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 a22 u 4500
controlfield tag 008 c20019999azu 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E11-00201
Educational policy analysis archives.
n Vol. 9, no. 4 (February 08, 2001).
Tempe, Ariz. :
b Arizona State University ;
Tampa, Fla. :
University of South Florida.
c February 08, 2001
Factors influencing GED and diploma attainment of high school dropouts / Jeffrey C. Wayman.
Arizona State University.
University of South Florida.
t Education Policy Analysis Archives (EPAA)
xml version 1.0 encoding UTF-8 standalone no
mods:mods xmlns:mods http:www.loc.govmodsv3 xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govmodsv3mods-3-1.xsd
mods:relatedItem type host
mods:identifier issn 1068-2341mods:part
mods:detail volume mods:number 9issue 4series Year mods:caption 20012001Month February2Day 88mods:originInfo mods:dateIssued iso8601 2001-02-08
1 of 19 Education Policy Analysis Archives Volume 9 Number 4February 8, 2001ISSN 1068-2341 A peer-reviewed scholarly journal Editor: Gene V Glass, College of Education Arizona State University Copyright 2001, the EDUCATION POLICY ANALYSIS ARCHIVES. Permission is hereby granted to copy any article if EPAA is credited and copies are not sold. Articles appearing in EPAA are abstracted in the Current Index to Journals in Education by the ERIC Clearinghouse on Assessment and Evaluation and are permanently archived in Resources in Education .Factors Influencing GED and Diploma Attainment of High School Dropouts Jeffrey C. Wayman Colorado State UniversityAbstract This study examined correlates of degree at tainment in high school dropouts. Participants were high school dropouts of Mexican American or non-Latino white descent who had no degree, a hi gh school degree, or a GED certificate. This study was unique in that it accounted for sample bias of missing data through the use of multiple im putation, it considered students who had dropped out as early as 7th grade, and it was able to include variables found significant in previous res earch on returning dropouts. Logistic regression analyses identified a parsimonious set of factors which distinguished dropouts who held degre es (diploma or GED) from those who did not. Similar analyses were performed to distinguish participants who had attained diplomas from those who had attained GEDs. It was estimated that 59.2% of dropo uts return to obtain high school credentials. School capability, age at dropout, and socio-economic status significantly predicted degre e attainment. Presence of children, higher school capability and socio-economic status were associated with GED attainment, while later gr ade at dropout was
2 of 19associated with diploma attainment. These relations hips did not vary by ethnicity, although degree attainment was less like ly for Mexican American dropouts. The study concludes that droppin g out is not the end of a student's education, and more research should be directed toward returning dropouts. Further, the focus of such rese arch should be expanded to include a more positive and broader ran ge of correlates.Introduction Dropping out of high school is a well-docum ented social problem, and often presents daunting circumstances for adolescents. Dr opping out is often associated with delinquency, substance use, and low school achievem ent (Chavez, Oetting, & Swaim, 1994; Ekstrom, Goertz, Pollack, & Rock, 1986; Ellio tt, Huizinga, & Ageton, 1985). Further, people without high school degrees often e xperience lower wages and higher unemployment, and more dependency on welfare and ot her social services (Catterall, 1987; Rumberger, 1987). Research also shows that dropping out of hi gh school does not have to be, and is not necessarily, a permanent condition. Estimates o f the percentage of dropouts who eventually attain either high school diplomas or Ge neral Educational Development certifications (GEDs) have been as high as 44% (Kol stad & Kaufman, 1989). Thus, study of the correlates of degree attainment in dro pouts could be an effective tool in reducing the dropout rate, but unfortunately, few s tudies have been conducted in this area. Balancing the welldeveloped research on dro pout correlates with a research base of return correlates not only provides information on why dropouts gain degrees, but also provides a different perspective from which to augment dropout prevention efforts. Dropouts Who Return to School Settings Studies of returning dropouts have examined either dropouts who return to school (Borus & Carpenter, 1983; Chuang, 1997) or dropouts who obtain high school degrees or GEDs (Kaufman, 1988; Kolstad & Owings, 1986; Kol stad & Kaufman, 1989). Studies of this type have compared factors present in returning dropouts to a Â“typical dropout profileÂ”. From the vast amount of dropout l iterature, these studies have been able to identify factors associated with dropping o ut and have analyzed variables identified in this profile, hypothesizing that thos e dropouts who do not fit the profile are more likely to return to high school. This body of research is not yet sufficient ly developed to identify a complete picture of why dropouts return to school settings, although some factors appear to be fairly robust. For instance, achievement test score s were found in all studies reviewed here (except Borus & Carpenter (1983)) to be positi vely related to return for more education. Early dropouts are less likely to return as shown by all the studies except Kaufman (1988), which did not include this variable Nonetheless, the sparsity of studies on re turning dropouts have left many questions as to other variables affecting return. Ethnic effe cts are an inconsistent mix in these studies, and other factors, such as socio-economic status, are significant in some studies and not in others. Further, questions remain as to the effects of sampling on significant relationships identified Â– none of these studies we re able to consider dropouts who left school before 10th grade, and none were able to est imate effects due to inability to
3 of 19longitudinally follow each participant in the sampl e. Previous research has laid the foundation for knowledge regarding degree attainment in high school dropouts. However, such r esearch should be extended and clarified. The next logical step is a study which c an pull together significant factors found in previous studies and present estimates whi ch infer to the entire population of dropouts. The present study will address these issu es. The Present Study The present study examines Mexican American and non-Latino white dropouts who have gained high school diplomas, GEDs, or neit her, identifying factors which are associated with attainment of high school credentia ls. In doing so, this study will address several important problems left unsolved by previou s studies on returning dropouts. First, the present study accounts for bias introduced by dropouts who did not respond to the second wave of data collection. Long itudinal dropout studies naturally suffer from an inability to resurvey each and every dropout. However, each of the reviewed studies conducted analyses on only those d ropouts who were successfully followed up. Such treatment of missing dropouts ass umes that the dropouts who remained in the study are similar to the ones who d id not, an assumption which leaves the study