USF Libraries
USF Digital Collections

Evaluating land use methods for altering travel behavior

MISSING IMAGE

Material Information

Title:
Evaluating land use methods for altering travel behavior transit patronage as a product of land use potential and connectivity : the Sacramento case
Portion of title:
Transit patronage as a product of land use potential and connectivity
Physical Description:
1 online resource (vii, 47 leaves) : maps. ;
Language:
English
Creator:
Thompson, Gregory Lee, 1946-
Frank, James E
United States -- Dept. of Transportation. -- University Research Program
Florida State University -- Dept. of Urban and Regional Planning
National Urban Transit Institute (U.S.)
Publisher:
Florida State University, Department of Urban and Regional Planning
Place of Publication:
Tallahassee, Fla
Publication Date:

Subjects

Subjects / Keywords:
Local transit -- Planning -- Mathematical models -- California -- Sacramento Metropolitan Area   ( lcsh )
Land use -- Mathematical models -- California -- Sacramento Metropolitan Area   ( lcsh )
Suburbs -- Mathematical models -- California -- Sacramento Metropolitan Area   ( lcsh )
Choice of transportation -- Mathematical models   ( lcsh )
Genre:
bibliography   ( marcgt )
technical report   ( marcgt )
non-fiction   ( marcgt )

Notes

Bibliography:
Includes bibliographical references (leaves 45-47).
Statement of Responsibility:
Gregory L. Thompson, James E. Frank.
General Note:
Title from e-book t.p. (viewed Aug. 16, 2011).
General Note:
Performed by Florida State University, Dept. of Urban and Regional Planning for the the National Urban Transit Institute and the U.S. Dept. of Transportation, University Research Institute Program.
General Note:
"January 1995."
General Note:
Report.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 029019397
oclc - 746955899
usfldc doi - C01-00286
usfldc handle - c1.286
System ID:
SFS0032368:00001


This item is only available as the following downloads:


Full Text
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 2200421Ia 4500
controlfield tag 001 029019397
005 20110816174830.0
006 m d
007 cr bn|||||||||
008 110816s1995 caub obt 000 0 eng d
datafield ind1 8 ind2 024
subfield code a C01-00286
035
(OCoLC)746955899
040
FHM
c FHM
043
n-us-ca
090
HE4491.S23 (ONLINE)
1 100
Thompson, Gregory Lee,
d 1946-
0 245
Evaluating land use methods for altering travel behavior.
p Final report for Task 1b
h [electronic resource] :
b transit patronage as a product of land use potential and connectivity : the Sacramento case /
Gregory L. Thompson, James E. Frank.
3 246
Transit patronage as a product of land use potential and connectivity
260
Tallahassee, Fla. :
Florida State University, Department of Urban and Regional Planning,
[1995]
300
1 online resource (vii, 47 leaves) :
maps.
500
Title from e-book t.p. (viewed Aug. 16, 2011).
5 FTS
513
Report.
May 1993-Dec. 1994.
Performed by Florida State University, Dept. of Urban and Regional Planning for the the National Urban Transit Institute and the U.S. Dept. of Transportation, University Research Institute Program.
FTS
"January 1995."
FTS
504
Includes bibliographical references (leaves 45-47).
650
Local transit
z California
Sacramento Metropolitan Area
x Planning
Mathematical models.
Land use
California
Sacramento Metropolitan Area
Mathematical models.
Suburbs
California
Sacramento Metropolitan Area
Mathematical models.
Choice of transportation
Mathematical models.
700
Frank, James E.
710
United States.
Dept. of Transportation.
University Research Program.
2
Florida State University.
Dept. of Urban and Regional Planning.
National Urban Transit Institute (U.S.)
776
i Print version:
Thompson, Gregory Lee, 1946-
t Evaluating land use methods for altering travel behavior. Final report for Task 1b.
Tallahassee, Fla. : Florida State University, Department of Urban and Regional Planning; Springfield, VA : Available through the National Technical Information Service, 1995]
w (OCoLC)32754355
773
Center for Urban Transportation Research Publications [USF].
4 856
u http://digital.lib.usf.edu/?c1.286
FTS
y USF ONLINE ACCESS
951
10
SFU01:001952625;
FTS



