERCOC EVERCRK EVERHER Dallas D&Eq ND&NEq Eq D&Eq ND&NEq ND&NEq ND&NEq D&Eq Eq Eq ND&NEq Denver Eq D Eq Eq ND&NEq D D&Eq D&Eq D&Eq D&Eq Eq Indianapolis D&Eq Eq D&Eq D ND&NEq ND&NEq ND&NEq Eq Eq Eq D&Eq Miami Eq D D&Eq Eq ND&NEq ND&NEq D&Eq Eq Eq ND&NEq Eq New Orleans D&Eq D&Eq D&Eq Eq D&Eq D&Eq Eq Eq D D Eq Phoenix D&Eq D D D&Eq D ND&NEq D&Eq Eq Eq Eq D Portland Eq D&Eq D Eq D&Eq D&Eq D&Eq D&Eq D Eq D San Antonio Eq Eq ND&NEq Eq ND&N Eq Eq ND&NEq D&Eq Eq D&Eq Eq San Jose D&Eq Eq D D&Eq ND&NEq Eq ND&NEq D&Eq D&Eq D&Eq Eq D = Different; Eq = Equivalent; D&Eq = Different and Equivalent; ND&NEq = Not Different and Not Equivalent Table 5.4 Summary Table for the Distribution of Drug Use Values Across the Outcome Categories Drug use Variables Outcome MJ MJ72 EVERMJ COC COC72 EVERCO C CRK72 EVERCRK OP HER72 EVERHER Total D 0 1 0 3 1 2 1 1 3 0 2 14 Eq 4 5 4 3 0 5 2 4 2 1 5 35 D&Eq 5 3 5 2 2 2 2 3 3 4 1 32 ND&NEq 0 0 0 1 6 0 4 1 1 4 1 18 D = Different; Eq = Equivalent; D&Eq = Different and Equivalent; ND&NEq = Not Different and Not Equivalent
132 Table 5.3. shows the classification for each drug use value and site. The drug use variables are ordered in their original way, that is, the first three variables are the urine test results, followed by the variables for self -reported drug use with in the past 72 hours, and finally, the self-reported lifetime drug use. Table 5.4. shows the same data but summarized for each outcome category and drug use variable. Also, the drug use variables are ordered different ly. The first three drugs are the variables for marijuana, followed by the variables for cocaine and finally opiates. This order makes patterns more obvious. The results show that there are indeed certain patterns with regard to the distribution of drug use values across the outcome categories. First, the marijuana variables fall almost exclusively into the two categories Equival ent and Not Different (Eq) and Different but E quivalent (D&Eq). Only one value (4%) for all marijuana variables shows substantial diffe rences (D). This value is Marijuana Used within the Past 72 Hours in Indianapolis None of the values for ma rijuana were classified as indeterminate. Thus, the drug use values for marijuana can be said to show no substantial differences for any site and overall for all sites combined. Figure 5.1. summarizes the distribution of the marijuana values across the outcome categories. For the purpose of si mplification, the outcome categories were recoded to better reflect their meaning. Th e four original outcome categories were collapsed into three categories. Specifi cally, the category Statistically Different and Not Statistically Equivalent was renamed in to Substantially Different. The outcome categories Statistically Equivalent and Not Statistically Different and Statistically Different but Statistically Equivalent were co llapsed into Not Subs tantially Different.
133 Finally, the category Not Stat istically Different and Not Statistically Equivalent was renamed Indeterminate. Again, Figure 5.1. very clearly demonstrates that all but one variable show no substantial differences. Figure 5.1. Distribution of Marijuana Values Across the Outcome Categories Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t Second, for the drug use variables for cocain e use, the data displayed in Tables 5.3. and 5.4. implies that the results are not as clear cut as for marijuana. The urine test results for cocaine (COC) show that three va lues were substantially different (Denver, Miami, and Phoenix), five values were not substantially different (Indianapolis, New Orleans, Portland, San Antonio, and San Jose), and one value could not be statistically determined (Dallas). For self-reported powder cocaine use within the past 72 hours (COC72), the analysis demonstrates that six values coul d not be statistically determined (Dallas, Denver, Indianapolis, Miami, San Antonio, and San Jose). Of the remaining three values, one value was substantially different (Phoenix) and two were not substantially different
134 (New Orleans and Portland). The pattern for self-reported crack cocai ne use within the past 72 (CRK72) hours demonstrates that one value was substantiall y different (Denver), four values were not substantially different (New Orleans, Portla nd, San Antonio, and San Jose). Finally, four values were classified as indeterminate (Dal las, Indianapolis, Miami, and Phoenix). The outcomes for self-reported powder co caine use over the lifetime (EVERCOC) suggest that two values were substantially different (New Orleans and Portland) and seven values were not substantially differe nt (Dallas, Denver, Indianapolis, Miami, Phoenix, San Antonio, and San Jose). None of the values was indeterminate. Similarly, for self-reported crack cocain e use over the lifetime (EVERCRK), the distribution is as follows: one value was classified as s ubstantially different (New Orleans), seven values did not show substantial differences (D allas, Denver, Indianapolis, Phoenix, San Antonio, and San Jose), and one value was statistically indeterminate (Miami). Figure 5.2. summarizes the results for the drug estimates for cocaine. To reiterate, the following variables were included in this diagram: COC, COC72, CRK72, EVERCOC, and EVERCRK. Of th e 45 values for cocaine, eight values (18%) fell into the category Statistically Different and Not Statistically Equivalent; 25 values (56%) were classified as either Statistically E quivalent and Not Statistically Different or Statistically Different But Statistically Equi valent; and 12 values (27%) were neither statistically different nor stat istically equivalent. Even though the results for cocaine are not as clear as for marijuana, overall the conc lusion would be that the drug use estimates are not substantially different because less th an 20% of the drug estimates demonstrated
135 substantial differences. Figure 5.2. Distribution of Cocaine Values Across the Outcome Categories Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t Third, Tables 5.3. and 5.4. also show the results for the opia te variables. The findings for the urine test results for opiates (O P) show that three values were classified as substantially different (P hoenix, Portland, and San Jose), five values did not show substantial differences (Dallas, Denver, Indianapolis, Phoenix, and Miami), and one value could not be assessed because it was not different and not equivalent (San Antonio). The results for the values for heroin use within the past 72 hours (HER72) demonstrates that none of the values were substantially different, five values were classified as not substantially different (Denver, Miami, New Orleans, Phoenix, and Portland), and four values were statistica lly indeterminate (Dallas, Indianapolis, San Antonio, and San Jose). With regard to lifetime heroin use (EVE RHER), the analysis demonstrates that
136 two values were substantially different (Phoe nix and Portland), six values did not show substantial differences (Denver, Indianapolis, Phoenix, Miami, San Antonio, and San Jose), and one value was statistically indeterminate (Dallas). Figure 5.3. summarizes the distribution of the values for opiates across the outcome categories. Of the 27 values, five values (19%) showed substantial differences, 16 values (59%) were not substantially different; and six values (22 %) could not be statistically determined. Overall, the conclusion is that the drug use estimates for opiates are not substantially different because less than 20% of the values demonstrated substantial differences. After having di scussed the overall results a nd the results for each drug category, the next section analyzes the findings for each site. Figure 5.3. Distribution of Opiate Valu es Across the Outcome Categories Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t
137 Site-Specific Findings Dallas Table 5.5. presents the results of the equivalence analys is for Dallas. The findings demonstrate that none of the drug us e variables fell into the category different that is, none of them were statistically different and not statistically equiva lent (D). Further, three variables (27%) were not statistically diffe rent in the traditional test and statistically equivalent in the equivalence test, whic h means that they were classified as equivalent (E). These three variables were Urine Te st Result for Opiates, Ever Used Powder Cocaine, and Ever Used Crack Cocaine. Th e smaller equivalence z with the higher pvalues for these three variable s is larger than 1.645, indicatin g that they are equivalent between DUF and ADAM. The z-values for the tr aditional test were not larger than 1.96, which means that they were not statistically different. The 90% confidence intervals of these three variables fall within the equivale nce interval and include zero. For instance, the equivalence interval for Ever Used Powder Cocaine is .0648. The 90% confidence interval has a lower limit of -.040 and an upper limit of .028. Three drug use estimates (27%) were f ound to be statistically different and statistically equivalent (DE) and thus classified as different but equivalent These variables were Urine Test Result for Mariju ana, Marijuana Used within the Past 72 Hours, and Ever used Marijuana. The z-va lues for the traditional test were greater than 1.96 and thus displayed a significant difference, but th e smaller z-values for the smaller equivalence test were also greater than 1.645 indicating that the confidence intervals overlap with the equivalence interval. In fact, the smaller z-values for the equivalence test were also greater than 1.96 indicating significance at the .025 level.
138 The remaining five variables were statis tically indeterminate because they were not statistically d ifferent and not statis tically equivalent (NDNE). The traditional z was not larger than 1.96 and the smaller equi valence z-value was not larger than 1.645, indicating that neither test was significant and as a result neit her of them can be rejected. The confidence limits of these variables we re not contained within the equivalence interval. Rather they were partially inside and outside the equivalence margin. The six drug use estimates that were indeterminate were Urine Test for Cocaine, Powder Cocaine Used within the Past 72 Hours, Crack Cocaine Used within the Past 72 Hours, Heroin Used within the Past 72 Hours, and Ever Used Heroin. Figure 5.4. summarizes the results for Dalla s. Overall, six variables (55%) were not substantially different and five variables (45%) could not be classified because they were statistically indeterminate. In accordance with the decisions ru les outlined in the previous chapter, this means that no conclusions can be drawn for Dallas because more than one third, that is, 45%, of the variables could not be statistically assessed.
139 Figure 5.4. Distribution of Variables across the Outcome Categories in Dallas Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t
140Table 5.5.: Equivale nce Test DUF and ADAM in Dallas, TX ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditional 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z P LCL UCL z pa LCL UCL Urine Test Marijuana 0.327 802 0.434 1547 -0.107 0.021 .0654 -5.141 000 -0.148 -0.066 -1.999 .022* -0.141 -0.073 Cocaine 0.271 802 0.305 1547 -0.034 0.020 .0542 -1.737 .041 -0.072 0.004 1.032 .159 -0.066 -0.002 Opiates 0.042 802 0.035 1547 0.007 0.008 .0084 0. 825 .203 -0.010 0.024 1.815 .035* -0.007 0.021 Self-Report Drug Use Within 72 Hours Marijuana 0.227 802 0.326 1547 -0.099 0.019 .0454 -5.212 000 -0.136 -0.062 -2.822 .002* -0.130 -0.068 Cocaine 0.059 802 0.060 1547 -0.001 0.010 .0118 -0.097 .461 -0.021 0.019 1.051 .159 -0.018 0.016 Crack 0.121 802 0.123 1547 -0.002 0.014 .0242 -0.141 .444 -0.030 0.026 1.561 .059 -0.025 0.021 Heroin 0.020 802 0.021 1547 -0.001 0.006 .0040 -0.163 .435 -0.130 0.011 0.488 .316 -0.011 0.009 Ever Used Drug Marijuana 0.685 802 0.783 1547 -0.098 0.019 .1370 -5.035 .000 -0.136 -0.060 2.004 .023* -0.130 -0.066 Cocaine 0.324 802 0.330 1547 -0.006 0.020 0648 -0.294 .384 -0.046 0.034 2.883 .002* -0.040 0.028 Crack 0.264 802 0.270 1547 -0.006 0.019 0528 -0.312 .378 -0.044 0.032 2.434 .008* -0.038 0.026 Heroin 0.094 802 0.101 1547 -0.007 0.013 .0188 -0.545 .293 -0.032 0.018 0.919 .181 -0.028 0.014 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of the two one-sided tests has been reported; bThe equivalence interv al was defined as % of the baseline va lue (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
141 Denver Table 5.6. shows the findings for Denver including the 95% confidence interval for the traditional null hypothesis test and 90% confidence limits for the equivalence test and how they fall with regard to the equiva lence interval. For Denver about 18% (2) of the variables were substantially different. These two variables were Crack Cocaine Used in the Past 72 Hours, and Urine Test Resu lt for Cocaine. For these two variables, the z-test for the traditional test was larger than 1.96 and as a result the p-value was smaller than .025 indicating that the difference was sta tistically significant fo r the traditional test. With regard to the equivalence test, the smaller z-values with the larger p-values were not larger than 1.645 and as a result the e quivalence test was not significant. These equivalence 90% confidence intervals of thes e two variables were not contained in the equivalence intervals, supporting the finding th at they are not equi valent for DUF and ADAM. Four (36%) of the variables displayed equivalence, that is, the smaller equivalence z with the higher p-value was greater than 1.645, i ndicating statistical equivalence, and the traditional z test was not statistically significant. The four variables that were found to be equivalent were Uri ne Test Result for Marijuana, Urine Test Result for Opiates, Marijuana Used in the Pa st 72 Hours, and Ever Used Heroin As evident in Table 5.6, the 90% equivalence c onfidence intervals of these variables are contained within the equivalen ce interval and include zero. Four (36%) variables fell into the category different but equivalent. These four variables were Heroin Used in the Past 72 Hours, Ever Used Marijuana, Ever Used Powder Cocaine, and Ever Used Crack Co caine. The traditiona l z test of these