vulnerable to sample bias. The present st udy, through the use of multiple imputation, accounts for bias caused by missing dat a. Second, previous studies were limited to pa rticipants who dropped out in tenth grade or later. Although against the law in many st ates, the truth is that many students leave school before age sixteen. The present study is able to consider students who dropped out earlier than tenth grade Â– some as earl y as seventh grade. Inclusion of these students, along with the estimation of missing data described above enables the present study to estimate return correlates for the full dr opout population. Third is the breadth of variables studied in this work. Previous studies independently drew upon factors known to be associa ted with dropping out and did not purposely examine variables shown to be significant in previous studies of returning dropouts. Therefore, it is not clear whether identi fied significances are due to omission of other important factors. To truly assess the sig nificance of factors associated with returning dropouts, these factors should be conside red in tandem. The present study addresses this need, as all variables considered we re chosen based on their significance in previous return research. Fourth, only Kolstad and Kaufman (1989) co nsidered diploma attainment and GED attainment separately. The present study will a lso discern differences between students with no degree, students with diplomas, an d students with GEDs.Method The data for this study were gathered as pa rt of a longitudinal project designed to study substance use and other correlates of high sc hool dropout among Mexican American and non-Latino white dropouts. The sample for this study consisted of Mexican American and non-Latino white adolescent dr opouts from three communities in the southwestern United States: a city with 400, 000 people, a mid-sized town with 90,000 people, and a small town with 30,000 people. Dropouts were defined as students in grades 7 Â– 12 who had not attended school for at least 30 days, had not transferred to another school, were not being home-schooled, and h ad not contacted the school system about re-admission. This definition is more stringe nt than that recommended by Morrow
4 of 19(1986), whose standard definition of a dropout call s for a period of unexcused absence from school of two weeks or more. The adoption of a period of absence of one month or longer provides a sufficient period of time to ensu re that youth are, in fact, high school dropouts. Potential participants were adolescents fro m dropout lists provided by school personnel in the aforementioned communities. Once t hey were identified and contacted, refusal rates were low (4 Â– 6%), so the resulting s ample is a random sample from the population of dropouts from these three communities Results from this study will be inferred to the population of Mexican American and non-Latino white dropouts in the United States. Although the sampling frame is limit ed geographically, previous results published from this data set have been comparable t o other studies of high school dropouts (e.g., Chavez, Oetting & Swaim, 1994; Chav ez, Deffenbacher, & Wayman, 1996). Therefore, inferring to this population from the present sample is appropriate. Measures All survey items used in this study were em bedded in a larger survey which took approximately one and a half hours to complete. Nea rly all surveys were completed in English, with less than 1% completed in Spanish. Dependent variable. Graduation from high school, possession of a GED, or no degree attainment were based on self-report measure s. Demographic information. Ethnicity was determined from school records and w as double-checked by field workers with the participan t. Gender and socio-economic status (SES) were based on self-reports from a demographic section of the initial survey. SES was a composite measure of the following items: edu cation of mother, education of father (possible responses of 6 th Â– 12 th grade, 1 Â– 4 years of college, or 5 or more years of college were coded as 6 Â– 17), Â“do your parents have good jobsÂ” (possible responses Â“they do not workÂ”, Â“poorÂ”, Â“not too goodÂ”, Â“goodÂ”, or Â“very goodÂ”), Â“what is your parents' incomeÂ” (possible responses were Â“very low Â”, Â“lowÂ”, Â“averageÂ”, Â“highÂ”, or Â“very highÂ”) and Â“does your family have enough mone y to buy the things you wantÂ” (possible responses Â“almost neverÂ”, Â“some of the ti meÂ”, Â“yes, most of the timeÂ”, or Â“yes, all of the timeÂ”). Since these items were not unifo rm in range of possible answers, responses were standardized before being summed to create the composite. The Cronbach alpha reliability of this scale was .65. Independent variables. Achievement test scores, age at dropout, grade at dropout, and grade point average were obtained from high sch ool records. Achievement test scores were used as a proxy for ability (or Â“school capabilityÂ”), which was measured by averaging mathematics, reading and vocabulary score s (Kaufman, 1988) for each participant. Data were collected on achievement tes ts administered at many times during the participant's school career, but due to inconsi stent record keeping, students transferring from districts using different procedu res, etc., neither the time frame nor the quantity of test scores was uniform across particip ants. Thus, the highest available mathematics, reading and vocabulary scores were use d. This not only provided consistency, but reduced noise in the test scores a s measures of school capability Â– few students would attain a test score which was a high er representation of their true capability. Whether the participant had or was expectin g children was based on selfreports from the initial survey, as was teacher caring. To assess a participant's feeling of teacher caring, an item asking Â“how much did teachers care about you during this last yearÂ” was included on the survey, with possible responses of Â“not at allÂ”, Â“not muchÂ”, Â“someÂ”, and
5 of 19Â“a lotÂ”. Marriage was not used in this analysis bec ause only three of the participants reported being married at the time of dropout.Procedure For the first wave of data collection, drop outs were chosen randomly from monthly lists of dropouts, provided by the school district. Field workers, employed by the district and fluent in English and Spanish, first contacted potential participants. After the project was described, potential participants were asked if they wished to be involved. If they expressed interest and were over 18, they completed consent forms. If they were under 18, parents were contacted, the project was fully e xplained, and written parental consent was obtained. Those who refused were replaced in th e sampling frame by another randomly sampled dropout. Following informed consent, arrangements we re then made for an individual administration of the survey. The survey was comple ted at school or at another public building such as a library, and participants were g iven as much time as needed to complete the survey. The survey administrator gave participants the survey, answered general questions and helped participants with read ing problems, but did not see participant responses. When the survey was complete the participant put it in a large envelope and sealed it personally. Based on the par ticipant's