PAGE 1

EVALUATING LAND USE METHODS FOR ALTERING TRAVEL BEHAVIOR Final Report for Task 1b: Transit Patronage as a Product of Land Use Potential and Connectivity: The Sacramento Case Principal Investigators Gregory L. Thompson James E Frank January 1995 Florida State University Department of Urban and Regional Planning Tallaha s see, Florida 32306-2030 (904) 644-4510 office (904) 644-6041

PAGE 2

. : The contents of this report reflect the views of the authors, who are responsible for the facts and the accuracy of the Information presented herein Thi s document Is dl$$amlnated under the sponsorship of the Department of Transportation, University Research Institute Program I n the Interest of information exchange. The U. S. Govemment assumes no liability for the contends or use thereof.

PAGE 3

. "'"'' "' NUTI93FSU4 1-2 2. Govf'II'Mf'll HI). s. flo. 4 llllo end &J:rlltlo tl. Repon o:uo EVALUATING LAND USE METHODS FOR ALTERING TRAVEL January 1995 BEHAVIOR: Final Report for Task 1b, Transit Pattonage as a Product of 6. Cod"e Land Use Potential and Connectivity: The Sacramento Case 1. Aa.lloOO e. Organ"lialion Aepon NO. Gregory L. Thompson and James E. Frank 9. 1'efformrng Org!lnizotbrl tlome ol'ld AdO$S 10. \Y'Qtk (.Mit No. Nat i onal U r ban Transit Institute Department of Urban and Regional Planning, Florida State Uni versity 11. Contr ooaorG r.,No. DTRS 93-G-0019 Tallahassee, Florida 32306-2030 12.. 6pons0f'lrlg Al)ff'IC'f Name Ill Acklft:U 13. TWill of Ropott end Per iod CoYcrcl Office of Research and Speci al Programs May 1993 through December 1994 U S Departtnent of Transportation, Washington, D.C. 20690 1 4 16. Not Supported by a grant from the U.S. Department of Transportation, University Research Institute Program 1 6 This study uses Sacramento Regiona l Transit dat a to determine conditions under which transit can attract traffic in suburbs. It specifies a direct demand model i n which tri p p r oductions and attractions are a function of zonal characteristics and transi t connect ivity from a given zone to places where transit users want t o go in the region T h e mode l der i vation i dentif ies a new concept, trip production and attraction potential, which is Independent of transportation connectivity but which is a function of zonal socioeconomic and design characteristics. We estimate potentials from a samp l e of zones where it I s known how many transit trips occur between each pair of zones in the sample. The direct demand model infers potentials from the magnitude of trips between each pair of zones i n relation to the leve l of transi t service linking t h e zones. Estimated po t entials can be transformed into transit trip productions and attractions through the levels of ttansit connect ivity to the rest of the region Analysts a l so may regress potentials against zona l characteristics to determine what qualities of the z ones influence transit production and attraction poten t ials. The study estimates production and attraction potentials and connectivity for 70 Sacramento County census tracts. Wrth this information, n: identifies suburban tracts where the tran sit system succeeds I n taping potentia l traffic, and it identifies qualities of the transit system assoc i ated with such succe ss It a lso shows wher e the transit system fails to tap potential. It a l so illustrates with graphs the relationsh i p between trans i t trip production and attraction and transit connectivity for land uses typical of post Wor l d War II suburbs and for transit oriented development. Fina lly, It regresses production and attraction potent i als against ce n sus tract cha r acteristics to determine what tract qua li ties are associated with potential transit demand. 1 7. Key WordS 1 e. D:slrlbutkln acce.ssibilfty connectivity, production potential, attraction Availa b le to the public through the National Tec h n i ca l Information potentia l t i med transfer, gravity Service {NTIS), 5285 Port Royal Road, Springf i eld, VA 22181, ph (703) model, transit oriented development 487-4650 19. Stcllllty C lmlf. IOIIW.JCIDOffl 20, s.turitv Clmll. IOI tllll p;,90) 21. No. of Pli!IIIS 22. P'1ico I . I ,:,,,.,. nnT F 1700. 7 tS.89l