142 variables implied a statistically significant di fference, but the smaller equivalence z was also greater than 1.645. As described previously, this finding indicates that although there is a statistically significant difference this di fference is not believed to be substantial, and as a result these drug use estimates can be treated as being comparable. The remaining variable Powder Cocaine Used within the Past 72 Hours was classified as not different and not equivalent The status of this dr ug use value with regard to equivalence or a substantial difference c ould not be statistically determined because neither the traditional test nor the equiva lence test was statistically significant. The results for Denver are summarized in Figure 5.5., showing the distribution of the variables across the outcome categories. Overall, two variables were substantially different, eight variables were not substant ially different, and one value could not be determined. The findings show that less than 20% of the drug use values were substantially different, which implies that the drug use estimates from DUF and ADAM are not substantially different for the site of Denver. Figure 5.5. Distribution of Variables across the Outcome Categories in Denver Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t
143Table 5.6.: Equivale nce Test DUF and ADAM in Denver, CO ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditional 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.404 1319 0.413 1908 -0.009 0.018 0808 -0.511 .305 -0.043 0.025 4.081 .000* -0.038 0.020 Cocaine 0.334 1319 0.399 1908 -0.065 0.017 .0668 -3. 789 .000 -0.099 -0.031 0.105 .460 -0.093 -0.037 Opiates 0.046 1319 0.038 1908 0.008 0.007 .0092 1.105 .136 -0.006 0.022 2. 375 .009* -0.004 0.020 Self-Report Drug Use Within 72 Hours Marijuana 0.303 1319 0.291 1908 0.012 0.016 .0606 0.733 .232 -0.020 0. 044 4.433 .000* -0.015 0.039 Cocaine 0.058 1319 0.073 1908 -0.015 0.009 0116 -1.711 .044 -0.032 0.002 -0 .388 .352 -0.029 -0.001 Crack 0.125 1319 0.152 1908 -0.027 0.012 .0250 -2 .201 .014 -0.051 -0.003 -0 .163 .436 -0.047 -0.007 Heroin 0.026 1319 0.003 1908 0.023 0.005 .0052 5.047 .000 0.014 0. 032 6.188 .000* 0.016 0.030 Ever Used Drug Marijuana 0.761 1319 0.855 1908 -0.094 0.014 1522 -6.600 .000 -0.122 -0. 066 4.086 .000* -0.117 -0.071 Cocaine 0.440 1319 0.482 1908 -0.042 0.018 .0880 -2.356 009 -0.077 -0.007 2.581 006* -0.071 -0.013 Crack 0.356 1319 0.392 1908 -0.036 0.017 .0712 -2.083 019 -0.070 -0.002 2.037 .023* -0.064 -0.008 Heroin 0.152 1319 0.156 1908 -0.004 0.007 .0304 -0.310 .378 -0.029 0. 021 2.045 .023* -0.025 0.017 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of the two one-sided tests has been reported; bThe equivalence interv al was defined as % of the baseline va lue (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
144 Indianapolis The results for Indianapolis, as presente d in Table 5.7., demonstrate that one drug use value (9%) was substantially different. Spec ifically, the value for Marijuana Used in the Past 72 Hours was statistically different and not statistically equivalent The traditional z was -2.194, which equals a p-value of .014 and the equivalence z was 1.293, which equals a p-value of .098. Thus, the variab le was significant for the traditional test at the .025 level and not significan t for the equivalence test. Further, four (36%) drug use variables demonstrated statistical equivalence. For the variables Urine Test Results for Cocai ne, Ever Used Marijuana, Ever Used Powder Cocaine, and Ever Used Crack Cocai ne, the smaller equivalence z was greater than 1.645 and a traditional z th at was smaller than 1.96. The 90% confidence interval of these four variables was completely contained within the equivalence interval and included zero. Additionally, three variables were different but equivalent suggesting trivial differences only. These variable s were Urine Test Results for Marijuana, Urine Test Results for Opiates, and Ever Used Heroi n. The traditional z was greater than 1.96, implying a significant difference, but the equiva lence z of these three variables was also greater than 1.645, suggesting equi valence in the sense that the 90% confidence interval was contained within the equivalence marg in. The 90% confidence interval did not include zero, however. Finally, three variables (27%) were neither statistically different nor statistically equivalent The 90% confidence interval for the va riables Powder Cocaine Used in the Past 72 Hours, Crack Cocaine Used in the Past 72 Hours, and Marijuana Used in the
145 Past 72 Hours fell partially in side and outside the equivale nce margin. The traditional z was smaller than 1.96, implying that they we re not statistically different, and the equivalence z was smaller than 1.645, sugges ting that they were not statistically equivalent either. Figure 5.6. displays the overall results for Indianapolis. The chart shows that 9% (1) of the variables were s ubstantially different, 64% (7) of the variables were not substantially different, and 27% were statistically indetermin ate. These findings support a conclusion of no substantial difference becau se less than 20% of the variables were substantially different. Figure 5.6. Distribution of Variables across the Outcome Categories in Indianapolis Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t
146Table 5.7.: Equivalenc e Test DUF and ADAM in Indianapolis, IN ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditional 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.487 1362 0.443 1545 0.044 0.019 .0974 2.375 .009 0.008 0. 080 7.633 .000* 0.014 0.074 Cocaine 0.311 1362 0.325 1545 -0.014 0.017 .0622 -0 .809 .209 -0.048 0.020 2.786 .002* -0.042 0.014 Opiates 0.051 1362 0.025 1545 0.026 0.007 .0132 3.630 000 0.012 0.040 5.054 .000* 0.014 0.038 Self-Report Drug Use Within 72 Hours Marijuana 0.302 1362 0.340 1545 -0.038 0.017 0604 -2.194 014 -0.072 -0.004 1.293 .098 -0.066 -0.010 Cocaine 0.030 1362 0.032 1545 -0.002 0.006 0060 -0.311 .378 -0.015 0. 011 0.622 .268 -0.013 0.009 Crack 0.096 1362 0.117 1545 -0.021 0.011 .0192 -1.838 .033 -0.043 0.001 -0.158 .440 -0.040 -0.002 Heroin 0.013 1362 0.012 1545 0.001 0.004 0026 0.242 .405 -0.007 0. 009 0.871 .192 -0.006 0.008 Ever Used Drug Marijuana 0.811 1362 0.801 1545 0.010 0.015 .1622 0.681 .248 -0.019 0. 039 11.725 .000* -0.014 0.034 Cocaine 0.318 1362 0.348 1545 -0.030 0.017 0636 -1.715 .043 -0.064 0. 004 1.920 .027* -0.059 -0.001 Crack 0.291 1362 0.316 1545 -0.025 0.017 0582 -1.465 .071 -0.058 0. 008 1.945 .026* -0.053 0.003 Heroin 0.066 1362 0.112 1545 -0.046 0.010 0132 -4.393 .000 -0.067 -0.025 -3.133 .001* -0.063 -0.029 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; a The highest p value of the two one-sided tests has been reported; bThe equivalence interval was defined as % of the baseline value (ADAM); p .025 for traditional signi ficance test two-tailed; *p .05 or equivalence test, one-tailed
147 Miami Table 5.8. shows the outcome of the equiva lence analysis for Miami. One value (9%) (Urine Test Result for Cocaine) was substantially different because the traditional z-score was -3.509, which is significant at the .025 level, and the z-score for the equivalence test was -0.008, which is be low 1.645 and therefore not statistically significant. The equivalence interval was .0898 and the 90% equi valence confidence interval was -.132 for the lower limit and -.048 for the upper limit. Thus, the 90% equivalence interval fell part ially inside and outside the eq uivalence interval and did not include zero. Five variables (45%) were statistically e quivalent and not diffe rent and therefore can be considered equivalent These equivalent variables were Urine Test Result for Marijuana, Marijuana Used in the Past 72 Hours, Ever Used Marijuana, Ever Used Cocaine, and Ever Used Heroin. The traditional z for the five variables was below 1.96 and as a result not si gnificant and the equivalen ce z was above 1.645, indicating statistical equivalence. The 90% equivalence confidence inte rval for these variables was fully contained within the equiva lence margin and included zero. Two variables (18%) fell into the category different but equivalent. These variables were Urine Test Result for Opiate s, and Heroin Used in the Past 72 Hours. Both variables had traditional z-scores above 1.96, which suggests a significant difference. Their equivalence z-scores we re, however, also above 1.645, implying that they were equivalent for DUF and ADAM. T hus, the differences, al though statistically significant, are probably small. Finally, three variables (27 %) could not be judged either way because they were
148 neither statistically different nor statistically equivalent. The variables classified into the indeterminate category were Powder Cocaine Used in the Past 72 Hours, Crack Cocaine Used in the Past 72 Hours, and Ever Used Crack. As shown in figure 5.7., only one value (9%) was substantially different and seven variables (64%) did not show substantial differences. Additi onally, three variables (27%) had to be classified as indeterminate. Figure 5.7. Distribution of Variables across the Outcome Categories in Miami Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t Due to the finding that less than 20% of the drug use estimates were substantially different and less than one-third was statistically indeterminate, the overall conclusion for Miami is that the drug use estimates are not substantially different.
149Table 5.8.: Equivale nce Test DUF and ADAM in Miami, FL ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditi onal 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.353 535 0.308 1272 0.045 0.024 .0706 1.846 .033 -0.003 0. 093 4.741 .000* 0.005 0.085 Cocaine 0.449 535 0.539 1272 -0.090 0.026 .0898 -3 .509 .000 -0.140 -0.040 -0 .008 .497 -0.132 -0.048 Opiates 0.047 535 0.022 1272 0.025 0.010 .0094 2.492 .006 0.005 0. 045 3.429 .000* 0.008 0.042 Self-Report Drug Use Within 72 Hours Marijuana 0.221 535 0.223 1272 -0.002 0.021 .0442 -0.093 .463 -0.044 0.040 1.972 .024* -0.037 0.033 Cocaine 0.097 535 0.113 1272 -0.016 0. 016 .0194 -1.027 .152 -0.047 0.015 0.218 .416 0.042 0.010 Crack 0.121 535 0.149 1272 -0.028 0.017 .0242 -1.621 .053 -0.062 0. 006 -0.220 .413 0.056 0.000 Heroin 0.036 535 0.015 1272 0.021 0.009 .0072 2.401 .008 0.004 0.038 3.225 .000* 0.007 0.035 Ever Used Drug Marijuana 0.602 535 0.633 1272 -0.031 0.025 .1204 -1.235 .108 -0.080 0.018 3.560 .000* -0.072 0.010 Cocaine 0.370 535 0.384 1272 -0.014 0. 025 .0740 -0.562 .287 -0.063 0.035 2.406 .008* -0.055 0.027 Crack 0.219 535 0.257 1272 -0.038 0.022 .0438 -1.735 .040 -0.080 0. 004 0.268 .397 -0.074 -0.002 Heroin 0.090 535 0.076 1272 0.014 0. 014 .0180 0.970 .166 -0.014 0. 042 2.217 .014* -0.010 0.038 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of he two one-sided tests has been reported; bThe equivalence interval wa s defined as % of th e baseline value (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
150 New Orleans Table 5.9. displays the findings for New Orleans. The drug use estimates for Ever Used Powder Cocaine and Ever Us ed Crack Cocaine were categorized as different The 90% equivalence confidence interval for both variables fell partially inside and partially outside the equivalence interval and did not include zero. Specifically, the equivalence interval for Ever Used Powd er Cocaine was .0558. The lower limit of the 90% equivalence confidence interval was -.090 and the upper limit was -.036. Similarly, the equivalence interval for Ever Used Crack Cocaine was .0512. The lower limit of the 90% equivalence confidence interval wa s -.083 and the upper limit was -.029. Also, the traditional z-value for Ever Used Po wder Cocaine was -3.781 and for Ever Used Crack Cocaine was -3.449, which is associat ed with a significant p-value of .000 for both variables. The smaller equivalence zvalue for the two va riables was below 1.645, indicating that there is no e quivalence. Thus, these two va riables were classified as substantially different. Four variables (36%) demonstrated equivalence for DUF and ADAM, that is, the drug use estimates for Marijuana Used in th e Past 72 Hours, Heroin Used in the Past 72 Hours, Ever Used Marijuana, and Ever Us ed Heroin were statistically equivalent and not statistically different. The 90% equivale nce confidence interval of these variables fell within the equivalence margin and incl uded zero. Also, the trad itional z-value was below 1.96, suggesting that no statistically si gnificant differences existed between the drug use estimates for DUF and ADAM and th e equivalence z-value was greater than 1.645, indicating that the drug use estimat es were statistically equivalent. Additionally, five drug use variables (45%) showed trivial differences only
151 because they fell into the category different but equivalent These five variables were Urine Test Results for Marijuana, Urine Test Results for Cocaine, Urine Test Results for Opiates, Powder Cocaine Used in the Past 72 Hours, and Crack Cocaine Used in the Past 72 Hours. The traditional z-value for these variables was above 1.96, indicative of a statistically significant difference. At the same time, the smaller equivalence z was also statistically significan t demonstrating that th e drug use estimates of these five variables were equivalent for DUF and ADAM. Accordingly, the five variables were classified as not substantially different. All of the drug use variables were classi fied as either subs tantially different, equivalent, or equivalent but different. Figure 5.8. shows th e distribution of the drug use estimates across the outcomes categories. About 18% of the drug use estimates were substantially different, and 82% were classified as not subs tantially different. Based on these results, the overall finding for New Orlean s is that there is no substantial difference between the drug estimates for DUF and ADAM because less than 20% of the drug use variables demonstrated substantial differences.