choice, the survey was mailed to the research office either by the survey administrator or was taken immediately to a mailbox by the participant and survey administ rator. These steps assured confidentiality; at no time was an unsealed, comple ted survey out of the participant's sight. Participants received $25 for completion of the survey. Accuracy and reliability of data were assur ed as surveys were subjected to 40 checks for inconsistency or exaggeration (e.g., end orsing a fake drug, claiming daily use of three or four drugs). Only 2% of initial surveys failed either review and were not replaced. Four years after the first assessment, foll ow-up of dropouts 18 or older began, with an average time to completion of the follow-up surv ey of 4.29 years. Follow-up contact was first attempted through the address given at th e first assessment. If this failed, staff contacted three people (e.g., parents, relatives, g ood friends) whom the participant indicated at the time of informed consent would alw ays know where the participant lived. If these efforts failed, public records such as phone books, motor vehicle records, etc., were checked to locate an address. A total of 519 (49%) of the 1071 original participants were successfully followed up. Once th e individual was contacted and gave his/her consent, survey administration was parallel to the first administration. Data Analysis Multiple imputation. Missing data presented a potential problem in this project, since not all participants had responded to the sec ond wave of data collection. Typically, data such as these are analyzed by using only the c ases with fully completed responses in both waves on all relevant variables, discarding in complete responses. Treating the data in this fashion not only results in a reduction of sample size, but more importantly, implicitly assumes the group of participants who an swered all questions to be similar to the group who did not. Should this assumption not h old true, sample bias results. Specific to the present work, analyzing only partic ipants who were followed up presumes these dropouts to have similar characteris tics to the dropouts who were not
6 of 19successfully located or who refused to participate. Further, inclusion of only those participants who answered all items would result in a substantially reduced sample size. To address issues of bias and power, multiple imput ation was used to account for the missing data in this study (Rubin, 1987; Schafer, 1 997). Multiple imputation has been shown to be an appropriate and robust method for es timating missing data in social science settings (Graham, Hofer, Donaldson, MacKinn on, & Schafer, 1997). In multiple imputation, missing values for any variable are predicted using existing values from other variables. The predicted values, called Â“imputesÂ”, are substituted for the missing values, resulting in a full data set ca lled an Â“imputed data setÂ”. This process is performed multiple times; results from the imput ed data sets are combined for the analysis. Multiple imputation accounts for missing d ata by restoring not only the natural variability in the missing data, but also by incorp orating the uncertainty caused by estimating missing data. Maintaining the original v ariability of the missing data is done by creating values which are modeled as a function of variables correlated with the missing data and with the causes of "missingness." Random errors from a normal distribution are added to these predicted values to produce the imputed values. Imputed values produced from an imputation model are not in tended to be Â“guessesÂ” as to what a particular missing value might be; rather, this mod eling is intended to create an imputed data set which maintains the overall variability in the population while preserving relationships with other variables. To incorporate the uncertainty associated with estimating missing data, K multiple models are drawn from the distribution of plausible models for the population. These models are used to produce K imputed data sets. Parameter estimates are then ob tained by combining these K imputed data sets. The parameter of interest in the current s tudy is the log odds, denoted by in the formulas below. Parameter estimates are computed by averaging the point estimates, obtained from the imputed data sets thusly: The total variance of is given by the formula T = W' + (1 + KÂ–1 ) B where W' = the average of the K imputed variances, and the between-imputation variance of the estimates of Thus, the total variance of is made up of a within-imputation component, W' which estimates the natural variability in the data, and a between-imputation component, B which estimates uncertainty caused by estimating mi ssing data (Rubin, 1987). Confidence intervals (95%) for are given by the usual formula, with confidence intervals for odds ratios obtained by exponentiating the bounds of the confidence intervals for theta. Degrees of freedom for t-statistics are given by the formula
7 of 19df = ( K Â– 1)[1 + KW' ( K + 1) Â–1 BÂ–1] 2Multiple imputation and combination of parameter es timates was performed using the NORM for Windows software package (Schafer, 1999). Multiple imputation is an appropriate meth od for treating missing data if correlates of the dependent variable are considered and if the causes of the missing data are measured and available for analysis. To this end, i t is important the imputation model is carefully chosen, ensuring that biases introduced b y "missingness" are eliminated. The variables which were included in the logistic regre ssion models were necessarily included in the imputation modeling. Also utilized were items correlated with "missingness": location (city or mid-sized communit y), substance involvement, whether the participant had ever been suspended from school whether the participant moved into the district from another district, current living arrangements, and whether the participant's family rented or owned their house. Logistic regression modeling. The research questions in the present study were answered through logistic regression analysis, defi ning two separate dichotomies as dependent variables Â– degree/no degree, and diploma /GED. Thus, one set of logistic regression models was estimated to ascertain factor s which significantly predict attainment of a high school education (either a dip loma or GED) or attainment of nothing. Then, the sample was restricted to partici pants who have attained a high school education, and models were estimated which distingu ish between possession of a diploma versus possession of a GED. Model selection was performed using a hier archical backward selection process. In each model, all main effects were examined, along w ith two-way interactions involving ethnicity, gender and SES (Other interactions were too numerous to examine in one analysis, and no theoretical base was available to justify inclusion or exclusion of particular interactions. The demographic variables ethnicity, gender and SES are the most commonly included variables in return research and are therefore the most pertinent to include in interactions). From this Â“f ullÂ” model, interactions were examined separately for significance at the .05 level, using the Wald statistic. The interaction with the smallest Wald statistic was eliminated from the model, then the model