PAGE 4

Table of Contents Table of Contents ............................................... List of Figures . . . . . . . . . . . . . . . . . . . ii List of Maps . . . . . . . . . . . . . . . . . . . . . . . . . . i i List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii Ack.nowledgments . . . . . . . . . . . . . . . . . . . . . . . v Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . vii Executive Summary . . . . . . . . . . . . . . . . . . . . . . . 1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . 5 LITERATURE REVIEW . . . . . . . . . . . . . . . . . . . . . . . 7 MODEL SPECIFICATION 0 0 0 12 MODEL ESTIMATION 0 20 RESULTS .......... ; . . . . . . . . . ... . . . . . . . . . . 22 DISCUSSION OF RESULTS . . . . . . . . . . . . 25 Tables 1 and 2 . . . . . . . . . . . . . . . . . . . . . . 25 Estimations of Trip Production and Attraction Potentials . . . . . . 26 Tract Characteristics Associated with Trip Production and Attraction Potentials 0 0 0 39 CONCLUSIONS . . . . . . . . . . . . . . 43 REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . 45

PAGE 5

list of Figures Figure 1: Transit Trip Productions as A Function of Zone Potential and Connectivity 37 Figure 2 : Transit Trip Attractions as A Function of Zone Potential and Connectivity 38 Map 1. Map 2. Map 3. Map 4. Map 5. Map 6. Map 7. List of Maps Census Tract Identifiers 0 0 0 PDN Distribution (Percent\ ......... .................... ATN1 Distribution (Percent) ........................ E, Distribution (Relative to Average) ........................... 81 Distribution (Relative to Average) ...................... %P/%PDN, (for those Tracts with PDN, Greater Than .9) ............ %A;f%ATN1 (for those Tracts with ATN1 Greater Than .7) .......... list of Tables 27 28 29 31 32 35 36 Table 1: Transportation System Variables in the Equation . . . . . . 24 Table 2: Elasticities of T11' with Respect to the Transportation Variables . . . 24 Table 3 : Explaining PDN1 41 Table 4: Explaining the ATN1 42 i i

PAGE 6

Preface This report presents the results of Task 1 b of NUTI93FSU4, Evaluating Land Use Methods for Altering Travel Behavior. Results of Tasks 1 a, 2, and 3 are found in another report, NUTI93FSU4.1. NU TI 93FSU4 was conceived originally as a five-year project in which we would contribute to the debate on the degree to which land use variables can influence travel behavior, particularly in regard to transit u sage We defined Task 1 as an update of earlier work that examined changing journey to work travel patterns in Florida based on the 1 970 and 1980 census. The schedule called for the completion of this work during the first year of the study. Task 2 called for the use of the nation-wide 1990 PUMS to analyze the degree which state-level policy variables influence modal sp l it. Task 3 called for a case study in Florida in which we would merge county assessor data tapes defining land uses as the parcel level with census STF3 data describing socio-economic characteristics of census tracts as well as journey to work travel behavior. The schedule called for Tasks 2 and 3 to be initiated dur ing year 1 but not completed. Approval was given for year one of the study in May 1993. At about the same time the Center for Urban Transportation Research 11994) produced the first draft of its own report comparing 1990 STF3 journey to work data with that from the 1980 and 1970 censuses. Their report obviated the need for our Task 1. We then asked and rece ived approval to substitute in Task 1 a project in which we wanted to explain transit ridership between pairs of census tracts in Sacramento by the quality of transit service between the census tracts as well as variables describing the census tracts. We hoped to contribute to the debate by controlling for transit connectivity from each zone In the analysis to destinations of importance to transit users The literature on the impact of such variables on transit behavior did not control for transit service leve l adequately in our opinion. The Sacramento study was approved as Task 1 b, while the original Task 1 was renamed Task 1 a. This report describes Task 1 b. It has undergone two revisions in response to criticisms and comments on earlier drafts. We wrote the first draft as a contribution to the literature on whether or not transit oriented developments influenced transit travel behavior. Our contribution was a more precise measure of transit connectivity from one zone to another. We submitted the first draft to the Transportation Research Board Committee on Public Transportation Planning and Development in early August 1994. We also sent copies to Wade White of Gannett Fleming in Tampa, Tom Matoff, General Manager of the Regional Transit Authority in Seattle, Pilka Robinson, General Manager of Regional Transit in Sacramento, and to the Center for Urban Transportation Research in Tampa. Our own internal review of the paper as well as extensive comments from Wade White and comments from Tom Matoff prompted us to extensively rewrite the first draft. The most substant ive change involved redefining the dependent variable to include all trips j i i