152 Figure 5.8. Distribution of Variables A cross the Outcome Categories in New Orleans Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t
153Table 5.9.: Equivalence Test DUF and ADAM in New Orleans, LA ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditi onal 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.454 1236 0.382 1959 0.072 0.018 .0908 4.018 .000 0.037 0. 107 9.086 .000* 0.043 0.101 Cocaine 0.350 1236 0.457 1959 -0.107 0.018 .0700 -6 .070 .000 -0.142 -0.072 -2 .099 .018* -0.136 -0.078 Opiates 0.151 1236 0.118 1959 0.033 0.013 0302 2.635 .004 0.008 0.058 5.046 .000* 0.012 0.054 Self-Report Drug Use Within 72 Hours Marijuana 0.350 1236 0.331 1959 0.019 0.017 .0700 1.102 .136 -0.015 0. 053 5.163 .000* -0.009 0.047 Cocaine 0.057 1236 0.085 1959 -0.028 0.009 0114 -3.070 .001 -0.046 -0.010 -1.820 .034* -0.043 -0.013 Crack 0.121 1236 0.173 1959 -0.052 0.013 .0242 -4.123 .000 -0.077 -0.027 -2 .204 .014* -0.073 -0.031 Heroin 0.112 1236 0.092 1959 0.020 0. 011 .0224 1.803 .036 -0.002 0. 042 3.821 .000* 0.002 0.038 Ever Used Drug Marijuana 0.758 1236 0.773 1959 -0.015 0.015 .1516 -0.972 .166 -0.045 0. 015 8.855 .000* -0.040 0.010 Cocaine 0.279 1236 0.342 1959 -0.063 0.017 0558 -3.781 .000 -0.096 -0.030 -0.432 .333 -0.090 -0.036 Crack 0.256 1236 0.312 1959 -0.056 0.016 .0512 -3.449 .000 -0.088 -0.024 -0 .296 .394 -0.083 -0.029 Heroin 0.206 1236 0.208 1959 -0.002 0.015 .0412 -0.136 .444 -0.031 0. 027 2.665 .004* -0.026 0.022 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of the two one-sided tests has been reported; bThe equivalence interval wa s defined as % of th e baseline value (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
154 Phoenix The results for Phoenix, as presented in Table 5.10., are somewhat different than the patterns of the previous five sites. The main difference is that Phoenix has the largest amount of drug use estimate s that are substantially different. Specifically, in Phoenix four variables (36%) fell into the category stat istically different a nd not statistically equivalent. These were Urine Test Result for Cocaine, and Urine Test Result for Opiates, Cocaine Used in the Past 72 Hour s, and Ever Used Heroin. The traditional z-values for these drug estimates were greate r 1.96 and as a result significant at the .025 level. Also, the smaller equivalence z-valu es were not significant, suggesting that equivalence does not exist. The 90% equiva lence confidence intervals for these four variables were not contained within the equi valence margin. Rather, they were partially inside and partially outside the ma rgin and did not include zero. Three drug use variables (27%) fell into the category statistica lly equivalent and not statistically different and were classified as equivalent These three variables were Ever Used Marijuana, Ever Used Powder Cocaine, and Ever Used Crack Cocaine. The 90% equivalence confidence intervals for all three variables fell completely inside the equivalence margin, and the equivalence z-values were statistically significant, indicating equivalence. Finall y, the traditional z-values we re not significant, suggesting that no statistically significant differences existed. Another three drug use variablesUrine Test Result for Marijuana, Marijuana Used in the Past 72 Hours, and Heroin Us ed in the Past 72 H ours fell into the category different but equivalent These three variables show ed statistically significant differences but they were also statistically equivalent and thus categorized as not
155 substantially different. One drug use estimate (Crack Cocaine Used in the Past 72 Hours) was classified as indeterminate because the dr ug use variables for DUF and ADAM were not statistically different and not statistically equivalent. The 90% e quivalence confidence interval for this value fell partially inside and partially outside the equivalence margin and did not contain zero. Th e equivalence margin was .0252. The lower limit of the confidence interval was -.038 and the upper limit was -.002. Figure 5.9. summarizes the results for Phoe nix. These findings suggest that there is some evidence for substantial differences between the drug estimates for DUF and ADAM. Specifically, 36% of the drug use es timates indicated substantial differences. Only about 55% of the drug use estimates showed no substantial differences and one variable could not be assessed. In accordance with the decision rules outlined earlier the results for Phoenix are interpreted as substa ntially different because more than 20% of the drug use estimates were classified as different. Figure 5.9. Distribution of Variables across the Outcome Categories in Phoenix Substantially Different Not Substantially Different Indeterminate 0 20 40 60 80 100 P e r c e n t
156Table 5.10.: Equivale nce Test DUF and ADAM in Phoenix, AZ ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditi onal 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.359 2850 0.311 1611 0.048 0.015 0718 3.283 .000 0.019 0. 077 8.194 .000* 0.024 0.072 Cocaine 0.276 2850 0.318 1611 -0.042 0.014 .0552 -2. 935 .002 -0.070 -0.014 0.923 .179 -0.066 -0.018 Opiates 0.062 2850 0.079 1611 -0. 017 0.008 .0124 -2.099 .018 -0.033 -0 .001 -0.568 .285 -0.030 -0.004 Self-Report Drug Use Within 72 Hours Marijuana 0.261 2850 0.220 1611 0.041 0.013 .0552 3.106 .000 0.015 0.067 7.062 .000* 0.019 0.063 Cocaine 0.050 2850 0.066 1611 -0.016 0.007 .0100 -2 .159 .015 -0.031 -0.001 -0 .810 .209 -0.028 -0.004 Crack 0.126 2850 0.146 1611 -0.020 0.011 0252 -1.857 .032 -0.041 0.001 0. 483 .319 -0.038 -0.002 Heroin 0.045 2850 0.073 1611 -0.028 0.008 .0090 -3.706 .000 -0.043 -0. 013 -2.515 .006* -0.040 -0.016 Ever Used Drug Marijuana 0.807 2850 0.814 1611 -0.007 0.012 .1614 -0.574 .283 -0.031 0. 017 12.665 .000* -0.027 0.013 Cocaine 0.508 2850 0.498 1611 0.010 0.016 .1016 0.642 .261 -0.021 0.041 7.161 000* -0.016 0.036 Crack 0.395 2850 0.416 1611 -0.021 0.015 .0790 -1.371 .085 -0.051 0. 009 3.786 .000* -0.046 0.004 Heroin 0.175 2850 0.219 1611 -0.044 0.013 .0350 -3 .514 .000 -0.069 -0.019 -0 .719 .764 -0.065 -0.023 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of the two one-sided tests has been reported; bThe equivalence interv al was defined as % of the baseline va lue (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
157 Portland The results for Portland, as presented in Table 5.11., demonstrate that three (27%) of the drug use variables were classified as different. These variables were Urine Test Results for Opiates, Ever Used Powder Co caine, and Ever Used Heroin. All three variables had traditional z-scores above 1.96, and smaller equivalence z-values fell below 1.645. The 90% equivalence confidence interval of these variables was not contained within the equivalence margin and did not include zero. Also, three drug use variables (27%) fell into the category equivalent ; including Urine Test Results for Marijuana Marijua na Used Within the Past 72 Hours, and Ever Used Crack Cocaine. For these three variables, the 90% e quivalence confidence interval fell within the equivalence margin and included zero. The traditional z was not statistically significant indicating that there was no significant difference, and the smaller equivalence z was statistically significant, suggesting that th e drug use estimates for these variables were equivalent for DUF and ADAM. Additionally, five (45%) of the drug use estimates were different but equivalent These estimates were Urine Test Result fo r Cocaine, Powder Cocaine Used Within the Past 72 Hours, Crack Cocaine Used W ithin the Past 72 Hours, Heroin Used Within the Past 72 Hours, and Ever Used Marijuana. For these variables the traditional z was statistically significant, indi cating that there were significant differences for the drug estimates of DUF and ADAM. At the same time, the smaller equivalence z was statistically significant and the 90% equivalence confid ence interval fell within the equivalence margin, suggesting that the diffe rences between the drug estimates for DUF and ADAM were only slightly different.
158 None of the drug use variables were stat istically indeterminate, that is, all variables were classified as either substantially different, equivalent or equivalent but different Figure 5.10. shows the findings for Portland. Overall, more than 20% of the drug use estimates are substantially different for DUF and ADAM and as a result the conclusions for Portland are that th ere are substantial differences. Figure 5.10. Distribution of Variables Acro ss the Outcome Categories in Portland
159Table 5.11.: Equivale nce Test DUF and ADAM in Portland, OR ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditi onal 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.354 1295 0.375 1401 -0.021 0.019 0708 -1.132 .129 -0.057 0. 015 2.685 .000* -0.052 0.010 Cocaine 0.236 1295 0.327 1401 -0.091 0.017 .0472 -5 .286 .000 -0.125 -0.057 -2.544 .005* -0.119 -0.063 Opiates 0.114 1295 0.148 1401 -0.034 0.013 .0228 -2.623 .004 -0.059 -0.009 -0.864 .194 -0.055 -0.013 Self-Report Drug Use Within 72 Hours Marijuana 0.247 1295 0.248 1401 -0 .001 0.017 .0494 -0.060 .476 -0.034 0.032 2.909 .002* -0.028 0.026 Cocaine 0.050 1295 0.079 1401 -0.029 0.009 0100 -3.081 .001 -0.047 -0.011 -2.018 .022* -0.044 -0.014 Crack 0.099 1295 0.140 1401 -0.041 0.012 0198 -3.295 .000 -0.065 -0.017 -1.704 .044* -0.061 -0.021 Heroin 0.065 1295 0.110 1401 -0.045 0.011 0130 -4.164 .000 -0.066 -0.024 -2.961 .002* -0.063 -0.027 Ever Used Drug Marijuana 0.841 1295 0.874 1401 -0.033 0.013 .1682 -2.447 .007 -0.059 -0. 007 10.025 .000* -0.055 -0.011 Cocaine 0.465 1295 0.541 1401 -0.076 0.019 0930 -3.955 .000 -0.114 -0.038 0.885 .812 -0.108 -0.044 Crack 0.411 1295 0.445 1401 -0.034 0.019 0822 -1.784 .037 -0.071 0.003 2.529 .006* -0.065 -0.003 Heroin 0.242 1295 0.310 1401 -0.068 0.017 0484 -3.964 .000 -0.102 -0.034 -1.142 .127 -0.096 -0.040 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of the two one-sided tests has been reported; bThe equivalence interv al was defined as % of the baseline va lue (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
160 San Antonio The results for San Antonio are shown in Table 5.12. None of the drug use variables imply a substantial difference that is, none of them we re statistically different and not statistically equivalent. Six variables (55%) were classified as equivalent because the smaller equivalence z-values of these drug estimates were statistically significant, indicating that they were equivalent for DUF and ADAM. Contrary, the traditional zvalues were not statistically significant, pr oviding evidence that there were no significant differences for these drug estimates for DUF and ADAM. These variables were the drug use estimates for Urine Test Result for Ma rijuana, Urine Test Result for Cocaine, Marijuana Used in the Past 72 Hours, C rack Cocaine Used in the Past 72 Hours, Ever Used Marijuana, and Ever Used Heroin. The 90% equivalence confidence interval for these six variables was completely contained within the equivalence margin and included zero. Two variables (18%) fell into the category different but equivalent. This category included the variables Ever Used Mariju ana and Ever Used Crack Cocaine. The traditional z-value for these two variables showed a statistically significant difference between the drug estimates but the smaller equi valence z was also stat istically significant, indicating that the differen ce was not substantial. The 90% equivalence confidence interval for these two variables fell completely within the equivalence margin but did not include zero. Finally, three variables (27%) fell into the category not different and not equivalent and as a result were classified as statistically indeterminate. These variables were Urine Test Analysis for Opiates, P owder Cocaine Used in the Past 72 Hours,
161 and Heroin Used in the Past 72 Hours. Figure 5.11. displays the summary results fo r San Antonio. These results indicate that 73% of the drug use estimates were not substantially different for DUF and ADAM and that none of drug use estimates suggested substantial differences. Three variables could not be judged either way. Overall, the results can be sa id to be not substantially different. Figure 5.11. Distribution of Variables Across the Outcome Categories in San Antonio
162Table 5.12.: Equivalenc e Test DUF and ADAM in San Antonio, TX ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditi onal 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.365 1122 0.377 1840 -0.012 0.018 0730 -0.656 .256 -0.048 0. 024 3.337 .000* -0.042 0.018 Cocaine 0.253 1122 0.266 1840 -0.013 0.017 .0506 -0.785 .216 -0.045 0.019 2. 269 .011* -0.040 0.014 Opiates 0.091 1122 0.099 1840 -0.008 0.011 .0182 -0.724 .235 -0.030 0.014 0. 923 .179 -0.026 0.010 Self-Report Drug Use Within 72 Hours Marijuana 0.263 1122 0.250 1840 0.013 0.017 .0526 0.784 .218 -0.019 0.045 3.958 .000* -0.014 0.040 Cocaine 0.085 1122 0.089 1840 -0.004 0.011 .0170 -0.376 .353 -0.025 0.017 1.221 .111 -0.022 0.014 Crack 0.033 1122 0.028 1840 0.005 0.007 0066 0.760 .224 -0.008 0.018 1. 764 .039* -0.006 0.016 Heroin 0.058 1122 0.051 1840 0.007 0.009 .0116 0.808 .212 -0.010 0.024 2.148 .016 -0.007 0.021 Ever Used Drug Marijuana 0.700 1122 0.639 1840 0.061 0.018 .1400 3.450 .000 0.026 0.096 11.370 .000* 0.032 0.090 Cocaine 0.356 1122 0.323 1840 0.033 0.018 .0712 1.836 .034 -0.002 0. 068 5.796 .000* 0.003 0.063 Crack 0.148 1122 0.122 1840 0.026 0.013 .0296 1.991 .023 0.000 0.052 4.257 .000* 0.005 0.047 Heroin 0.135 1122 0.136 1840 -0.001 0.013 0270 -0.077 .469 -0.026 0.024 2.006 .023* -0.022 0.020 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of the two one-sided tests has been reported; bThe equivalence interval wa s defined as % of th e baseline value (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
163 San Jose Finally, the results for San Jose are show n in Table 5.13. Only one variable (9%) proved to be different (Urine Test Result for Opiates). The traditional z-value showed a statistically significant difference for th e drug estimates for DUF and ADAM. The smaller equivalence z-value fell below 1.645, indicating that equivalence did not exist. The equivalence margin for this variable was .0070, and the 90% equivalence confidence interval was -.030 for the lower limit and -.004 for the upper limit. That shows that the confidence interval falls partially inside and outside the equivalence margin and does not include zero. Three variables (27%) fell into the category equivalent including Urine Test Result for Cocaine, Crack Used within th e Past 72 Hours, and Ever Used Heroin. The traditional z-value did not demonstrate a statistically significant difference and the smaller equivalence z-value did suggest th at there is equivalence between the drug estimates. The 90% equivalence confidence in terval fell fully within the equivalence margin for these drug estimates and included zero. Additionally, five variables (45%) we re categorized as statistically different and equivalent suggesting the presence of only trivial differences These five variables included Urine Test Result fo r Marijuana, Marijuana Used within the Past 72 Hours, Ever Used Marijuana, Ever Used Powder Cocaine, and Ever Used Crack Cocaine. These five variables showed statistically significant differences but they also showed statistically significant equivalence. Two variables (18%) could not be st atistically determined. They were neither statistically different nor equivalent These two variables were Cocaine Used in the Past
164 72 Hours and Heroin Used in the Past 72 Hours. The 90% confid ence interval for these two variables fell partially inside and partially outside the e quivalence margin but included zero. Figure 5.12. summarizes the results for San Jose. Overall, 71% of the variables demonstrated no substantial differences. Only one variable was substantially different and two variables could not be statistically as sessed. Consistent with the decision criteria the site of San Jose can be said to have drug estimates for DUF and ADAM that are not substantially different. Figure 5.12. Distribution of Variables A cross the Outcome Categories in San Jose
165Table 5.13.: Equivale nce Test DUF and ADAM in San Jose, CA ADAM 2000/2001 DUF 1997/1998 Difference Equivalence Traditi onal 95% CI Equivalence 90% CI Variable p1 n1 p2 n2 DIF. S.E. Criterion z p LCL UCL z p LCL UCL Urine Test Marijuana 0.351 1248 0.275 1319 0.076 0.018 .0702 4. 160 .000 0.040 0.112 8.003 .000* 0.046 0.106 Cocaine 0.120 1248 0.118 1319 0.002 0.013 .0240 0.156 .440 -0.023 0.027 2.033 .023* -0.019 0.023 Opiates 0.035 1248 0.052 1319 -0.017 0.008 .0070 -2. 118 .017 -0.033 -0.001 -1.246 .106 -0.030 -0.004 Self-Report Drug Use Within 72 Hours Marijuana 0.263 1248 0.189 1319 0.074 0.016 .0526 4.491 .000 0.042 0. 106 7.683 .000* 0.047 0.101 Cocaine 0.032 1248 0.040 1319 -0.008 0.007 .0064 -1.089 .138 -0.022 0. 006 -0.218 .414 -0.020 0.004 Crack 0.042 1248 0.034 1319 0.008 0.008 0084 1.058 .159 -0.007 0.023 2. 170 .015* -0.004 0.020 Heroin 0.018 1248 0.028 1319 -0.010 0.006 .0036 -1.695 .045 -0.022 0. 002 -1.085 .139 -0.020 0.000 Ever Used Drug Marijuana 0.714 1248 0.670 1319 0.044 0.018 .1428 2.418 .008 0.008 0.080 10.264 .000* 0.014 0.074 Cocaine 0.443 1248 0.379 1319 0.064 0.019 .0886 3.300 .000 0.026 0.102 7.868 .000* 0.032 0.096 Crack 0.276 1248 0.207 1319 0.069 0.017 .0552 4.090 .000 0.036 0.102 7.363 .000* 0.041 0.097 Heroin 0.118 1248 0.107 1319 0.011 0.012 0236 0.881 .189 -0.013 0.035 2. 772 .003* -0.010 0.032 Note: Dif. = difference p1-p2; S.E. = standard error; CI = confidence interval; LCL = Lower Confidence Limit; UCL = Upper Confidence Limit; aThe highest p value of the two one-sided tests has been reported; bThe equivalence interv al was defined as % of the baseline va lue (ADAM); p .025 for traditional significance test two-tailed; *p .05 for equivalence test, one-tailed
166 The Impact of Different Alpha Levels and the Inclusion/Exclusion of Indeterminate Values Table 5.14 shows that changes in the outcomes due to usi ng different alpha levels and manipulating the inclusion and exclusion of the values found to be statistically indeterminate. The data shows that these manipulations can make a difference with regard to the findings. The analysis of all possible combinations is beyond the scope of this dissertation, and as a resu lt, only four possibilities wi ll be discussed. These four possibilities are shown in Table 5. Column one shows the results for the original analysis using a .05 alpha level for the traditional null hyp othesis test and including the values that were statistically indeterminate in the analysis. Column two presents the findings if the indeterminate values would have been excl uded from the calculation. The alpha level was kept at .05 to determine changes only due to the exclusion of cer tain values. Column three shows the results if a .01 alpha level would have been used instead of a .05 level. The indeterminate values were included in the analysis (as in the original analysis) to determine changes due solely to the more conservative alpha level. Column four illustrates the findings for a change in th e alpha level to .01 and the exclusion of indeterminate values. With regard to the indeterminate values, it could be argued that the results of the analysis might be different if the values that were indete rminate would be excluded. For instance, assume that one site had two valu es that showed substantial differences, six values that were similar, and three values that could not be determined as either different or similar because neither the traditional nor th e equivalence test were significant. In the original analyses the total num ber of values would be eleve n. Out of these eleven values,
167 two values (18%) were substantially different six values (55%) were similar, and three values (27%) were indeterminate. The conclu sion would be that there are no substantial differences because less than 20% of the va lues demonstrated substantial differences. This conclusion might be different if the th ree values that could not be statistically determined would be excluded; that is, the total number of values that can be evaluated is eight rather than eleven. For the current exam ple this means that two out of eight values would be substantially different, which equals 25%; and six values are similar, which equals 75%. The conclusion would be that this site demonstrates s ubstantial differences because more than 20% of the values demonstrated substantial differences. Thus, the question is whether the findings of the current study would be different if the indeterminate values would be excluded. Column two presents the results for this analysis. Column one is the reference category presenting the results from the orig inal analysis. Column two suggests that excluding the indeterminate values would lead to differences in the results for two sites. In Dallas, the results would now indicate that there are no substantial differences. Specifically, in Dallas, none of the variables were categorized as s ubstantially different, three variables were equivalent, three variab les were different but equivalent, and five variables could not be statis tically assessed. Thus, the ne w number of total variables would be six (11 5) and the conclusion w ould be that since no variable showed substantial differences the data for Dallas can be said to be similar and not substantially different. This is different from the origin al conclusion in that Dallas was categorized originally as indeterminate because more th an one third of the values could not be statistically judged to be different or similar.