was reestimated with the remaining main effects and inter actions. This process was repeated until only main effects and significant interaction s remained, if any interactions were significant. If interactions were significant, the main effects supporting these interactions were necessarily retained in the model. The process then was performed similarly for main effects not involved in significant interactio ns. This process was repeated until the remaining model consisted only of significant facto rs. These factors were then retained as the most parsimonious set of factors which descr ibed the outcome. For each model, slope estimates ('s) and standard deviations of slope estimates were obtained by performing a separate logistic reg ression analysis on each imputed data set. These slope estimates and standard errors were then combined as described in Â“Multiple imputationÂ” above, producing one set of s lope estimates and standard deviations, similar in appearance to what would res ult from a logistic regression analysis which did not use multiple imputation. Wald statist ics were computed and significance was assessed using these combined estimates.ResultsSample Demographics
8 of 19 Participants were 1,071 adolescents who qui t high school at some point during their schooling. Because of budget constraints, the small town was eliminated from the followup sample. Of these participants, 204 (19%) were non-Latino white males, 163 (15%) were non-Latino white females, 400 (37%) were Mexican American males, and 304 (28%) were Mexican American females. The urban location contributed 795 (74%) participants, while 276 (26%) were from the midsi zed location. The age at dropout of these participants ranged from 13 to 21, with 6 par ticipants (1%) having dropped out in 7 th grade, 24 (2%) in 8 th grade, 251 (23%) in 9 th grade, 314 (29%) in 10 th grade, 299 (28%) in 11 th grade, and 177 (17%) in 12 th grade. Note that a full 26% of the participants in the present study dropped out at 9 th grade or earlier, a group previously not included in studies of returning dropouts. Follow-up surveys were completed by 519 (49 %) of the participants. Of these, 508 (47%) responded to the items regarding high school completion. There were 217 (43%) with no high school credentials, 175 (34 %) with GE D certificates, and 116 (23%) with a high school diploma. Table 1 gives breakdowns of degree attainment for ethnicity and gender.Table 1 Description of Degree Attainment, for Ethnicity and Gender No DegreeGEDDiploma Male 114 (43%)97 (36%)56 (21%) Female 103 (43%)78 (32%)60 (25%) Non-Latino White 55 (34%)34 (42%)39 (24%) Mexican American 162 (47%)107 (31%)77 (22%) Table 2 gives means and standard deviation s for the other variables considered in this study. The categorical variable (children) is included with a percent response to one category. The last column of Table 2 gives the perc ent of missing data for each independent variable considered in the present stud y. Possession of high school credentials was the only variable from the second w ave of data utilized in this study. Accordingly, this variable has the greatest proport ion of missing values. The variable measuring teacher caring was not included in the fi nal two years of data collection, so it also has a high percentage of missing responses. Be cause of incomplete records, achievement tests were not always available for the se students, resulting in the high percentage of missing data for this variable. Final ly, since the socio-economic status measure included questions about both parents, many students who did not have two parents left blank the item inquiring about the abs ent parent. Multiple imputation was used to account for missing data in these and other variables.Table 2 Description of Independent VariablesContinuous Variables
9 of 19 Table 3 gives means or percentages for each variable used in the logistic regression models, broken down by respondents and non-responde nts (participants with and without follow-up data). Using statistical signific ance as a guide (alpha =.10), Mexican American participants and female participants were overrepresented in the follow-up sample. Mexican American participants comprised 68. 6% of the respondents, as opposed to 63.0% of the nonrespondents, and 47.0% o f the respondents were female, as opposed to 40.4% of the nonrespondents. Respondents scored slightly higher on achievement tests and were slightly younger.Table 3 Means and Percentages, by Respondents and Non-respo ndentsFactorRespondentNon-respondentpEthnicity68.6% MA63.0% MA0.03Gender47.0% female40.4% female0.06SES0.040.070.48Test scores55.8452.350.03Age at dropout16.5416.670.09Grade at dropout10.3010.320.79GPA220.127.116.11Have or expectingchildren 18.2% yes17.1% yes0.64 Teacher caring2.702.730.69 Distribution of Degree Attainment Combining estimates of degree attainment ac ross the twenty imputed data sets estimated that 40.8% of high school dropouts had no degree, 35.0% had a GED certificate, and 24.2% had a high school diploma.Final Logistic Regression Models Since the variables of interest were dicho tomous (degree/no and diploma/GED), logistic regression was an appropriate analysis. Fo r each logistic regression analysis in this section, predicted odds ratios are presented, and each estimate of an odds ratio is accompanied by a 95% confidence interval. All estimates were obtained using multiple imputation (see Method). Typically, no more than ten data sets are needed for multiple imp utation. However, preliminary examination of results using 10 imputed data sets i ndicated a greater amount of imputed data was needed to ensure stability of the estimate s and to guarantee that variability due to imputation would be properly estimated. This is analogous to the practice of drawing
10 of 19 a large sample to ensure that results will properly infer to the population. Therefore, 20 imputed data sets were used. Tables 4 and 5 give the estimated odds rat ios with 95% confidence intervals for significant factors in each model. Estimates of odd s ratios are given in terms of the increase in odds for one unit change of the indepen dent variable. Degree vs. no degree. As described in Table 4, socio-economic status, te st scores and age at dropout were the only variables shown to be significantly related to returning for a degree. Socio-economic status was positively associated with degree attainment, with a one point increase on the SES scale associat ed with an increase in the odds of returning of 1.34. A participant's test scores were positively related to degree attainment. A one point increase in average test score increase d the odds of gaining a high school degree by a factor of 1.02, while a 10 point increa se in test scores increased the odds of gaining a high school degree by a factor of 1.21 (1 .21 = 1.0210). Participants who dropped out as older adolescents were more likely t o gain some form of high school credentials. For every year of age, the odds a part icipant would return for a degree was increased by 1.28. Thus, a participant who dropped out at age 18 was 2.12 times more likely to get a degree than a participant who dropp ed out at age 15 (2.12 = 1.28 3 ).Table 4 Final Model Describing Degree Attainment: Variables From Previous Dropout LiteratureFactorOdds Ratio 95% Conf. Interval(Lower Bound, Upper Bound)se()tdfp SES1.341.01, 