PAGE 7

with either end at home as home-based tri p This chan g e substantially affected the, results of the secondary anal yses, in which we regressed production and attraction potentials estimated in the primary analysis against zonal variables. The other change, which proved more farreaching, was our discovery that the potentials were observed trips divided bv accessibilities. This resulted in the insight. achieved just as we fina l ized the second draft, that there was an important reason to regress potentials rather than observed trips against zonal characteristics, because potentials were independent of connectivity whereas observed trips were not. Because this insight came just as the fina l draft was final i zed, we did not develop this argument well in it. At the same time we received the reviews of the first draft from the TAB. In genera l whil e one review was positive, and one was neutral the substance of the negative review suggested that we needed to justify our use of a direct demand model when disaggregate techn i ques and the Urban Transportation Modeling Systems were available. Shortly thereafter Alan Horowitz at the University of Wisconsin, Milwaukee, gave us a critique of the second draft. He viewed direct demand aggregate models as a valid approach but wanted us to justify our model in terms of the literature on joint models that s i multaneous l y pred ict destinat i on and mode. He also wanted justification for using potentials rather than observed trips as the dependent variable in the analyses of what zonal characteristics influence transit trip generation. The third draft addresses the TAB and Horowitz criti ques. It focuses more on model development and drops the last section on s imulati ons. We also have added maps showing the distributions of potentials and trans i t accessibilities. We presented an outline of it to the 1994 annua l meeting of the Association of American Collegiate Schools of Planning in Tempe, Arizona, in early November 1994 where we received additional comments suggesting the desirability of justifying our approach. Tom Matoff a l so wanted a clear presentation of the concept of potent i als. We believe that what we present here does so Our experience w ith the Sacramento study suggests that refinements should be made to our methods before they are used in future studies We intend to do so in the second year of the project in our analysis of of transit connectivity and land use in Orange County, F l orida. iv

PAGE 8

Acknowledgments This project is made possible through a grant from the U.S. Department of T r ansportation, University Research Institute Program. Their support is gratefully acknowledged The project team for this study includes: Mr. David Schmitt, Research Assistant, Department of Urban and Reg i onal P l anning, Florida State University. Mr. Chris Gray, Research Assistant, Department of Urban and Regional Planning, Florida State University. Both Messrs. Schmitt and Gray are second year master's students in the Transportation Planning Specialty in the Department of Urban and Regional Planning. Mr. Schmitt, funded by the project, collected census STF data and wrote computer programs to merge large data sets. The largest of these were the Orange County, Florida land use files which he merged with census STF3 files at the census tract level. He created another program to extract desired variables from the merged data base for analysis, and he ran trial regressions of journey to work va r iables against socio -economic and land use variables. He also wrote programs that created the data bases for the simulations in the first two drafts of the Sacramento study and collected the data for the analyses of variables explaining transit trip generation potentials. Finally, he wrote a program that is capable of uploading the entire one percent PUMS for the nation onto the Florida State University Garnet platform and has abstracted a trial set of variables from it for analysis Currently he i s working with Mr. Gray on assembling data sets (including setting up Tranplan on departmental computers and running Orlando road and transit networks supplied by KPMG Peat Marwick) for the second y e ar project. Mr. Gray already was funded by a Rorida State University Fellowship but volunteered his time to the project. He succeeded in clean ing up the Orange County land use file, which consisted of several million parcels level records and consolidating it into a census-tract based file that could be merged with STF3 files He also conceived and produced all of the maps for the Sacramento project. Cur rently, he is designing a research study as his master's thesis that will correct some of the noted shortcomings of the Sacramento study. He and Mr. Schmitt are beginning to assemble data and carry out this work as part of the second year grant. v