168 The second site that would be categorized differently if the indeterminate values would be excluded is Denver. In Denver one va riable was indeterminate, which results in a new total of variables. The new total would be 10 (11 1) In Denver two values were categorized as substantially different, which is exactly 20%. Thus, Denver could be said to show substantial differences across the DUF and ADAM data. None of the remaining seven sites would have been categorized differently. This means that overall, if the indeterminate values would be excluded from the analysis, six sites would be said to be sim ilar and three sites would be said to be substantially different (as shown in column two in Table 5). The c onclusions from the original analysis were that one site could not be categorized as either similar or different because too many values were indeterminate, two sites were substantially different, and six sites were similar. With regard to the alpha level, it could be argued that the alpha-level used for the analysis should be .01 instead of .05 because the sample sizes (N) used in the current analysis are large. Due to the fact that N influences the outcome of the traditional significance test, it would be reas onable to use the smaller alpha level to reduce the Type I error. A Type I error leads us to rej ect the null hypothesis of no difference and conclude that there is a difference when in fact the null hypothesis of no difference is true. At the .05 level there is a 5% chance that the null hypot hesis of no difference is correct given these data. Accordingly, at the .01 level there is a 1% chance that the null hypothesis of no difference is correct given these data. Thus, the .01 level is the more conservative level. The greater the sample size, the smaller or more conservative the alpha level should be in order to avoid a Type I Error.
169 If the alpha level is more conservative, however, the chances of making the Type II Error increase, that is, researchers will erroneously accept the null hypothesis of no difference when in fact there is a significan t difference and the null hypothesis should be rejected. Researchers have to decide which alpha level is suited best for their data. For the current study, the sample sizes vary between 535 and 2,850 to ADAM and 1,272 to 1,959 for DUF (as shown in Table 3.4 in Chapter Three). Thus, according to Cohen's Power Primer, an alpha level .01 could be used instead of the .05 level. Column Three in Table 5 presents the results for the analysis if an alpha leve l of .01 would have been used in the current analysis instead of .05 (as shown in Column One). Seve n of the nine sites would have been classified into the same category at the .01 and the .05 alpha level. These seven sites are Dallas (Unknown), De nver (Similar), Miami (Similar), New Orleans (Similar), Portland (Different) San Antonio (Similar), and San Jose (Similar). Two sites would have been categorized differently. First, Indianapolis was originally classified as Similar. Using th e .01 alpha level the resu lts for Indianapolis would have changed and it would have been classified as unknow n. Specifically, in Indianapolis, marijuana used w ithin the last 72 hours was sta tistically different at the .05 level but not at the .01 level. There was no st atistical equivalence fo r this variable, and thus the variable was classified as indeterm inate at the .01 level, increasing the number of indeterminate values to four, which is 36%. Thus, at the .01 alpha level, more than one third of the variables were st atistically indeterminate and as a result Indianapolis cannot be said to be either different or similar. No judgment can be made for this site and it would be categorized as unknown.
170 Second, Phoenix would have been classified as similar at the .01 level whereas it was different at the .05 level. This change occurred because at the .05 level four variables were statistically diffe rent and not equivalent, and as a result more than 20% of the variables were different leading to a classi fication of substantially different. At the .01 level, only two variables were statistically different and not equiva lent, which is 18%, and as a result Indianapolis would ha ve been categorized as similar. Finally, the question is what would happen if the variables that are statistically indeterminate would be excluded from the cal culation and an alpha level .01 would be used instead of .05. Column four shows th e findings for this analysis. Again, the reference category is Column one (origina l analysis). Comparing Column four and Column one shows that eight of the nine sides would have be en categorized into the same category. Only one side (Dallas) would have been judged differently, that is Dallas would have been classified as similar instead of unknown. The results presented in Table 5 suggest that the classification for five sides (Miami, New Orleans, Portland, San Antonio, a nd San Jose) did not change regardless of the intervention. Stated differently, the findings for these five sides were the same in the original analysis, at the .01 le vel, and excluding the values th at were indeterminate. This suggests that the results for th ese sites are very stable and the conclusions can be drawn with confidence. For four siTes, however, some changes took place when using a different alpha level and excluding the values that were inde terminate from the analysis. Dallas showed the most changes. In the original analys is (Column One) and at the .01 alpha level (Column Three) Dallas was categorized as unknown. This classification would have
171 changed if the indeterminate values would ha ve been excluded (Column Two) and if the analysis would have employed a .01 alpha level and excluded the indeterminate values (Column Four). In this case Dallas w ould have been judged to be similar. The findings for Indianapolis would ha ve changed from being similar to unknown if the .01 level would have been us ed with everything else being the same (Column Three). At all other c onditions, Indianapolis was clas sified as similar. The site Phoenix would be classified as different under all conditions except when using the .01 alpha level instead of .05 (Column Three). De nver was classified as similar in all circumstances except when the indeterminate variables were excluded from the analysis (Column Two), in which situation Denver would have been cl assified as different. Overall, it appears that the results of the current analysis are very consistent for some sites, but also show some difference s across different condi tions. The consistency of the findings is especially obvious in colu mn four as compared to column one, where only one site (Dallas) would have been classi fied differently, as similar rather than unknown, if both changes (using a .01 alpha level and excluding i ndeterminate values) would have been made. This would strength en the conclusion that the drug estimates contained in the DUF and ADAM data are simila r. Further, the data suggests that the drug use estimates contained in the DUF and ADAM data show no differences for five sites regardless of the intervention. These si tes are Miami, New Orleans, Portland, San Jose, and San Antonio. There are, however, also some differences. Classification would have changed for four sites: Dallas, Denve r, Indianapolis, and Phoenix. There is no consistency with regard to the direction of change. The classification of Dallas and Indianapolis varied between similar and unknown, Denver and Phoenix varied
172 between similar and different. The implications of these changes will be discussed in the final chapter. Certainly, the changes of the alpha level and the inclusion an d exclusion criteria of indeterminate values can be expected to lead to some ch anges in the findings, which is also true for the current study. Using a more conservative alpha level decreases the number of values that are statistically differe nt in the traditional null hypothesis test, and as a result reducing the number of variables and sides ca tegorized as substantially different. Excluding values that are statistically indeterminate lead to only one change, namely that Denver was classified as diffe rent rather than similar. A comprehensive analysis of all possible cond itions and changes is beyond the scope of this study. These changes due to different condi tions are, however, very impor tant as they can lead to substantial differences in the results of empirical studies. Thus, future research should address how different alpha levels, inclus ion and exclusion crite ria and equivalence intervals influence study outcomes and inferen ces and possible policy implications made from these studies.
173Table: 5.14 Differences in Outcomes by Using different Alpha Levels and Changes in the Inclusion and Exclusion Criteria Original Indeterminate Excluded Alpha-Level .01 Alpha-Level .01 & Indeterminate Excluded Site Different Similar Unknown Different Similar Unknown Different Similar Unknown Different Similar Unknown Dallas x x x x Denver x x x x Indianapolis x x x x Miami x x x x New Orleans x x x x Phoenix x x x x Portland x x x x San Antonio x x x x San Jose x x x x
174 CHAPTER SIX DISCUSSION AND CONCLUSION This Chapter provides a review of th e study purpose, the results that emerged from the statistical analysis, and the inferen ces that can be drawn. This is followed by a discussion of the implications and limita tions of the current study. This Chapter concludes by conferring the contributions and possible extensions of this work and opportunities for future research in this area. Major Goal and Possible Outcomes of the Study The major goal of the current study is to assess whether the drug estimates for selected drugs are similar or different be tween DUF and ADAM. It was hypothesized that the drug use information in the two samples mi ght not be substantially different for two main reasons: (1) both the probability samp le of ADAM and the non-probability sample of DUF rely on volunteers; and (2) both DUF and ADAM were only able to sample arrestees who were held in the facility long enough, resulting in a sample of more serious offenders and offenders who did not have the financial means to post bail. The analysis included the following nine sites: Dalla s, Denver, Indianapolis, Miami, New Orleans, Phoenix, Portland, San Antonio, and San Jose. These nine sites were chosen because they have the same catchment area for DUF and ADAM and sufficient data for each time period to allo w for a meaningful comparison. The variables included in the analysis are the same for a ll sites. These variables are urine analysis results (positive/negative) for marijuana, cocaine, and opiates; self-reported drug use
175 within the last 72 hours for marijuana, powde r cocaine, crack cocaine, heroin; and selfreported drug use for the question whether th e arrestee had ever used marijuana, powder cocaine, crack cocaine, or heroin. For the purpose of examining the rese arch question, the current study employed equivalence analysis. Using equivalence anal ysis, it was determined that there existed four possible outcomes for the current st udy. These four possible outcomes were: 1) DUF and ADAM can be said to be equivalent (Eq) 2) DUF and ADAM are different (D) 3) DUF and ADAM are different and equivalent (D&Eq) 4) DUF and ADAM are not different a nd not equivalent (indeterminate) (ND&NEq) The interpretation of the results for each site and for the selected drugs was completed in accordance with the three decisi on criteria laid out in Chapter Four. These three decision criteria were: 1) The findings of Equivalent (Eq) and Different but Equivalent (D&Eq) will be interpreted as not substantially different or similar. 2) Sites will be classified as unable to be assessed if more than one third of the drug use values fall into the category Not Statistically Different and Not Statistically Equivalent (ND&NEq). 3) Third, sites will be classified as substant ially different if 20% or more of the drug use values show substantial differences (D). Analytical Strategy of the Current Study The current study uses a research strate gy that is commonly used in medical
176 research: equivalence testing. Equivalence testing has also become more popular in other fields, however, because it is useful to a ssess the comparability of scales, groups, and other outcomes. The underlying idea is to te st whether two outcomes can be said to be equivalent or whether they are substantially different. To reiterate, substantially different does not simply mean that there is a st atistically significant difference using the traditional null hypothesis test but that the difference is of practical importance. For this type of analysis, two simultaneous tests are carried out: the traditional null hypothesis test and the equivalence test. A substantial difference can only be established if the traditional test is statistically significan t and the equivalence test is not statistically significant. The results can be said to be equivalent if the equivalence test is statistically significant and the traditional test is not. A result of different but equivalent is typically interpreted as a lack of substantial differences and an indica tion that the two outcomes are comparable (as discussed in Chapter 4). Finally, it is possible that the result of the analysis show that neither the traditional test nor the equivalence te st is statistically significant. In this case it cannot be statistically determ ined whether the results are different or equivalent. No conclusion can be drawn either way (Tryon, 2001). Equivalence testing is typically applied to test the comparability of the effect of different drugs or treatments (i.e., established drug versus alternative drug). The current study is different from clinical studies in that superiority of a certa in treatment cannot be established. That, however, was not the goal of this study. Instead, the main purpose was to determine whether the percentage outcome s for 11 drug use variables are substantially different or whether they are similar.
177 Main Findings Overall Findings for All Sites Out of a total of 99 drug use va lues, 14 (14%) were found to be substantially different 67 (68%) were classified as not substantially different or similar, and 18 (18%) values were deemed indeterminate because they were neither statistically different nor statistically equivalent. Thus, the overall analysis of all drug use values suggests that there are no substantial differences because le ss than 20% of the values were categorized as different. The site with the greatest number of values classified as equivalent and different but equivalent was New Orleans with nine values (82%), followed by Denver, Portland, San Antonio, and San Jose with eight values (73%). Seven values (64%) were classified as equivalent and different but equivalent in Indianapolis and Mi ami. Finally, in Phoenix and Dallas, six values fell into either of thes e categories. None of the sites had less than six values classified as equivalent and different but equivalent Further, the analysis shows that each site has a certain pattern with regard to the distribution of the drug use estimates across the outcome categories. In Dallas and San Antonio, neither of the drug use values was substantially different. In Indianapolis, Miami, and San Jose, one of the 11 drug use valu es was substantially different, but it was not the same for the three sites. Rather, at each site a different value was categorized as substantially different. In Denver and New Or leans, two of the drug use values were classified as substantially di fferent. Again, both sites had di fferent drug use values that were classified as substantially different. Portland had three drug use values that were substantially different, and Phoenix had four values that were classified as substantially
178 different. None of the sites had more than f our values that fell into this category. With regard to the outcome category not different and not equivalent the analysis demonstrates that only one site that had more than one-third of the drug use values categorized as indeterminate, which was Dallas with five values (45%). In Indianapolis, Miami, and San Antonio, three values (27%) could not be statistically assessed. In San Jose, two values (18%) fell into this categ ory, and in Denver and Phoenix one value (9%) was classified as indeterminate. The remain ing sites, New Orleans and Portland, had no values that were indeterminate. Following the examination of the distribution of drug use values overall, the next analysis step was to take a closer look at the specific drugs to determine whether there are pa tterns for each drug with regard to their outcomes. Overall Findings by Drug The overall results for each drug (marijuana, cocaine, and opiates) are that none was substantially different between DUF a nd ADAM. Specifically, all but one of the drug estimates for marijuana fell into the categories equivalent or different and equivalent Thus, 26 of the 27 drug use estimates fo r marijuana use, in cluding urine test results for marijuana, self-reported marijuan a use within the past 72 hours, and selfreported marijuana use over the lifetime were classified into the category no substantial difference In accordance with the decision cr iteria, the drug use estimates more marijuana can be said to show no subs tantial differences between DUF and ADAM because less than 20% of the drug use estimates were significantly different and not equivalent. The drug use estimates for cocaine consisted of urine test results for cocaine, selfreported powder cocaine use within the past 72 hours, self-reported crack cocaine use
179 within the past 72 hours, self -reported powder cocaine use ov er the lifetime, and selfreported crack cocaine use over the lifetime. Altogether, the analysis included 45 drug use estimates for cocaine. Of those 45 values for cocaine, eight values (18%) fell into the category Statistically Different and Not Statis tically Equivalent; 25 values (56%) were classified as either Statistically Equi valent and Not Statis tically Different or Statistically Different but St atistically Equivalent; and 12 values (27%) were neither statistically different nor stat istically equivalent. Less than 20% of the drug estimates demonstrated substantial differences, and as a result the drug use estimates for cocaine can also be said to show no substantial di fferences between DUF and ADAM. The results for cocaine were, however, not as clear cut as for marijuana. Specifically, the drug use estimates for cocaine demonstrat e some differences, that is, one or more of the drug use values for cocaine were classified as s ubstantial different in Denver, Miami, New Orleans, Phoenix, and Portland. With regard to opiates, the analysis in cluded the following vari ables: urine test results for opiates, self-reported heroin use within the past 72 hour s, and self-reported heroin use over the lifetime resulting in a total of 27 drug use values. Of these 27 values, five values (19%) showed substantial di fferences. Sixteen values (59%) were not substantially different; and si x values (22%) could not be statistically determined. Overall, the conclusion is that the drug use estimates for opiates are not substantially different because less than 20% of the values demonstrated substantial differences. It is also apparent, however, that there are a few di fferences. Specifically, one or more of the opiate values was categorized as substantia lly different in Phoe nix, Portland, and San Jose.