1.790.290.1452.03910.045Test scores1.021.01, 1.030.020.0054.07500.000Age at dropout1.281.12, 1.470.250.0693.571170.001Intercept0.010.00, 0.10-4.681.199-3.901040.000 Note. Dependent variable is degree/no degree. High school diploma vs. GED. As described in Table 5, socioeconomic status, t est scores, children and grade at dropout significantly predicted the choice between a diploma or GED. Socio-economic status was positivel y associated with GED attainment. A one-point increase in the SES score was associate d with an increase of 1.47 in the odds of GED attainment (an increase of .68 in the odds o f diploma attainment). Higher test scores were also associated with GED attainment. Si milar to the previous model, a one point increase in test scores was associated with a n increase in the odds of GED attainment by a factor of 1.02, (an increase of .98 in the odds of diploma attainment) while a 10-point increase raised these odds by a fa ctor of about 1.21. Having or expecting a child at the time of dropout was also associated with GED attainment. Degree holders having or expecting children were 1.92 times as lik ely to have a GED than a diploma (.52 times as likely to have a diploma than a GED). The amount of school a participant
11 of 19 completed was a strong predictor of the type of deg ree held. A participant was approximately twice as likely to have a diploma for each increase in grade at dropout. To illustrate, someone who dropped out in 11 th grade was estimated to be 7.46 times more likely to have a diploma than someone who dropped o ut in 8 th grade.Table 5 Final Model Describing Choice of Degree: Variables From Previous Dropout LiteratureFactor Odds Ratio 95% Conf. Interval(Lower Bound, Upper Bound)se()tdfp SES0.680.47, 0.99-0.380.188-2.01930.047Test scores0.980.97, 0.99-0.020.006-3.11680.003Grade at dropout 1.951.52, 2.510.670.1265.31790.000 Children0.520.28, 0.95-0.650.305-2.141110.035Intercept0.000.00, 0.03-6.261.296-4.83790.000 Note. Dependent variable is diploma/GED. Note. Children is Y/N.Discussion The present study extended and clarified p revious work regarding degree attainment in high school dropouts. Previous studie s had provided information on returning dropouts, but had been unable to include students who dropped out before 10 th grade and students who were unavailable for subsequ ent followup. The present study was able to estimate relationships within the entir e dropout population by including students who dropped out before 10 th grade, and by using multiple imputation to estimate effects of students not included in follow up data collection. Also, although previous studies had identified factors significant ly associated with returning, each study contained omissions of factors deemed important by other studies. The present study was able to consider a broader view of the dropout' s situation by collecting factors found significant in other studies, thus answering questi ons regarding the significance of these factors in the presence of other important factors. Finally, the present study compared dropouts without degrees to those with either a dip loma or GED, performed in return studies only by Kolstad and Kaufman (1989). Two separate logistic regression models we re estimated, one discerning between dropouts with some sort of degree and those with no degree, the other discerning between dropouts with diplomas and those with GEDs. Results indicated that dropouts of higher socio-economic status, higher achievement test scores and greater age at dropout were more likely to attain some sort of deg ree. Analyses further showed that
12 of 19dropouts of higher socio-economic status, with high er test scores, and who dropped out having or expecting children were more likely to ha ve GED certificates than high school diplomas, while those who dropped out in later grad es were more likely to have diplomas than GEDs. Commonly identified factors suc h as ethnicity and gender were not significantly associated with either dependent measure. How Many Gain Degrees? One of the most striking findings of the p resent study is perhaps the simplest, that an estimated 59.2% of the high school dropouts from this study have returned to gain either a high school diploma or GED certificate. Th is result supports the assertions of previous studies that dropping out does not represe nt the end of a student's education. Further, it gives evidence of an increasing trend i n degree attainment over the last ten years, as the estimate is 15.2% higher than the 44% estimate given by Kolstad and Kaufman (1989). The difference is even more notewor thy when one considers that the present study includes participants who dropped out between seventh and twelfth grades, while the Kolstad and Kaufman study only included p articipants who dropped out in the tenth through twelfth grades. Grade has been shown in both studies to be positively associated with degree attainment, so the Kolstad a nd Kaufman estimates should be biased upward. Also important to note from this finding i s the role played by multiple imputation in reducing the bias introduced by participants who did not respond to the second wave of data collection. It has been commonly assumed (e .g., Kolstad, 1988) that dropouts who did not respond to subsequent waves of longitud inal data were Â“hard coreÂ” dropouts who were less likely to hold high school credential s. Such assumptions are admittedly conjecture, since degree estimates for this populat ion were unavailable. The present study, however, estimated that dropouts who do not participate in subsequent data collection actually are slightly more likely to hav e some form of high school credential. Degrees were held by 57.2% of the participants who participated in the follow-up wave; estimates using multiple imputation indicated that 59.2% of the total sample holds high school credentials.Degree vs. no degree The results from this study indicate that generally, dropouts who gain some form of high school degree are of higher socio-economic status (SES), possess higher school capability (as measured by test scores), and are ol der when they drop out. The age and capability findings are consistent with previous li terature and the fact that the present study proves these findings while accounting for ea rlier dropouts, participant nonresponse, and a wider breadth of factors suggest s that these factors are robust. The SES finding clarifies some confusion in previous li terature as to the significance of this factor. These findings stress the importance of tar geting students of low SES and low capability, in addition to continued emphasis on ea rly dropout prevention. Possibly the greatest contribution of the model describing degree attainment is in the clarification of factors which are not signific antly associated with returning for a degree. For instance, previous research had identif ied interactions involving ethnicity and SES, test scores and SES, gender and ethnicity, and gender and grade at dropout, but these interactions were not presented controlling f or other important variables (Kolstad and Kaufman (1989); Kolstad and Owings (1986)). Res ults from the present study