PAGE 9

We also appreciate work done for the Sacramento project by departmental research assistant Mr. Chris King. Mr. King was a first year master's student funded by a departmental fellowship. Dr. Mark Ellis, Associate Professor, Department of Geography, University of California, Los Angeles, also provided the project invaluable assistance in introducing the project team to the literature on Poisson regression and actually performing the first runs on the initial Sacramento data set. He subsequently has reviewed and commented on the results of other runs. Numerous persons have commented on drafts of our work. We would like to thank Tom Matott, Wade White, Alan Horowitz, Susan Handy, and three anonymous reviewers from the Public Transportation Planning and Development Committee of the Transportation Board. vi

PAGE 10

TRANSIT PATRONAGE AS A PRODUCT OF LAND USE POTENTIAL AND CONNECTIVITY': THE SACRAMENTO CASE Abstract This report presents a study that uses Sacramento Regional Transit system collected data to determine conditions under which transit can attract traffic in suburban environments. It does so by specifying a direct demand model in which trip productions and attractions are a function not only of zonal characteristics, but also of transit connectivity from a given zone to places where transit users want to go in t he reg ion The model derivation identifies a new concept, trip production and attraction potential, which we show i s independent of transportation connectivity and which is a function of zonal socio-economic and design characteristics. Such potential can be estimated for a sample of zones where it is known how many transit trips occur between each pair of zones in the sample. Potentials are estimated through an analysis of the magnitude of trips between each pair of zones in relation to the level of transit service linking the zones Once potentials are known, an analyst may estimate transit trip productions and attractions for the zones for different levels of transit connectivity to the rest of the region. Analysts also may regress potentials against zonal characteristics to determine what qualities of the zones infl uence transit production a nd attraction potentials. Having developed the model, the study estimates production and attraction potentials for 70 Sacramento County census tracts. It also estimates transit connectivity from each tract to both production and attraction potentials of all 70 tracts. With this information, it identifies suburban tracts where the transit system succeeds in taping potential traffic, and it identifies qualities of the t ransit system associat ed with such success. It also shows where the transit system fails to tap potential. It also illustrates with graphs the relationship between transit trip production and attraction and transit connectivity for land uses typical of post-World War II suburbs and for transit oriented development. Finally, it regresses production and attraction potentials against census tract characteristics to determine what tract qualities are associated with potential transit demand. Connectivity refers to how accessible a given zone is via transit service to productions or attractions in all zones lor a sample of zones) in a reg ion. We chose not to use the term accessibility, because while accessibility is used in the sense we use connectivity, particularly in the modeling literature, in many readers' minds accessibility refers to access to bus stops, or it refers to whether transit vehicles are accessible to wheel chairs. vii

PAGE 11

Executive Summary The salient characteristic of travel in a decentralizing society is ever-more dispersed trip making Users of the census describe this phenomenon in terms of overwhelming preponderance of suburb-to-suburb work trips compared to two other categories: travel within central areas and travel from suburbs to central areas (Pisarski, 1987; Rossetti and Eversole, 1993; Center for Urban Transportation Research, 1994). Policy analysts have commented on such decentralization and its counterpart, the decl i ne of central business districts (CBDs). since at least the 1960s (Meyer, Kain, and Wohl, 1966). Policy analysts disagree on how t r ansit systems should respond to such trends. One group argues that the suburb-to-CBD and the traditional inner city markets despite their relative declines remain the only markets where t r ansit can maintain modal share (U.S. Department of Transportation, 1987; Jones, 1985; Pisarski, 1992). Others argue that transit should attempt to serve the non-traditional suburb-to-suburb markets, because this is where the bulk of travel demand Is. Reasons for transit failure in these areas are more related to poorly thought-out route structures that fail to allow for many-to-many travel than to the nature of suburban physica l and social structure (Thompson, 1977; Washington and Stokes, 1988). The debate over appropriate transit markets has not been resolved, because previous studies have not controlled for variables necessary to resolve it. Studies claiming that urban form or socio-economic variables are of paramount importance in explaining transit success or failure do not control adequately for transit level of service. Studies concluding that trans i t level of service is of paramount importance do not contro l for urban form or socio-economic variables This study is an attempt to bridge gaps between the urban str ucture-oriented and the transit level-of-service oriented studies by controlling for both sets of variables. It does so by analyzing transit patronage between any two points as a function of how well transit and automobiles connect the two points, population and job densities of the two points, income and transit dependence characteristics of the two points, and design features of the two points. The degree to which transit connects two points is the control for transit level of service. Density, urban design, and socio-economic variables control for various aspects of urban structure. A useful finding of this study is that it i s possib l e to distingu i sh between the potential of an area, such as a census tract, for generating transit traffic and the traffic that actually is generated Because of this distinction, we are able to address the followi ng four questions: 1 What is the range of potentials of suburban census tracts for producing or attracting transit traffic? 1