180 The question why there are differences for some sites and for cocaine and opiates but not for marijuana cannot be answered with the data currently avai lable. This question is very problematic to assess given the nature of the issues that are be ing looked at in this study. It is, however, an important question th at should be explored in future research. There are some alternative techniques that c ould be used, such as Monte Carlo simulation and perhaps a Bayesian approach to this analytic problem. Simulation studies would permit some explor ation of different scenario outcomes under alternative models of dist ribution of drug use patterns for example, comparing an aggregate theoretical site derived from th e current empirical data for both DUF and ADAM as a comparator. A Bayesian paradigm might also fit here. If one thinks of Bayesian approaches as being a technique for the assessment of informational utility it might be that coupling Bayesian analysis, perhaps with a propor tional-reduction-of-error objective, would be a useful avenue. These type s of alternative analys is methods will be discussed in more detail later in the discussion. After having assessed the overall findings for each drug, the next step in the analysis was to explore the specific findings for each site. Site Specific Findings Dallas was the only site that could not be judged either way because more than one third of the drug use values were not different and not equivalent and as a result were classified as indeterminate. The remaining eight sites had 27% or less of the variables that fell into this category. Of these eight sites, two demonstrated substantial differences, that is, in Phoenix and Portland more than 20% of the drug use values were statistically different and not equivalent Specifically, in Phoenix four values (36%) demonstrated
181 substantial differences and in Portland three values (27%) were substantially different. At the sites of Denver, Indianapolis, Miami, New Orleans, San Antonio, and San Jose, less than 20% of the drug use values were classifi ed as substantially di fferent. The analysis suggests that the outcome is site specific, th at is, each site has a specific pattern of how the specific drug use values are distributed over the outcome categories. In sum, Dallas could not be assessed; Phoenix and Portland we re categorized as substantially different; and Denver, Indianapolis, Miami, New Or leans, San Antonio, and San Jose were categorized as not substantially different or similar. Discussion of the Findings The results of this study s uggest that the overall results for all sites combined are not substantially different and that the outcomes for three drug categories are not substantially different. The site-specific anal ysis implies that two sites are substantially different, five sites are similar, and one site could not be judged either way. The majority of findings demonstrate that the drug use es timates in the DUF and ADAM data for the three major drugs are similar. There are, however, some differences, which will be discussed in more detail now. As described above, there are some variab les that do show substantial differences. The variables that fell into the category different most often were uri ne test results for cocaine, urine test results for opiates, ever used pow der cocaine, and ever used heroin. The variable urine test results fo r cocaine was substantially different for three sites: Miami, Phoenix, and Denver. For all thr ee sites there was a substantial increase in the number of arrestees who tested positive. Additionally, ever used powder cocaine was substantially different in New Orleans and Portland. The resu lts are consistent in that
182 the data shows an increase in cocaine use fr om 1997/98 to 2000/01. The same is true for ever used crack cocaine, which demonstrated a substantial increase in New Orleans. These findings suggest that cocaine might have increased within that five-year period. Another variable that showed s ubstantial differences at some sites was urine test results for opiates. Opiate use significantly incr eased between the DUF and ADAM data in Phoenix, Portland, and San Jose. Also, the variable ever used heroin substantially increased in Phoenix and Portland. The questi on of why there are differences for those three sites cannot be determined with the current data. There is also no other data currently available to researchers to explore this question. Fu ture research should further explore data collection and analyt ic strategies that might allow for a better analysis of this question. This might be very difficult, how ever, because the DUF and ADAM data is historic data and cannot be altered at this point. Consistency of Findings Despite the differences apparent in the da ta the overall findings suggest that the drug estimates of DUF and ADAM are not substa ntially different. It is noteworthy that this finding supports what a study by NIJ found, that is, that although the charge distribution for arrestees of the DUF sample differed from the charge distribution of arrestees from the UCR, the drug use estimates derived from the DUF data were almost identical to the estimates for the UCR data (NIJ, 1990). The current findings are also consistent with the study from Anchorage, which suggested that the sample of females was representative of the female arrestee popu lation despite the fact that the sample of females was a convenience sample (Myrstol and Langworthy, 2005).
183 Explaining the Results One explanation for the results of this study, namely that DUF and ADAM are not substantially different, might be that both studies used vol unteers and suffered from nonresponse bias. This is not a problem for drug re search alone but for research in general. As stated earlier, in the case of ADAM, the interview refusal rate for all sites combined was 17.5%. Additionally, of the individuals that did choose to participate in the interview (82.5%), 15.6% refused to provide a urine sample (NIJ, 2000). The DUF data shows that approximately 10% of the selected arrestees re fused to interview. Of the arrestees that agreed to participate in the survey, about 20% refused to provide a urine sample (NIJ, 1995). Non-respondents constitute a problem for a study if they are different from respondents in ways that bias the study outcome. Research suggests that non-respondents share certain characteristics (Sharp and Feldt, 1959; Hill, 1997). Similarly, volunteers also appear to share certain characteristics. Although researchers recogni ze non-response bias, it is rarely quantified. A number of studies that do attempt to quantify non-response bias suggest that nonrespondents differ from respondents in various ways. A number of researchers have found differential survey responses, meaning that some population types of the sample have significantly higher re sponse rates than other popula tion subgroups. For instance, Sharp and Feldt (1959) found that younger pe rsons are significantly more likely to respond than older persons. Response rates decline with increasi ng age. Additionally, response rates varied considerably depending on the marital status. Widowed persons were the least likely to grant an interview (74%), followed by adults who had never been married (82%). Married adults with children had the highe st response rate (93%).
184 The Oslo Health Study also assessed non-response patterns and found that a number of population sub-groups were significantly unde rrepresented. The underrepresented population subgroups were males, young persons, single/never married and divorced/separated persons, persons not born in Norway, persons with lower or unknown education level, persons with a low socio-economic status, and persons receiving disability benef its (Sogaard, et al., 2004). Additionally, Vivienne (2002) assessed differences between respondents and non-respondents in a survey about alcohol consumption and found that abstainers were overrepresented among the non-respondents, biasing the sample towards indi viduals who drink alcohol. Hill, et al. (1997) examined non-response bi as in a lifestyle survey and found that respondents and non-respondents varied significantly with regard to current smoking, hazardous alcohol consumption, and lack of moderate or vigorous exercise. More specifically, non-respondents were significantly more likely to be current smokers. This finding was also confirmed by Bostram, et al (1993), who studied smoking behavior in Sweden, and Smith and Nutbeam (1990). C ontrary, respondents showed significantly higher hazardous alcohol consumption. This finding should caution researchers against believing that non-respondents are always engaging in more risky behavior and unhealthy lifestyles than respondents. Non-response is of concern es pecially if it is associat ed with the variable of interest (Oberski, 2008). For example, if the variable of in terest is the prevalence of illicit drug abuse among school children and the majori ty of children who are using illicit drugs are either absent from school or refuse to participate in the inte rview, the researcher might draw the conclusion that illicit drug a buse among school children is very rare. If
185 the children who were absent and who refuse d would have truthfully reported their drug abuse, the researcher might have come to a different conclusion. With regard to DUF and ADAM, it is possibl e that the similar results of drug use estimates can be attributed to the fact that probability samples suffer from some of the same shortcomings as do non-probability samp les, that is, the sample consists of volunteers. Arrestees who did not volunteer to participate in the DUF study might also have refused to complete the interview in the ADAM program. To date, there is very little research that assesses this question. To this author's knowledge, there is only one study from Anchorage that looked at demogr aphic differences between respondents and non-respondents in the ADAM program. Th ey found no differences for the male probability sample and the female convenience sample. Specifically, the female convenience sample was just as representative of the population of female arrestees as the male probability sample was of the population of male arrestees (Myrstol and Langworthy, 2005). This supports the results of the current study, which showed that both DUF and ADAM produced drug use estimates that were not substantially different despite the differences in the sampling design. Even though the current study only examined the drug use information it is like ly that other information contained in DUF and ADAM is also not substantia lly different. This question s hould be explored in future studies. Implications of the Current Study The main implication of the current st udy and the two studies by NIJ and Myrstol and Langworthy (2005) is that the genera l assumption of researchers that a nonprobability sample produces estimates that are substantially different from a probability
186 sample is not necessarily correct. These th ree studies assessed only one program, DUF and ADAM, but the findings are remarkably si milar and imply that researchers can use the drug use data from both studies and expl ore research questions that have not been assessed yet because the DUF data was said to be unreliable. This implication is strengthened by the fact that these three stud ies used different anal ytical strategies and sites and still arrived at similar conclusions. With regard to policy makers this would imply that the DUF and ADAM data would lead to similar conclusions about drug use prevalence and patter ns a. Thus, if the main concern is the implementation of programs aiming to reduce drug use, the nonprobability sample DUF data would very likely be sufficient. Specifically, the DUF data as well as the ADAM data showed that some drugs are more popular than others across sites and that these differences were apparent in DUF and ADAM. It is likely that policy makers would have implemented the same programs regardless of whether they used DUF or ADAM. This is a crucial finding b ecause budget restraints currently inhibit the collection of data from arrestees nationwide. Due to this lack of data and knowledge, necessary programs are not implemented and drug trends go undetected until they show up in the general population or until the problem ha s become epidemic. Limitations of the Current Study First, one of the major limitations stem s from the data itself, specifically, only nine sites could be examined because the majo rity of sites did not either have the same catchment area or had too few cases for the ADAM sample. The implication of this limitation is that the current study cannot ma ke inferences about the similarity or dissimilarity of the DUF and ADAM data for all sites. This means that the results and
187 conclusions of this study are limited to the available sites. It is possible that the analyzed sites are dissimilar from the sites excluded from this analysis. It is also possible that the analyzed sites are not different than the si tes not included. At this point this question cannot be answered. Second, for both DUF and ADAM, it is unknown whether the arrestees who volunteered to participate and who were in the facility l ong enough to be interviewed are representative of the arrest ee population overall w ith regard to their drug use prevalence and patterns. This question is important and should be assessed in future research to determine whether arrestees who are available and volunteer to participate differ in their drug use habits from arrestees who are either not available or refuse participation. As described earlier, programs are implemente d on the basis of what is known about drug use and these programs are only effective if the information they are based upon is accurate. Thus, future research should attempt to collect data from non-respondents and compare it to the individuals who participated in the study. Third, equivalence analysis has certain limita tions in itself. First, researchers have to determine the equivalence margin. At this time there are no standard rules that are applied by all researchers using equivalence te sting. Rather, it is a discretionary decision and as a result it is possible to manipulate th e results of the study. For example, setting a higher equivalence margin with improve the chances of finding equivalent result. The opposite is true also; researchers looking for substantial differences might choose a very small equivalence margin (Gotzsche, 2006) Thus, researchers suggest that the equivalence margin should be determined based on scientific grounds, past research results, and clinical standards (LeHen ann, 2006). These recommendations are fairly
188 vague and leave much room for discretion. For social scientists it is even more difficult to find an appropriate equivalence margin than for biomedical researchers. E quivalence is more difficult to assess when factors are not completely constant. Even t hough clinical research often assumes that a certain drug produces a constant outcome this might not nece ssarily be true (Gotzsche, 2006). This is also true for the current study. The constancy of drug use over time cannot be assumed. As described above, some change s are to be expected due to the natural fluctuations in drug use prevalence and pa tterns. This has implications for the equivalence margin. For the current study, the equivalence margin was based on previous drug use research and what has been established as a substant ial difference with regard to changes in drug use among arrestees. For researchers who are using this type of analysis it is important to find an equivalence margin th at is appropriate for the research topic. Fourth, the current study found no substantial differences for the majority of variables for DUF and ADAM. Also, all of the findings appear to be in congruence with data from other drug survey. Still, it is possi ble that the differences found are due to the sampling method. Due to the limitations of pr obability samples, this possibility cannot definitely be excluded. Unfortunately, there is no dataset that could be examined against the DUF and ADAM data to determine the reason for these differences. The question about why some sites and drug use values are substantially different relates to another question. How can we determine the error associated with using the DUF and ADAM data combined and separately? The current st udy used a 20% margin, that is, if less than 20% of the drug use values were substantially di fferent then the data for this site was said to be similar enough to be considered equiva lent. Using this margin, two sites (Phoenix
189 and Portland) were categorized as substantially different. Further, depending on the alpha level and the inclusion or exclusion of th e indeterminate values the findings changed. These changes were discussed in detail earlier. These findings suggest that researchers can use the DUF data for their analysis. There is, however, a certain risk or error a ssociated with doing so. The current analysis has shown that overall 14% of the drug use es timates were substantially different. Thus, if researchers would draw one of the test s at random assuming that they are using equivalent data, there would be a .14 probabi lity that the data is not equivalent. This probability of using data that is believed to be equivalent when it is not differs depending on the drug and site. For marijuana, the probability would be 0 because none of the estimates were substantially different. For co caine, the probability would be .18, and for opiates it would be .19. With regard to the ten sites, the probability would be .36 for Phoenix, .27 for Portland, .18 for Denver and Ne w Orleans, .9 for Indianapolis, San Jose, and Miami, and .0 for Dallas and San Antonio. Final Remarks In spite of these limitations, however, th e findings demonstrate that the drug use information collected via a non-probability sampling procedure in DUF are not substantially different than the drug use information collected via a probability sampling procedure in ADAM. As a result, this dissertation presents a contribution to researchers, especially drug use researchers using the DUF and ADAM data and researchers examining hard-to-access populations, po licy makers, and law enforcement. The termination of the DUF and ADAM study and the reinstatement in 2006 by the ONDCP in only ten counties across the United States (now called ADAM II) has
190 robbed researchers of the ability to track drug using behaviors among arrestees and inform police agencies and policy makers of changes in drugs used by arrestees and prevalence of drug use across different geograp hic areas at the loca l level. The current study hopes to lay the groundwork for the im plementation of a national study that systematically tracks drug use patterns am ong arrestees similar to the NSDUH and the MTF. Considering the budget cris is, it is understood that this national study must be cost effective, yet provide valid data. One solu tion might be to supplement a probability sample with a convenience sample. In fact, some researchers have suggested that it might be possible to use a convenience sample to supplement a probability sample as a means of saving research money and to produce a sample with a sma ller mean squared error than would be possible with the probability sample given cost and time restraints The bias in the convenience sample can be reduced, fo r example, by using the known population variables to calibrate the convenience sample (Kalton and Kasprzyk, 1986). This approach allows researchers to compute infere ntial statistics from the probability sample, including the calculation of confidence intervals, standard errors, and the representativeness of data with regard to the population of interest and draw a large enough sample that allows researchers to a ssess a variety of research questions. Considering the fact that the federal government spends $50 million to study drug abuse among the general populati on who use drugs rarely, it is reasonable to expect that the government also examines drug abuse amon g arrestees, who use drugs at much higher rates than the general po pulation and school children.