13 of 19indicate that although significantly associated wit h degree attainment, grade at dropout and SES operate independently of other factors. Fur ther, ethnicity and gender are not significant at all when controlling for other facto rs. The fact that ethnicity was not found to be significant in these models should not be construed as a statement that ethnicity is unrelate d to degree attainment. The univariate relationship between degree attainment and ethnicit y indicated that non-Latino white dropouts are 1.73 times more likely to return to ea rn some form of high school degree (95% CI: 1.23, 2.43). However, the multivariate mod el indicated that SES, achievement test scores and age at dropout sufficiently explain the ethnic differences involved in the univariate effect. Further inspection of these resu lts reveals that Mexican American dropouts display more risk in these factors than do non-Latino white dropouts. The test scores of Mexican American dropouts were on average 15.56 percentile points less than non-Latino white dropouts (95% CI: 12.08, 19.03), M exican American dropouts were 2 months younger than non-Latino white dropouts (95% CI: .08, 3.85), and Mexican American dropouts averaged .56 of a standard deviat ion less on the SES scale than non-Latino white dropouts (95% CI: .48, .64). That these factors account for the univariate effect helps clarify some contradictory findings from previous literature on returning dropouts Â– if a study includes sufficient covariates, ethnic effects should be rendered insignificant.Diploma vs. GED Dropouts who chose a GED over a high schoo l diploma were typically of higher socio-economic status (SES), possessed greater leve ls of school capability and were more likely to have children. Dropouts who chose to get a diploma rather than a GED typically dropped out at a later grade. The grade in which a student drops out of h igh school is a strong predictor of which degree (s)he will attain. This is not unexpected Â– for a student who dropped out early in her/his high school career, finishing a high school diploma takes more time and effort than would attaining a GED. The magnitude of the gr ade/attainment relationship is large, more so than found by Kolstad and Kaufman (1 989). This is likely due to the inclusion of younger dropouts in the present study. Students of higher SES and of higher schoo l capability were more likely to get a GED than a high school diploma. These results sugge st that many students have the resources and capability needed to complete high sc hool, but for some reason, school does not provide them with the fit they are looking for. It is possible that these students have specific aims in dropping out Â– given their hi gher social standing and ability, these students may have access to better jobs, schooling or training that require quick attainment of a high school credential. Or, these s tudents may not have a specific goal in mind, but feel they have the ability to succeed at something, and that school does not afford them the environment to succeed as they want to. Also, it is possible these students dropped out with no future plans, then as they decided to return, they had better access to GED programs, GED information, etc., and just saw a GED as a quicker and easier way to get a degree. Kolstad and Kaufman (1989) showed that par ticipants who were parents were more likely to return for some kind of degree, whil e Kolstad (1988) showed these students more likely to stay out of school. Results from the present study indicate that children don't affect overall degree attainment, bu t for those students who did attain a degree, those who had or were expecting children at the time of dropout were more likely to get a GED than a diploma. This is a reaso nable finding, as many of these
14 of 19students would not be able to put forth the time re quired to finish a high school diploma. Also interesting is that there is no interaction wi th this factor and gender, indicating that the effect is the same for males as it is for femal es. Many studies (e.g., Rumberger, 1983; Wehlage & Rutter, 1986) suggest that females are more likely to drop out for child-related reasons. However, the return process is not that way. Implications The results and conclusions presented here have implications for education, and more specifically, dropout prevention and retrieval Because of the breadth of factors considered, and the consideration of dropouts previ ously left out, this study has been able to clarify questions arising from previous res earch. In doing so, the present study has identified a group of factors which together ap pear to be most proximal in effecting degree attainment. This study has joined previous research in affirming that dropping out is not the end of a student's education. Degree attainment in drop outs is a common occurrence, and results from the present study suggest it is more c ommon now than ever. Despite these findings, the research devoted to dropping out of h igh school continues to weigh heavily toward studying causes and correlates of dropping o ut. It is imperative that research institutions and school systems greatly increase ef forts to help dropouts return for degrees if in fact, they do drop out. In some schoo ls, this may be an untapped resource in the fight to reduce dropout rates. The simplicity of the final models should b e helpful for practitioners. Based on factors considered here, degree attainment, whether by way of diploma or GED, can be explained in terms of a few important factors. Furt her, the decision to return for a degree operates similarly regardless of gender or ethnicit y (Mexican American or nonLatino white). Therefore, the models estimated here sugges t that dropout retrieval programs (and some facets of dropout prevention programs) ca n possibly be simplified, streamlined, and ultimately, less expensive. Important for practitioners also is the fin ding that of dropouts who return for degrees, GED-holders on average have higher school capability. As described above, the reasons why these factors are significant are not e vident. It is clear that these students have capability to do school work, and seemingly, s chool is not a good fit for them. However, it seems that these students are walking a dangerous line in opting for a GED instead of a diploma, since attainment of a high sc hool diploma is associated with more labor and economic success than is attainment of a GED certificate (Cameron & Heckman, 1993; Passmore, 1987). This is not to say that for all students, a high school degree is a better choice than a GED, but research suggests that a high school degree is better for most students unless there is a demonstr ated situation where the GED would be better. Therefore, schools should persevere to p rovide opportunities which could channel these students toward diploma attainment, a n endeavor which will likely be more positive for the student in the long run. Although there are positives associated wi th the simplicity of these models, the specific factors identified are also discouraging f or practitioners attempting to change the life trajectories of these dropouts. Starkly ob vious from the models presented here is the fact that degree attainment in dropouts is a fu nction of factors in a student's life which are very difficult for schools to change. Des pite the fact that this study has clarified many issues regarding returning dropouts, it is now clear that different frameworks should be explored in order to identify factors which are more easily changed by practitioners.