PAGE 12

2. How well does transit tap the potential that exists? 3. How much transit traffic couid lie generated from transit oriented developments and typical post-World War II suburbs at different levels of transit connectivity to the rest of the region? 4. What census tract characteristics are related with transit potential? The method that we developed is based on data that transit systems can collect; namely, transit origin-destination trip tables derived from onboard surveys. The survey used here was performed for Sacramento Regiona l Transit by Evaluation and Training Institute in May and June 1989 !Evaluation and Training Institute, 19891. The method is a direct demand model that distinguishes between potential and observed transit trips in various parts of the Sacramento regional area. The study has important policy implications. Many suburban areas produce and attract few transit trips, but it is unclear whether the small number reflects lack of potential demand or lack of service. Our study attempts to show which it is and to highlight suburban areas where potential transit demand exists but where there currently is little transit ridership. A suburban tract may lack potential demand, even if a high level of transit service were offered. Improvements to transit service in such areas would waste resources. On the other hand, potential may ex ist for transit traffic, but the service offered to the tract is inappropriate to demand, and traffic does not materialize. Service improvement might be warranted. The key to making such a determination is to distinguish between potential and observed transit traffic. This study does so by analyzing observed flows of transit traffic between pairs of census tracts in relation to the quality of transit and road service linking them. Where transit service is bad, but moderate flows of t r ansit trips occur anyway, there probably is a large potential for additional transit traffic. Where service is relatively good, but little traffic materializes, potential is probably lacking. Through a systematic analysis of transit traffic in relation to service, this study infers the potential of most of the tracts in Sacramento County for producing and attracting transit trips. This method contrasts to three other approaches for determining transit demand Disaggregate models of destination and modal choice often are used to infer what qualities of the choices influence the probability of persons of different socioeconomic characteristics for making a choice. While such models are soundly based in utility maximizing theory, they by themselves cannot predict flows of people from one area to another. Being able to determine what causes such flows is a major concern of transit managers and policy analysts attempting to determine the welfare consequences of transit investments. In order to obtain such insights from disaggregate choice models, the user 2

PAGE 13

must combine them with other models that determine the frequency of trip making from a given area. Another approach, the Urban frarisportatlon Modeling System !UTMS} simulates regional road and transit systems for urban regions thro ughout the U.S. and has the ability to forecast transit traffic for assumed transit systems in suburban regions. We do not use the UTMS for three reasons. First, it assumes that trips are produced and attracted as a function of land uses but not as a function of the transportation sys \ em. We argue later for a different assumption Second, it assumes that trips produced and attracted are split between all available modes. We argue that tota l trips produced and attracted are the sum of modal trip productions and attractions and that each of the modes has a degree of autonomy in trip generation. Third, the UTMS lacks the flexibility to allow transit managers or other policy analysts to translate information that they might have produced such a on-board rid ership surveys, into insights about how the systems might be restructured to better serve demand. Our method shows how ridership survey information can be used to the benef i t of transit policy analysts. A third approach is to infer what socio-demographic or land use characteristics produce or attract transit trips by regressing observed transit trip productions and attraction against such zonal variables. We do not use this approach, because observed trip productions and attractions are influenced as much by service offered as by potential demand that is derived from zonal variables. Such an approach can yield little insight about what land uses i nfluence transit traffic, unless the quality of transit service is controlled. The approach that we use, first inferring transit potential of zones, does control for the quality of transit service. Having obtained knowledge of census tract transit potential, we examined its relationship in each census tract to observed transit productions and attractions, and transit connectivity. We found that two zones with differing potentials produced negligible transit traffic if poorly connected to the rest of the region. As transit connectivity improved, traffic grew much more rapidly for zones with moderate potential than for zones with low potential. Transit oriented suburban developments, for example, have the potential for producing and attracting three to four times as much transit traffic as typical post-World War II suburbs, but the difference does not become obvious unless the quality of transit service measured in terms of connectivity, is raised to fairly high levels. We also found that Sacramento Regional Transit offered greater than average connectivity to several suburban census tracts. All of them were tracts with good transit connections to the CBD as well as to nearby suburban tracts with heavy employment. Most of the tracts with such service had one or more light rail transit stations that functioned as timed transfer bus centers, affording suburb-to-suburb as well as suburb-to CBD movement. We then drew inferences about what characteristics of the tracts cause potential transit demand. The inferences come from r egress i ng the inferred potentials (rather than 3