191 REFERENCES Abiona, T. C., Balogun, J. A., Adefuye A. S ., & Sloan, P. E. (2009). Pre-incarceration HIV risk behaviours of ma le and female inmates. International Journal of Prisoner Health, 5 (2), 59-70. Agresti, A. & Finlay, B. (2007). Statistical methods for social sciences. Prentice Hall. Allen, E. & Seaman, C. A. (2006). Different, equivalent, or both. Quality Progress, July. Altman, D. G. and Bland, J. M. (1995). Absen ce of evidence is not ev idence of absence. British Medical Journal, 311, 485. Appel, P. W., Hoffman, J. H., Blane, H. T ., Frank, B., Oldak, R., & Burke, M. (2001). Comparison of self-report and hair anal ysis in detecting cocaine use in a homeless/transient sample. Journal of Psychoactive Drugs 33, 47-55. Bachman, J. G., Johnston, L. D., OMalley, P. M., & Schulenberg, J. E. (2006). The Monitoring the Future Project after th irty-two years: Design and procedures. Monitoring the Future, Occasional Paper 64. Michigan University, Ann Arbor: Institute for Social Research. Batanero, C. and Diaz, C. (2006). Methodologi cal and didactical controversies around statistical inference. Journes de Statistique 1-10. Beckett, K., Nyrop, K., & Pfingst, L. (2006) Race, drugs, and polic ing: Understanding disparities in drug delivery arrests. Criminology, 44 (1), 105-137. Bickel, W. K., & DeGrandpre, R. J. (1996). Drug policy and human nature: psychological perspectives on the prevention, management, and treatment of drug abuse New York: Plenum Press Brecht, M. L., Anglin, M. D., & Lu, T. H. (2003). Estimating drug use prevalence among arrestees using ADAM data: An app lication of a logistic regression synthetic estimation procedure. Research Report, National Institute of Justice: U.S. Department of Justice. Brook, J. S., Whiteman, M., Finch, S. J., & Cohen, P. (1996). Young adult drug use and delinquency: childhood antecedents and adolescent mediators. Journal of the American Academy of Child & Adolescent Psychiatry 35, 1584-1592.
192 Bureau of Justice Statistics (1998). Substance abuse and trea tment of adults on probation,1995. NCJ 1666611. Bureau of Justice Assistance (1999). Integrating drug testing into a pretrial services system: 1999 update (Publication No. NCJ 176340). Washington, DC: Bureau of Justice Assistance, Office of Justice Pr ograms, U.S. Department of Justice. Bureau of Justice Statistics (2000). Census of state and federal corr ectional facilities. Washington, D.C.: U.S. Department of Justice. Bureau of Justice Statistics (2004a). Profile of jail inmates, 2002 Washington, D.C.: U.S. Department of Justice. Bureau of Justice Statistics (2004b). Drug use and dependence, state and federal prisoners, 2004, Washington, D.C.: U.S. Department of Justice. Burke, C. (2004). City of San Diego clean syringe exchange program: Final evaluation report San Diego Regional Planning Agency. Catania, J. A., Turner, H., Pierce, R. C., Golden, E., Stocking, C., Binson, D., & Mast, K. (1993). Response bias in surveys of AIDS-related sexual behavior. In D. G. Ostrow & R. C. Kessler (Eds.), Methodological issues on AIDS behavioral research. New York: Plenum. Centers for Disease Control and Prevention (CDC) (2004). HIV/AIDS surveillance report, 2003 Atlanta: U.S. Department of Health and Human Services. City of San Diego (2001). Clean syringe exchange pilot program. San Diego City Council. Chen, H, Stephens, R. C., Cochran, D. C., & Huff, H. K. (1997). Problems and solutions for estimating the prevalence of drug abuse among arrestees. Journal of Drug Issues, 27 689-701. Cleophas, T. J., Zwinderman, A. H., Cleophas, T. F., & Cleophas, E. P. (2009). Statistics applied to clinical trials. Springer. Cohen, J. (1992). A power primer. Psychological Bulletin, 112 (1), 155-159. Cone, E. J. (1997). New developments in biol ogical measures of drug prevalence. In L. Harrison and A. Hughes (Eds.), The Validity of Self-Reported Drug Use: Improving the Accuracy of Survey Estimates Rockville, MD: National Institute of Drug Abuse. Cook, S. W., & Selltitz, C. (1964). A mu ltiple-indicator approach to attitude measurement. Psychological Bulletin 62, 36-55.
193 Corman, H. and H. N. Mocan (2000). A time-se ries analysis of crime, deterrence, and drug abuse in New York City. American Economic Review 90 (3), 584-604. Crowne, D. and Marlow, D. (1964). The approval motive. New York: Wiley. Cunningham, J.K., Liu, L.-M., 2003. Impacts of federal ephedrine and pseudoephedrine regulations on methamphetamin e-related hospital admissions. Addiction, 98, 1229. Cunningham, J.K., Liu, L.-M., 2005. Impacts of federal precursor chemical regulations on methamphetamine arrests. Addiction, 100, 479. Decker, S. H., Pennell, S. & Caldwell, A. (1997). Illegal firearms: Access and use by arrestees. Research in Brief. National Institute of Justice: U.S. Department of Justice. Dembo, R.,Williams, L.,Wish, E. D., Berry, E., Getreu, A.,Washburn, M., & Schmeidler, J. (1990). Examination of the relationships among drug use, emotional/psychological problems, and crime among youths entering a juvenile detention center. The International Journal of the Addictions, 25, 1301. DeSimone, J. (2005). Needle exchange programs and drug injection users. Journal of Policy Analysis and Management, 24 (3), 559-577. Des Jarlais, D., C. (1998). Co mmentary: Validity of self-re ported data, scientific methods and drug policy. Drug and Alcohol Dependence 51 265 Drug Enforcement Agency (2007a). DEA staffing and budget. Available online at: http://www.usdoj.gov/dea/agency/staffing.htm. Drug Enforcement Agency (2007b). Organized crime drug enforcement task forces (OCDETF) Available online a t: http://www.usdoj.gov/dea/programs/ocdetf.htm. Drug Policy Alliance (2006). Proposition 36: Improving lives, delivering results. A review of the first four years of Cali fornias Substance Abuse and Crime Prevention Act Available online at: http://www.drugpolicy.org/docUploads/Prop36March2006.pdf. Drug Reform Coordination Network (2005). Marijuana law enforcement costs more than $7 billion a year and doe snt work says new report Available online at: http://stopthedrugwar.org/ch ronicle-old/379/report1.shtml. Edwards, A. L. (1953). The relationship betwee n the judged desirability of a trait and the probability that the trait will be endorsed. Journal of Applied Psychology 37, 90103.
194 Elliott, D. S., & Ageton, S.S. (1980). Rec onciling race and class differences in self-reported and official estimates of delinquency. American Sociological Review, 45(1), 95. Epstein, J., Klinkenberg, W. D., Wiley, D. & McKinley, L. (2001). Insuring sample equivalence across internet a nd paper-and-pencil assessments. Computers in Human Behavior, 17 (3), 339-346. European Medicines Agency (2000). Points to consider in sw itching between superiority and non-inferiority Committee for Proprietary Medi cinal Products: The European Agency for the Evaluation of Medicinal Products. Available at: http://www.ema.europa.eu/pdfs/human/ewp/048299en.pdf. Fagan, J. and K. L. Chin (1990). Violence as regulation of and social control in the distribution of crack. In: M. de la Rosa E. Y. Lambert, and B. Gropper (Eds.), Drugs and violence: Causes, correlates, and consequences NIDA Research Monograph No. 103, Rockville, MD: National Institute on Drug Abuse. Fendrich, M., Johnson, T. P. Sudman, S., Wislar, J. S., & Spiehler, V. (1999). Validity of drug use reporting in a high-risk community sample: A comparison of cocaine and heroin survey reports with hair tests. American Journal of Epidemiology, 149, 945962. Feucht, T. E. & Kyle, G. M. (1996). Methamphetamine use among adult arrestees: Findings from the Drug Use Forecasting (DUF) Program National Institute of Justice. U.S. Department of Justice. Franco, C. (2007). Methamphetamine: Legislation and issues in the 109th Congress. In: Gerald H. Toolaney. New Research on Methamphetamine Abuse New York: Nova Science Publisher. French, M. T. and Martin, R. F. (1996). The cost of drug abuse consequences: A summary of research findings. Journal of Substance Abuse Treatment, 13(6) 453-466. Gfroerer J, Eyerman J, and Chromy J, (Eds.) (2002) Redesigning an ongoing national household survey: Methodological issues DHHS Publication No. SMA 03. Rockville, MD: Substance Abuse and Ment al Health Services Administration, Office of Applied Studies. Goldstein, P. J. (1990). Crack and homicide in New York City: A conceptually based event analysis, Contemporary Drug Problems, 4, 651-687. Goldstein, P. J. (1987). Drug-related crime analysis and homicide A report to the National Institute of Justice, U.S. Department of Justice.
195 Golub, A. L., & Johnson, B. D. (1997). Cracks decline. Some surprises across U.S. cities. Research in Brief. National Institute of Justice. U.S. Department of Justice. Golub, A. L., & Johnson, B. D. (2001). The rise of marijuana as the drug of choice among youthful adult arrestees Research in Brief. Nati onal Institute of Justice. U.S. Department of Justice. Golub, A. L., & Johnson, B. D. (2005). The new heroin users among Manhattan arrestees: Variations by race/eth nicity and mode of consumption. Journal of Psychoactive Drugs 37, 5161. Grogger, J., & Willis, M. (2000). The emergence of crack cocaine and the rise in urban crime rates. The Review of Economics and Statistics 82, 519-529. Hammett, T. M., Harmon, P., & Rhodes, W. (2002). The burden of infectious diseases among inmates of and releasees from US correctional facilities, 1997. American Journal of Public Health 92, 1789. Harrell, A. V. (1985). Validation of self-report: the research re cord in self-report methods of estimating drug use and meeting current challenges to validity. In B. E. Rouse, & N. J. Kozol (Eds.), National institute on drug abuse research monograph, vol. Washington, DC: US Government Printing Office. Harris, R. J. (1997). Signifi cance tests have their place. Psychological Sciences, 8, 8-11. Harrison, L. D. (1992). Tre nds in illicit drug use in the USA; Conflicting results from national surveys. International Journal of Addictions, 27 817-847. Harrison, L. D. (1995). The validity of self-reported data on drug use. The Journal of Drug Issues, 25 91-111. Harrison, L. D., Martin, S. S., Enev, T, and Harrington, D. (2007). Comparing drug testing and self-report of drug use among youths and young adults in the general population Rockville, MD: DHHS Publication No. SMA 07-4249. Hartley, R., Maddan, S. & Spohn, C. (2007). Prosecu torial discretion: An examination of substantial assistance departures in Federal crack-cocaine and powder cocaine cases. Justice Quarterly, 24, 382. Hauck, W. W. & Anderson, S. (1986). A proposal for interpreting and reporting negative studies. Statistics in Medicine, 5, 203-209. Harder, V. S., & Chilcoat, H. D. (2007). Cocaine use and educational achievement: Understanding a changing association over the past 2 decades. American Journal of Public Health, 97 (10), 1790-1793.
196 Hersen, M., & Gross, A. M. (Eds.). (2008). Handbook of clinical psychology: Adults and children (2 volumes). Hoboken, NJ: John Wiley & Sons, Inc. Herz, D. C. (2000). Drugs in the heartland: Methamphe tamine use in rural Nebraska Research in Brief. National Institute of Justice. U.S. Department of Justice. Hindelang, M.J., Hirschi, T., & Weis, J.G. (1979). Correlates of delinquency: The illusion of discrepancy between self-report and official measures. American Sociological Review,44, 95. Hindelang, M.J., Hirschi, T., & Weis, J.G. (1981). Measuring delinquency. Beverly Hills: Sage Publications. Hindin, R., McCusker, J., Vickers-Lahti, M., Bigelow, C., Garfield, F., & Lewis, B. (1994). Radioimmunoassay of hair for determination of cocaine, heroin and marijuana exposure: Comparison with self-report. International Journal of the Addictions, 29 p. 771-789. Hora, P., Schma, W. G. & Rosenthal, J. T. A. (1998). Therapeutic jurisprudence and the drug treatment court movement: Revolutionizing the criminal justice systems response to drug abuse and crime, Notre Dame Law Review 101, 74 Hser, Y.-I., Maglione, M., & Boyle, K. (1999) Validity of self-report of drug use among STD patients, ER patients and arrestees. American Journal of Drug and Alcohol Abuse, 25, 81-91. Huizinga, D., & Elliott, D.S. (1986). Reasse ssing the reliability and validity of selfreport delinquent measures. Journal of Quantitative Criminology 2 (4), 293 327. Hunt, D., Kuck, S., & Truitt, L. (2006). Methamphetamine use: Lessons learned National Institute of Justice. U.S. Department of Justice. Hyman, H. (1944). Do they tell the truth. The Public Opinion Quarterly 8, 557-559. Inciardi, J. A., Martin, S. S., & Butzin, C. A. (2004). Five-year outcomes of therapeutic community treatment of drug-involved offenders after release from prison. Crime & Delinquency 50, 88-107. Jenkins, P., Earle-Richardson, G., Tucker Sl ingerman, D., and Ma y, J. (2002). Time Dependent Memory Decay. American Journal of Industrial Medicine, 41, 98-101. Johnston, L. D. and OMalley, P. M. (1997). The recanting of earlier reported drug use by young adults. In L. Harrison and A. Hughes (Eds.), Validity of Self-Reported Drug Use: Improving the Accuracy of Survey Estimates Rockville, MD: National Institute of Drug Abuse.