15 of 19 Educational research can inform decisions on where to turn next. Finn and Rock (1997) have argued that the research on academic su ccess has placed undue focus on relatively constant characteristics of the individu al, and that more focus should be placed on factors which can be changed by educators. Augme nting this notion is the assertion by Alva (1991), that subjective student appraisals are very important in the evaluation of the student's educational experience. School struct ure could play a role in helping dropouts return, in fact, many researchers (e.g., F inn & Rock, 1997; Wehlage & Rutter, 1986) believe that the secret to educating at-risk students lies in the alteration of factors related to school. Judicious alteration of school f actors could serve to aid in positive alteration of individual factors. Thus, there is room for future research on returning dropouts to expand into a less restrictive framework. Attention should be turned t o more positive correlates, ones associated with academic success rather than failur e, aiming to identify areas where both the school and student can more easily exact positi ve change. Candidates for such expansion include the roles of attitudinal factors, which are more malleable and more internal to the student, factors pertaining to peer s and family, factors pertaining to schools, such as teacher attitudes and communicatio n, and school opportunities and definitions of success.Conclusion The present study has extended previous res earch on dropouts who gain degrees. This study has found, as have other studies, that h igh school dropouts frequently return to gain degrees of some form, a finding which under scores the need for more research in this area. This study has also provided clarificati on of correlates of degree attainment. In doing so, it has presented a neat, concise package of factors which influence returning for a degree. Although concise, this group of facto rs also presents a problem, in that they are factors which are difficult to change in order to create a more positive situation for a dropout. Hence, this study has illuminated the need for additional studies on returning dropouts which can build upon knowledge presented h ere. Such studies should endeavor to consider more positive correlates of returning, ones which can more easily be effected by schools and practitioners.Notes This study was supported by the National Institute on Drug Abuse under grant R01 DA 04777. 1. The author is especially grateful to Randall C. Swa im, and also to Brian Cobb, Bill Timpson and Cori Mantle-Bromley for their insi ghtful comments. In addition, thanks to Ernest Chavez for making available the da ta used in this study. 2.ReferencesAlva, S.A. (1991). Academic invulnerability among M exican-American students: The importance of protective resources and appraisals. Hispanic Journal of Behavioral Sciences,13 (1), 18-34. Borus, M.E. & Carpenter, S.A. (1983). A note on the return of dropouts to high school.
16 of 19Youth and Society, 14, 501-507. Cameron, S.V., & Heckman, J.J. (1993). The nonequiv alence of high school equivalents. Journal of Labor Economics, 11 (1), 1-47. Catterall, J.S. (1987). On the social costs of drop ping out of school. High School Journal, 71, 19-30. Chavez, E.L., Deffenbacher, J.L, & Wayman, J.C. (19 96). A longitudinal study of drug involvement in Mexican American and white non-Hispa nic high school dropouts, academically at risk students and control students. Free Inquiry in Creative Sociology, 24, 185-193. Chavez, E.L., Oetting, E.R., & Swaim, R.C. (1994). Dropout and delinquency: Mexican-American and Caucasian nonHispanic youth. Journal of Clinical Child Psychology, 23 (1), 47-55. Chuang, H. L. (1997). High school youths' dropout a nd re-enrollment behavior. Economics of Education Review, 16 (2), 171-186. Ekstrom, R.B., Goertz, M.E., Pollack, J.M., & Rock, D.A. (1986). Who drops out and why? Findings from a national study. Teachers College Record, 87 356373. Elliott, D.S., Huizinga, D., & Ageton, S. S. (1985) Explaining delinquency and drug use. Newbury Park, CA: SAGE. Finn, J.D., & Rock, D.A. (1997). Academic success a mong students at risk for school failure. Journal of Applied Psychology, 82 (2), 221-234. Graham, J.W., Hofer, S.M., Donaldson, S.I., MacKinn on, D.P., & Schafer, J.L. (1997). Analysis with missing data in prevention research. In K. Bryant, M. Windle, & S. West (Eds.), The science of prevention: Methodological advances from alcohol and substance abuse research. (pp. 325-366). Washington, D.C.: American Psycholo gical Association. Kaufman, P. (1988). High school dropouts who return to school. Unpublished doctoral dissertation, Claremont Graduate School, Claremont, CA. Kolstad, A.J. & Owings, J.A. (1986). High school dropouts who change their minds about school. Washington, DC: Office of Educational Research and Improvement. (ERIC Document Reproduction Service No. ED 275 800)Kolstad, A.J. & Kaufman, P. (1989, March). Dropouts who complete high school with a diploma or GED. Paper presented at the annual meeting of the Ameri can Educational Research Association, San Francisco, CA.Morrow, G. (1986). Standardizing practice in the an alysis of school dropouts. Teachers College Record, 87 342-355. Passmore, D.L. (1987). Employment of young GED recipients. American Council on Education Research Brief (14).Rubin, D.B. (1987). Multiple imputation for nonresponse in surveys. New York: Wiley.