197 Johnston, L. D., O'Malley, P. M., & Bachman, J. G. (1997). National survey results on drug use from the Monitoring the Future study, 1975-1995. Volume II: College Students and Young Adults Rockville, M.D.: National Institute of Drug Abuse. Johnston, L. D., O'Malley, P. M., Bachman, J. G., & Schulenberg, J. E. (2007). Overall, illicit drug use by American t eens continues gradual decline in 2007 University of Michigan News Service: Ann Arbor, MI. [Online]. Available online: www.monitoringthefuture.org. Johnston, L. D., O'Malley, P. M., Bachman, J. G., & Schulenberg, J. E. (2008). Monitoring the Future national results on adolescent drug use: Overview of key findings, 2007 (NIH Publication No. 08-6418). Be thesda, MD: National Institute on Drug Abuse, 1-70. Johnson, E.O., & Schultz, L. (2005). Forward tele scoping bias in reported age of onset: an example from cigarette smoking. International Journal of Methods in Psychiatric Research 14 (3):119-129. Kalton, G. and Kasprzyk, D. (1986). The treatment of missing data. Survey Methodology, 12 1. Karberg, J. C. & James, D. C. (2005). Substance dependence, abuse, and treatment of jail inmates, 2002. U.S. Department of Justice Program s. Bureau of Justice Statistics Special Reports. Kelly, J. A, Stevenson L. Y., Hauth, A. C., Kalichman, S. C., Diaz, Y. E., Brasfield, T. L., Koob, J. J., & Morgan, M. G. (1992). Community AIDS/HIV reduction: The effects of endorsements by popular people in three cities. American Journal of Public Health, 82, 1483 1489. Kleiman, M. A. R. (2004). Flying blind on drug control policy. Issues in Science and Technology Summer 2004. Krantz, J., Ballard, J., & Scher, J. (1997). Co mparing the results of laboratory and WorldWide Web samples on the determinants of female attractiveness. Behavior Research Methods, Instruments, and Computers, 29(2), 264. Langan, P. A. & Levin. D. J. (2002). Recidivism of prisoners released in 1994 Special Report. Bureau of Justice Statistics. Leff, H. S., Wieman, D. A., McFarland, Bent son H., Morrissey, J. P., Rothbard, A., Shern, D. L., Wylie, A. M., Boothroyd, R. A., Stroup, T. S., & Allen, I. E. (2005). Assessment of medicaid managed behavior al health care for persons with serious mental illness. Psychiatric Services, 56, 1245-1253.
198 Le Henanff, A., Giraudeau, B., Baron, G., & Ravaud, R. (2006). Quality of reporting of noninferiority and equivalence randomized trials. JAMA, 295 1147 1151. Link, B. G., & Phelan, J. (1995). Social condit ions as fundamental causes of disease. Journal of Health and Social Behavior; Spec No 80. Magnusson, D. & Bergman, L. R. (Eds.) (1990). Data Quality in Longitudinal Research. Cambridge: Cambridge University Press. Mallender, J., Roberts, E., & Seddon, T. (2002). Evaluation of drug testing in the criminal justice system in three pilot areas. Home Office Research Findings No. 176, London: Home Office. Martin, S. S., Butzin, C. A., Saum, C. A., & Inciardi, J. A. (1999). Three-year outcomes of therapeutic community treatment for drug-involved offenders in Delaware: From prison to work release to aftercare. The Prison Journal, 79, 291-293. McBride, D. C., & Swartz, J. (1990). Drugs and violence. In Ralph Weisheit (Ed.), Drugs, Crime, and the Criminal Justice System Cincinnati: Anderson. Mieczkowski, T., Barzelay, D., Gropper, B ., & Wish, E. (1991). Concordance of three measures of cocaine use in an a rrestee population: Hair, urine and selfreport. Journal of Psychoactive Drugs, 23, 241-249. Mieczkowski, T., & Newel, R. (1997). Patterns of concordance between hair assays and urineanalysis for cocaine: Longitudinal analysis of probationers in Pinellas County, Florida. In L. Harrison and A. Hughes (Eds.), The Validity of SelfReported Drug Use: Improving the Accuracy of Survey Estimates Rockville, MD: National Institute of Drug Abuse. Moses, L. E. (1992). The reasoning of statistical inference. In D. C. Hoaglin & D. S. Moore (Eds.), Perspectives on contemporary statistics (pp. 107-122). Washington, DC: Mathematical Association of America. Murphy, S. A., Collins, L. M., & A.J. Rush (2007). Customizing treatment to the patient: Adaptive treatment strategies. Drug and Alcohol Dependence 88(2), S1-S72. Myrstol, B. and Langworthy, R. (2005). ADAM-Anchorage data: Are they representative? Working Paper Number 1. Just ice Center Working Papers: University of Alaska Anchorage. National Criminal Justice Association (1999). The rising methamphetamine crisis: An examination of state responses. Policy and Practice, 2(1), 1-12.
199 National Institute of Drug Abuse (1990). Drugs and violence: Caus es, correlates, and consequences. Research Monograph Series 103. Washington D.C.: National Institute of Health. National Institute of Drug Abuse (2006). Monitoring the Future National Survey Results on Drug Use, 1975-2007, Volume II: Colle ge Students and Adults Ages 19-45, 2007. Washington D.C.: National Institute of Health. National Institute of Justice (1990). Drug use forecasting DUF estimates of drug use applied to the UCR Washington D.C.: Nationa l Institute of Justice. National Institute of Justice (1993). Identifying and responding to new forms of drug abuse: Lessons learned from Crack and Ice. Washington D.C.: National Institute of Justice. National Institute of Justice (1994). Drug use forecasting Annual report on adult and juvenile arrestees Washington D.C.: National Institute of Justice. National Institute of Justice (1995). Drug use forecasting Annual report on adult and juvenile arrestees Washington D.C.: National Institute of Justice. National Institute of Justice (1997). Drug use forecasting Annual report on adult and juvenile arrestees Washington D.C.: National Institute of Justice. National Institute of Justice (1998). Drug use forecasting in 24 cities in the United States, 1987-1997 Washington, DC: U.S. Dept. of Jus tice, National Institute of Justice National Institute of Justice (1999). Drug use forecasting Annual report on adult and juvenile arrestees Washington D.C.: National Institute of Justice. National Institute of Justice (2000). Applying the New ADAM Method Washington, DC: U.S. Department of Justice. National Institute of Justice (2001). 2001 Arrestee Drug Abuse Monitoring Annual Report Washington, DC: U.S. Department of Justice. National Institute of Justice (2003). 2000 Arrestee Drug Abuse Monitoring Annual Report Washington, DC: U.S. Department of Justice. National Institute of Justice (2004). Arrestee Drug Abuse Monitoring (ADAM) Washington, DC: U.S. Department of Justice. Nurco, D. N. (1985). A discussion of validity. In B. A. Rouse, N. J. Kozel, & L G. Richards (Eds). Self-Report Methods of Estimating Drug Use Rockvillee, M.D.: National Institute on Drug Abuse.
200 Oberski, D. (2008). Self-selection versus non-response bias in the perceptions of mobility surveys. A comparison using multiple imputation The Hague: The Netherlands Institute for Social Research. Office of National Drug Control Policy (1996). Treatment protocol effectiveness study Washington, DC: Executive Office of th e President, Office of National Drug Control Policy. Office of National Drug Control Policy (2001). The economic costs of drug abuse in the United States 1992-1998. Washington, DC: Executive Office of the President. Office of National Drug Control Policy (2002). National drug control budget. Executive summary. Fiscal Year 2002. Executive Office of the President. Office of National Drug Control Policy (2003). Drug data summary. NCJ 191351. Executive Office of the President. Office of National Drug Control Policy (2007). National drug control strategy. FY 2008 Budget Summary NCJ 216432. Executive Office of the President. Office of National Drug Control Policy ( 2008). ADAM II. Annual Report. Executive Office of the President. Office of National Drug Control Policy (2009). National drug control budget. Executive summary. Fiscal Year 2010. Executive Office of the President. Office of Safe and Drug Free Schools (2006). Preliminary report to the secretary on the State grants program Available online: http://www.ed.gov/about/bdscomm/list/sdfscac/grantrpt1.html Parry, H. J., & Crossley, H. M. (1950). Va lidity of responses to survey questions. The Public Opinion Quarterly, 14, 61-80. Pasveer, K., & Ellard, J. (1998) The making of a personality inventory: help from the WWW. Behavior Research Methods, Instruments, and Computers, 30(2), 309 313. Peters, R.J. Jr, Yacoubian, G.S. Jr, Baumler, E.R., Ross, M. W., & Johnson, R. J. (2002). Heroin use among southern arrest ees: Regional findings from the Arrestee Drug Abuse Monitoring Program. Journal of Addictions & Offender Counseling 22, 50-60. Petersilia, J., Greenwood, P. W., & Lavin, M. (1978). Criminal careers of habitual felons Washington, DC: U.S. Government Printing Office.
201 Pocock, S. (2003). The pros and cons of noninferiroty trials. Fundamental and Clinical Pharmacology, 17, 483-490. Prendergast, M. L., & Wexler, H. K. (2004) Correctional substance abuse treatment Programs in California: A historical perspective. The Prison Journal, 84, 8-35. Presser, (1990). Can context ch anges reduce vote overreporting? Public Opinion Quarterly, 54 586-593. Resignato, A. J. (2000). Viol ent crime: A function of drug use or drug enforcement. Applied Economics 32, 681-688. Reuter, P. (2006). What drug policies co st. Estimating government drug policy expenditures. Addiction 101, 315-322. Richter, L., & Johnson, P. B. (2001). Current methods of assessing substance use: a review of strengths, problems, and developments. Journal of Drug Issues 46, 34. Riley, K. J. (1997). Crack, powder cocaine, and heroin: Drug purchase and use patterns in six U.S. cities National Institute of Justice. U.S. Department of Justice. Riley, K.J., Lu, N.T., & Taylor, T.G. ( 2000). Drug screening: A comparison of urinalysis resu lts from two independent laboratories Journal of Drug Issues. 30 (1), 173 185. Roberts, J., Mulvey, E. P., Horney, J., Lewis, J., and Arter, M. L. (2005). A test of two methods of recall for violent events. Journal of Quantitative Criminology, 21, 175-193. Rogers, J. L., Howard, K. I., & Vessey, J. T. (1993). Using significance tests to evaluate equivalence between two experimental groups. Psychological Bulletin, 113, 553 565. Roth, J. A. (1994). Psychoactive substances and violence Research in Brief. U.S. Department of Justice: National Institute of Justice. Sharp, H., & Feldt, A. (1959). Some factor s in a probability sample survey of a metropolitan community. American Sociological Review 24, 650-661. Sloan III, J. J., Bodapati, M. R., & Tucker T. A. (2004). Respondent misreporting of drug use in self-reports. Social desirability and other correlates. The Journal of Drug Issues 34, 269-292. Smith, C., & Nutbeam, D. (1990). Assessing non-response bias: A case study from 1985 Welsh Heart Health Survey. Health Education Research 5, 381-386.
202 Stegner, B. L., Bostrom, A. G., & Greenfield, T. K. (1996). Equivalence testing for use in psychosocial and services research : An introduction with examples. Evaluation and Program Planning, 19 (3), 193-198. Stone, A. A., Turkkan, J. S., Bachrach, C. A., Jobe, J. B., Kurtsman, H. S., & Cain, V. S. (2000). The science of self-report: Imp lications for research and practice Mahwah, NJ: Lawrence Erlbaum. Substance Abuse and Mental Hea lth Service Administration (2002). Results from the 2001 National Survey on Drug Use and Health: National Findings. Office of Applied Studies, Rockville, MD. Substance Abuse and Mental Health Service Administration (2002a). Emergency Department Trends From the Drug Abuse Warning Network, Final Estimates 1994-2001. Office of Applied Studie s, Rockville, MD. Substance Abuse and Mental Hea lth Service Administration (2002b). Drug Abuse Warning Network: Development of a New Design. Methodology Report: Office of Applied Studies. U.S. Department of Health and Human Services. Substance Abuse and Mental Hea lth Service Administration (2003). Drug and alcohol information systems: The DASIS report. Office of Applied Studies. U.S. Department of Health and Human Services. Substance Abuse and Mental Health Service Administration (2004). Meth Abuse Increases in the Midwest. Big Increase s Seen on the East coast also. News Release. Available online at: http://alcoholism.about.com /od/meth/a/blsam040822.htm Substance Abuse and Mental Health Services Administration. (2005). Results from the 2005 National Survey on Drug Use and Health. Department of Health and Human Services. Substance Abuse and Mental Health Services Administration. (2006). National Survey of Drug Use and Health: Summary of Methodological Studies. Department of Health and Human Services Substance Abuse and Mental Hea lth Service Administration (2008). Results from the 2007 National Survey on Drug Use and Health: National Findings. Office of Applied Studies, Rockville, MD. Tassipoulos, K., Bernstein, J., Heeren, T., Levenson, S., Hingson, R., & Bernstein, E. (2004). Hair testing and self-report of cocaine use by heroin users. Addiction 99, 590-597.
203 Tryon, W. W. (2001). Evaluating st atistical difference, equiva lence, and indeterminacy using inferential confidence intervals: An integrated alte rnative method of conducting null hypothesis statistical tests. Psychological Methods, 6, 371. Tryon, W. W. and Lewis, C. (2009). Evaluati ng Independent Proportions for Statistical Difference, Equivalence, Indeterminacy, a nd Trivial Difference Using Inferential Confidence Intervals, Journal of Educational and Be havioral Statistics, 34 (2), 171-189. United States General Accounting Office (1993). Drug use measurement: strength, limitations, and recommenda tions for improvement Report to the Chairman, Committee on Government Operations, House of Representatives. General Accounting Office, Washington DC. Office of Safe and Drug Free Schools (2006). Offices U.S. Department of Education. Available online at: http://www.ed.gov/ about/offices/list/o sdfs/index.html. Webb, V. J., & Delone, M. A. (1996). Drug use among a misdemeanant population. Crime, Law and Social Change 24(3), p. 241-255. Westlake, W., J. (1981). Bioequivalence testin gA need to rethink (Reader reaction response). Biometrics, 37, 591-593. Wiens, B., L. (2001). Something for nothing in noninferiority/superi ority testing: a caution. Drug Info Journal, 35, 241. Willis, J. J., Mastrofski, S. D., and Weisburg, D. (2003). Compstat in practice: An indepth analysis of three cities Police Foundation. Wish, E., & Gropper, B. (1990). Drug Testing by the Criminal Justice System. In M. Tonry, & J. Wilson, Drugs and Crime, (Eds.). Vol. 13 of Crime and Justice: A Review of Research. Chicago: University of Chicago Press. Yacoubian, G. S. (2003a). Correlates of be nzodiazepine use among a sample of arrestees surveyed through the Arrestee Drug Abuse Monitoring (ADAM) program. Substance Use & Misuse 38, 127-139. Yacoubian, G. S. (2003b). Does the cale ndaring method enhance drug use reporting among Portland arrestees? Journal of Subst ance Use, 8 (1), 27-32. Yacoubian, G. S. (2004). The sins of ADAM : Toward a new national criminal justice drug use surveillance system. International Journal of Drug Testing 3, 1-32. Yacoubian, G. S., Urbach, B. J., Larsen, K. L ., Johnson, R. J., & Peters Jr., R. J. (2000). A comparison of drug use between pr ostitutes and other female arrestees. Journal of Alcohol and Drug Education, 46, 12-25.
204 Yacoubian, G. S., Peters, R. J., Urbach, B. J., & Johnson, R.J. (2002). Comparing drug use between welfare-rece iving arrestees and non-welf are-receiving arrestees. Journal of Drug Education 32, 139-147. Yang, Y. M. (2004). Survey errors and survey costs: Experience from surveys of arrestees. American Statistical Associat ion Section on Survey Research Methods, 46564659. Zerbe, W. J., & Paulhus, D. L. (1987). So cially desirable resp onding in organization behavior: A reconception. Academy of Management Review 12, 250-264.
206 Appendix A Demographic Profile by Site Table A.1.: Dallas, TX DUF 97/98 ADAM 00/01 Variable N % N % Race Black 903 58.4 394 49.1 White 450 41.4 280 34.9 Hispanic 152 9.8 119 14.8 Other 14 0.9 9 1.1 Not obtained 28 1.8 Employment Full time 892 57.7 473 59.0 Part time 269 17.4 81 10.1 Unemployed 71 4.6 189 23.6 Other 305 19.7 58 7.2 Not obtained 197 12.7 1 0.1 Highschool Graduate Yes 904 58.4 576 71.8 Offense Category Violent Offense 448 29.0 156 19.5 Property Offense 675 43.6 160 20.0 Drug Offense 468 16.0 201 25.3 Other Offense 234 15.1 283 35.2 Age in years (mean) 30 30 N 1,547 802
207 Table A.2.: Denver, CO DUF 97/98 ADAM 00/01 Variable N % N % Race Black 662 34.7 350 26.5 White 545 28.6 364 27.6 Hispanic 645 33.8 547 41.5 Other 42 2.2 57 4.3 Not obtained 14 0.7 1 0.1 Employment Full time 979 51.3 667 50.6 Part time 392 20.5 181 13.7 Unemployed 141 7.4 355 26.9 Other 382 20.0 116 8.8 Not obtained 14 0.7 Highschool Graduate Yes 1,028 53.9 858 65.0 Offense Category Violent Offense 524 27.5 313 23.7 Property Offense 361 18.9 247 18.7 Drug Offense 565 23.9 274 20.8 Not obtained 2 0.1 485 36.8 Age in years (mean) 32 32 N 1,908 1,319
208 Table A.3.: Indianapolis, IN DUF 97/98 ADAM 00/01 Variable N % N % Race Black 884 57.2 771 56.6 White 586 37.9 577 42.4 Hispanic 64 4.1 11 0.8 Other 9 0.6 3 0.2 Not obtained 2 0.1 Employment Full time 846 54.8 771 56.6 Part time 280 18.1 182 13.4 Unemployed 116 7.5 281 20.6 Other 277 17.9 120 8.8 Not obtained 26 1.7 Highschool Graduate Yes 824 53.3 848 62.3 Offense Category Violent Offense 404 26.1 359 26.4 Property Offense 372 24.1 291 21.4 Drug Offense 326 21.1 339 24.9 Other 443 28.7 17 1.2 Age (mean) 31 31 N 1,545 1,362
209 Table A.4.: Miami, FL DUF 97/98 ADAM 00/01 Variable N % N % Race Black 588 46.2 283 52.9 White 200 15.7 231 43.2 Hispanic 480 37.7 21 3.9 Other Not obtained 4 0.3 Employment Full time 627 49.3 296 55.3 Part time 257 20.2 65 12.1 Unemployed 118 9.3 134 25.0 Other 263 20.7 40 7.5 Not obtained 7 0.6 Highschool Graduate Yes 512 40.3 359 67.1 Offense Category Violent Offense 446 35.1 119 22.2 Property Offense 296 23.3 129 24.1 Drug Offense 404 31.8 123 23.0 Other 126 9.9 164 30.7 Age in years (mean) 33 33 N 1,272 535
210 Table A.5.: New Orleans, LA DUF 97/98 ADAM 00/01 Variable N % N % Race Black 1,747 87.0 1,070 86.6 White 216 11.0 158 12.9 Hispanic 15 0.8 2 0.2 Other 12 0.6 4 0.3 Not obtained 12 0.6 2 0.2 Employment Full time 905 46.2 556 45.0 Part time 454 23.2 204 16.5 Unemployed 73 3.7 321 26.0 Other 503 25.7 155 12.5 Not obtained 24 1.2 Highschool Graduate Yes 847 43.2 670 54.2 Offense Category Violent Offense 628 32.1 176 14.2 Property Offense 728 37.2 214 17.3 Drug Offense 187 9.5 256 20.7 Age in years (mean) 30 30 N 1,959 1,236
211 Table A.6.: Phoenix, AZ DUF 97/98 ADAM 00/01 Variable N % N % Race Black 221 13.7 340 11.9 White 790 49.0 1,585 55.6 Hispanic 520 32.3 730 25.6 Other 77 4.8 188 6.6 Not obtained 3 0.2 7 0.2 Employment Full time 953 59.2 1,597 56.0 Part time 210 13.0 313 11.0 Unemployed 106 6.6 674 23.6 Other 339 21.0 264 9.3 Not obtained 3 0.2 2 0.1 Highschool Graduate Yes 956 59.3 1,914 67.2 Offense Category Violent Offense 271 16.8 607 21.3 Property Offense 414 25.7 625 21.9 Drug Offense 317 19.7 703 24.7 Other 609 37.8 914 32.1 Age in years (mean) 31 31 N 1,611 2,850
212 Table A.7.: Portland, OR DUF 97/98 ADAM 00/01 Variable N % N % Race Black 421 30.0 318 24.6 White 865 61.7 837 64.6 Hispanic 78 5.6 95 7.3 Other 35 2.5 41 3.2 Not obtained 2 0.1 4 0.3 Employment Full time 478 34.1 480 37.1 Part time 312 22.3 152 11.7 Unemployed 80 5.7 479 37.0 Other 527 37.6 184 14.2 Not obtained 4 0.3 Highschool Graduate Yes 828 59.1 963 74.4 Offense Category Violent Offense 217 15.5 269 20.8 Property Offense 225 16.1 219 16.9 Drug Offense 407 29.1 448 34.6 Other 11 0.8 Age in years (mean) 33 33 N 1,401 1,295
213 Table A.8.: San Antonio DUF 97/98 ADAM 00/01 Variable N % N % Race Black 203 11.3 142 12.7 White 606 32.9 396 35.3 Hispanic 995 54.1 577 51.4 Other 6 0.3 4 0.4 Not obtained 25 1.4 3 0.3 Employment Full time 1,043 56.7 691 61.6 Part time 306 16.6 119 10.6 Unemployed 216 11.7 207 18.4 Other 272 14.8 103 9.2 Not obtained 3 0.2 2 0.2 Highschool Graduate Yes 874 47.5 685 61.1 Offense Category Violent Offense 449 24.4 145 12.9 Property Offense 453 24.6 150 13.4 Drug Offense 281 15.3 198 17.6 Other 657 35.7 627 55.9 Not obtained 2 0.2 Age in years (mean) 29 29 N 1,840 1,122
214 Table A.9.: San Jose DUF 97/98 ADAM 00/01 Variable N % N % Race Black 147 11.1 149 11.9 White 424 32.1 379 30.4 Hispanic 568 43.1 574 46.0 Other 170 12.9 148 11.4 Not obtained 10 0.8 144 11.5 Employment Full time 739 56.0 743 59.5 Part time 206 15.6 121 9.7 Unemployed 158 12.0 317 24.5 Other 214 16.2 111 8.9 Not obtained 2 0.2 Highschool Graduate Yes 657 49.8 969 77.6 Offense Category Violent Offense 551 41.8 383 30.7 Property Offense 338 25.6 197 15.8 Drug Offense 168 12.7 427 34.2 Other 338 25.6 239 19.2 Not obtained 8 0.6 2 0.2 Age in years (mean) 31 32 N 1,319 1,248
215 Appendix B Drug Use Freque ncies by Year and Site Table B.1.: Dallas, TX DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 42943.824342.917733.8 8530.6 Cocaine 31131.716128.414928.4 6824.5 Opiates 424.3122.1214.0 134.7 Self-Report Drug Use Within 72 Hours Marijuana 32132.818432.512123.1 6121.9 Cocaine 626.3315.5346.5 134.7 Crack 13313.65810.26011.5 3713.3 Heroin 242.491.6122.3 41.4 PCP 80.850.951.0 00.0 Amphetamines 242.4132.310.2 10.4 Barbiturates 60.630.510.2 00.0 Ever Used Drug Marijuana 79481.041873.736970.4 18064.7 Cocaine 33434.117631.017232.8 8831.7 Crack 27127.714625.714427.5 6824.5 Heroin 979.95910.4458.6 3010.8 N 980 567 524 278
216 Table B.2.: Denver, CO DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 39441.439441.225441.1 27939.8 Cocaine 38340.237939.621034.0 23032.8 Opiates 343.6384.0233.7 385.4 Self-Report Drug Use Within 72 Hours Marijuana 25927.229631.017528.3 22532.1 Cocaine 626.5788.2325.2 446.3 Crack 16116.912913.57311.8 9213.1 Heroin 262.7323.3142.3 202.9 PCP 00.000.000.0 00.0 Amphetamines 121.3121.350.8 60.9 Barbiturates 40.440.450.8 10.1 Ever Used Drug Marijuana 81485.581885.647877.3 52675.0 Cocaine 46548.845447.525841.7 32346.1 Crack 38240.136638.321835.3 25235.9 Heroin 15115.914715.49515.4 10515.0 N 952 956 618 701
217 Table B.3.: Indianapolis DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 41143.927344.930247.9 36149.3 Cocaine 29531.520734.018729.7 23632.2 Opiates 283.0111.8233.7 466.3 Self-Report Drug Use Within 72 Hours Marijuana 32935.119732.417527.8 23632.2 Cocaine 293.1213.5172.7 243.3 Crack 11111.86911.3548.6 7710.5 Heroin 161.730.540.6 141.9 PCP 10.110.210.2 10.1 Amphetamines 131.440.720.3 10.1 Barbiturates 161.7152.520.3 10.1 Ever Used Drug Marijuana 74779.749180.850780.5 59781.6 Cocaine 32534.721234.919831.4 23532.1 Crack 29931.918931.116225.7 23432.0 Heroin 11912.7548.9396.2 517.0 N 937 608 630 732
218 Table B.4..: Miami DUF ADAM 1997 1998 2000 Variable N % N % N % Urine Test Marijuana 27031.612229.218935.3 Cocaine 38845.419847.424044.9 Opiates 182.1102.4254.7 Self-Report Drug Use Within 72 Hours Marijuana 19222.59222.011822.1 Cocaine 10011.74410.5529.7 Crack 12414.56515.66512.1 Heroin 141.651.2193.6 PCP 10.100.010.2 Amphetamines 00.000.010.2 Barbiturates 50.630.700.0 Ever Used Drug Marijuana 53863.026763.932260.2 Cocaine 32938.515938.019837.0 Crack 21925.610825.811721.9 Heroin 708.2276.5489.0 N 854 418 535
219 Table B.5.: New Orleans DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 38138.336838.128046.4 28144.5 Cocaine 45645.944045.620533.9 22735.9 Opiates 10610.712613.19215.2 9515.0 Self-Report Drug Use Within 72 Hours Marijuana 31832.033134.319432.1 23837.7 Cocaine 989.9687.0264.3 457.1 Crack 16516.617418.06610.9 8413.3 Heroin 848.59710.16711.1 7111.2 PCP 10.130.300.0 00.0 Amphetamines 30.330.310.2 10.2 Barbiturates 70.7101.010.2 10.2 Ever Used Drug Marijuana 76276.775277.947578.6 46273.1 Cocaine 33233.433734.916727.6 17828.2 Crack 31031.230231.315425.5 16225.6 Heroin 19019.121722.513221.9 12219.3 N 994 965 604 632
220 Table B.6.: Phoenix DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 29930.420232.242333.4 60137.9 Cocaine 31832.319531.137829.9 40825.8 Opiates 929.4365.7766.0 1016.4 Self-Report Drug Use Within 72 Hours Marijuana 21121.514322.827621.8 46829.5 Cocaine 656.6416.5624.9 805.1 Crack 15015.38613.717513.8 18411.6 Heroin 808.1386.1624.9 674.2 PCP 30.310.270.6 20.1 Amphetamines 171.7182.9221.7 191.2 Barbiturates 121.250.860.5 40.3 Ever Used Drug Marijuana 81382.749879.397977.3 132083.3 Cocaine 49450.330949.261948.9 83052.4 Crack 42843.524238.550439.8 62239.3 Heroin 23123.512219.422918.1 26917.0 N 983 628 1266 1584
221 Table B.7.: Portland DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 24638.128037.021235.3 24635.4 Cocaine 23836.922029.113021.6 17625.4 Opiates 8913.811815.68013.3 679.7 Self-Report Drug Use Within 72 Hours Marijuana 16926.217823.511819.6 20229.1 Cocaine 599.1526.9244.0 415.9 Crack 10416.19212.2396.5 8912.8 Heroin 7311.38110.7416.8 436.2 PCP 00.020.310.2 00.0 Amphetamines 284.3101.371.2 71.0 Barbiturates 50.850.710.2 10.1 Ever Used Drug Marijuana 56287.166387.751485.5 57582.9 Cocaine 34854.041054.227846.3 32446.7 Crack 30146.732342.723539.1 29742.8 Heroin 18428.525133.215025.0 16423.6 N 645 756 601 694
222 Table B.8.: San Antonio DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 31934.337441.117734.0 23238.5 Cocaine 24426.224527.012423.8 16026.6 Opiates 9610.3879.6479.0 559.1 Self-Report Drug Use Within 72 Hours Marijuana 22524.223525.911822.7 17729.4 Cocaine 838.9808.8397.5 569.3 Crack 303.2222.4152.9 223.7 Heroin 525.6424.6305.8 355.8 PCP 10.100.010.2 10.2 Amphetamines 101.1111.210.2 40.7 Barbiturates 80.970.820.4 10.2 Ever Used Drug Marijuana 58863.258764.633664.6 44974.6 Cocaine 31633.927930.715830.4 24140.0 Crack 12012.910511.66813.1 9816.3 Heroin 12813.712213.46412.3 8714.5 N 931 909 520 602
223 Table B.9.: San Jose DUF ADAM 1997 1998 2000 2001 Variable N % N % N % N % Urine Test Marijuana 25628.910724.716731.7 27137.5 Cocaine 12013.6358.17113.5 7910.9 Opiates 495.5194.4244.6 202.8 Self-Report Drug Use Within 72 Hours Marijuana 17119.37818.011822.4 21029.1 Cocaine 394.4143.2122.3 283.9 Crack 404.551.2234.4 294.0 Heroin 293.381.2101.9 121.7 PCP 50.640.940.8 162.2 Amphetamines 3184.108.40.206 131.8 Barbiturates 40.510.200.0 40.6 Ever Used Drug Marijuana 59967.728565.735467.3 53774.4 Cocaine 33237.516838.720839.5 34547.8 Crack 19822.47517.313124.9 21329.5 Heroin 9610.84510.35510.5 9212.7 N 885 434 526 722
About the Author Janine Kremling graduated from the University of Leipzig in Germany with a Masters in Sociology in 2001. She received he r Masters in Criminology in 2004 from the University of South Florida. Ms. Kremlings research interests are predominantly focused on capital punishment, racial discrimination, and drug use and abuse. She started teaching while in the PhD program at the University of South Florida. Ms. Kremling was President of the Crimi nology Graduate Student Organization at the University of South Florida for two years. Additionally, served as a Justice on the Student Government Supreme Court for three years. From 2006 until 2008, Ms. Kremling was a research assistant at the Florid a Mental Health Institu te. In that capacity, she co-authored several publica tions. Ms. Kremling is currently an Assistant Professor at California State University at San Bernardino.