17 of 19 Rumberger, R.W. (1983). Dropping out of high school : The influence of race, sex and family background. American Educational Research Journal, 20, 199220. Rumberger, R.W. (1987). High school dropouts: A rev iew of issues and evidence. Review of Educational Research, 57 (2), 101-121. Schafer, J.L. (1999) NORM: Multiple imputation of i ncomplete multivariate data under a normal model, version 2. Software for Windows 95/ 98/NT, available from http://www.stat.psu.edu/~jls/misoftwa.html.Schafer, J.L. (1997). Analysis of incomplete multivariate data. New York: Chapman and Hall.Waxman, H.C., Huang, S.L., Padron, Y.N. (1997). Mot ivation and learning environment differences between resilient and nonresilient Lati no middle school students. Hispanic Journal of Behavioral Sciences, 19 (2), 137155. Wehlage, G.G., & Rutter, R.A. (1986). Dropping out: How much do schools contribute to the problem? Teachers College Record, 87 374-392.About the AuthorJeffrey C. Wayman Ph.D. Tri-Ethnic Center for Prevention Research101 Sage Hall, Colorado State UniversityFort Collins, CO 80523.Phone: (970) 491-6969Fax: (970) 491-0527.Email: firstname.lastname@example.orgUpdated Email (October, 2002): email@example.comJeff Wayman is a Research Associate with the Tri-Et hnic Center for Prevention Research, at Colorado State University. He holds a Ph.D. in Education and a Masters in Statistics. His current educational research intere sts include at-risk students, educational resilience, and cultural issues, and how these issu es relate to teacher training and school reform. Current methodological interests include mi ssing data issues and multilevel modeling.Copyright 2001 by the Education Policy Analysis ArchivesThe World Wide Web address for the Education Policy Analysis Archives is epaa.asu.edu General questions about appropriateness of topics o r particular articles may be addressed to the Editor, Gene V Glass, firstname.lastname@example.org or reach him at College of Education, Arizona State University, Tempe, AZ 8 5287-0211. (602-965-9644). The Commentary Editor is Casey D. C obb: email@example.com
18 of 19EPAA Editorial Board Michael W. Apple University of Wisconsin Greg Camilli Rutgers University John Covaleskie Northern Michigan University Alan Davis University of Colorado, Denver Sherman Dorn University of South Florida Mark E. Fetler California Commission on Teacher Credentialing Richard Garlikov firstname.lastname@example.org Thomas F. Green Syracuse University Alison I. Griffith York University Arlen Gullickson Western Michigan University Ernest R. House University of Colorado Aimee Howley Ohio University Craig B. Howley Appalachia Educational Laboratory William Hunter University of Calgary Daniel Kalls Ume University Benjamin Levin University of Manitoba Thomas Mauhs-Pugh Green Mountain College Dewayne Matthews Western Interstate Commission for HigherEducation William McInerney Purdue University Mary McKeown-Moak MGT of America (Austin, TX) Les McLean University of Toronto Susan Bobbitt Nolen University of Washington Anne L. Pemberton email@example.com Hugh G. Petrie SUNY Buffalo Richard C. Richardson New York University Anthony G. Rud Jr. Purdue University Dennis Sayers Ann Leavenworth Centerfor Accelerated Learning Jay D. Scribner University of Texas at Austin Michael Scriven firstname.lastname@example.org Robert E. Stake University of IllinoisÂ—UC Robert Stonehill U.S. Department of Education David D. Williams Brigham Young UniversityEPAA Spanish Language Editorial BoardAssociate Editor for Spanish Language Roberto Rodrguez Gmez Universidad Nacional Autnoma de Mxico email@example.com
19 of 19 Adrin Acosta (Mxico) Universidad de Guadalajaraadrianacosta@compuserve.com J. Flix Angulo Rasco (Spain) Universidad de Cdizfelix.firstname.lastname@example.org Teresa Bracho (Mxico) Centro de Investigacin y DocenciaEconmica-CIDEbracho dis1.cide.mx Alejandro Canales (Mxico) Universidad Nacional Autnoma deMxicocanalesa@servidor.unam.mx Ursula Casanova (U.S.A.) Arizona State Universitycasanova@asu.edu Jos Contreras Domingo Universitat de Barcelona Jose.Contreras@doe.d5.ub.es Erwin Epstein (U.S.A.) Loyola University of ChicagoEepstein@luc.edu Josu Gonzlez (U.S.A.) Arizona State Universityjosue@asu.edu Rollin Kent (Mxico)Departamento de InvestigacinEducativa-DIE/CINVESTAVrkent@gemtel.com.mx email@example.com Mara Beatriz Luce (Brazil)Universidad Federal de Rio Grande do Sul-UFRGSlucemb@orion.ufrgs.brJavier Mendoza Rojas (Mxico)Universidad Nacional Autnoma deMxicojaviermr@servidor.unam.mxMarcela Mollis (Argentina)Universidad de Buenos Airesmmollis@filo.uba.ar Humberto Muoz Garca (Mxico) Universidad Nacional Autnoma deMxicohumberto@servidor.unam.mxAngel Ignacio Prez Gmez (Spain)Universidad de Mlagaaiperez@uma.es Daniel Schugurensky (Argentina-Canad)OISE/UT, Canadadschugurensky@oise.utoronto.ca Simon Schwartzman (Brazil)Fundao Instituto Brasileiro e Geografiae Estatstica firstname.lastname@example.org Jurjo Torres Santom (Spain)Universidad de A Coruajurjo@udc.es Carlos Alberto Torres (U.S.A.)University of California, Los Angelestorres@gseisucla.edu