PAGE 14

observed transit productions and attractions! against variables describing zonal socio economic and design characteristics. Such regressions show which variables are important for explaining potential. Results here were disappointing. Only one variable, the absolute magnitude of a tract's population, was significant in explaining a tract's potential for producing transit traffic. This was highly significant, however, and the model explained over 70 percent of the variance in production potential. Three variables explained attraction potential: employment density had by far the largest effect, but the absolute magnitude of employment, and a dummy variable for core CBD tracts also were significant. These explained over BO percent of the variance in attraction potential. Variables denoting economic status, the degree of census tract coverage by transit, or the percent of dwelling units built before 1940 had no explanatory power. We suspect that the results are determined in part at least by the large range In size of the census tracts, which range from a traction of a square kilometer in central Sacramento to many square kilometers in some suburban areas. We were restricted to census tracts, because our source of transit rider behavior was a survey that was geo-coded to the census tract level. We consider the results as preliminary and as indicating that the method developed in this study holds promise for producing insights if modifications are made in future studies. These include disaggregation of the dependent variable by trip type, the coding of separate peak and off-peak transit and road networks with congested auto times used for the latter, the use of smaller zones, such as traffic analysis zones, and the use of actual rather than straight-line walking distance to transit stops. 4

PAGE 15

TRANSIT PATRONAGE AS A PRODUCT OF LAND USE POTENTIAL AND CONNECTIVITY: THE SACRAMENTO CASE INTRODUCTION The salient characteristic of travel in a decentralizing society is ever-more dispersed trip making. Users of the census describe this phenomenon in terms of overwhelming preponderance of suburb-to-suburb work trips compared to two other categories: travel within central areas and travel from suburbs to central areas (Pisarski, 1987; Rossetti and Eversole, 1993; Center for Urban Transportation Research, 1994). Policy analysts have commented on such decentralization and its counterpart, the decline of central business districts ICBDs), since at least the 1960s (Meyer, Kain, and Wohl, 1966). Policy analysts disagree on how transit systems should respond to such trends One group argues that the suburb-to-CSD and the traditional inner city markets, despite their relative declines, remain the only markets where transit can maintain modal share (U.S. Department of Transportation, 1987; Jones, 1 g85; Pisarski, 1992). Others argue that transit should attempt to serve the non-traditional suburb-to-suburb markets, because this is where the bulk of travel demand is. Reasons for transit failure in these areas are more related to poorly thought-out route structures that fail to allow for many-to-many travel than to the nature of suburban physical and social structure (Thompson, 1977; Washington and Stokes. 1988). The debate over appropriate transit markets has not been resolved, because previous studies have not controlled for variables necessary to resolve it. Studies claiming that urban form or socio-economic variables are of paramount importance in explaining transit success or failure do not control adequately for transit level of service. Studies concluding that transit level of service is of paramount importance do not control for urban form or socio-economic variables. This study is an attempt to bridge gaps between the urban structure-oriented and the transit level-of-service-oriented studies by controlling for both sets of variables. It does so by analyzing transit patronage between any two points as a function of how well transit and automobiles connect the two points, population and job densities of the two points, income and transit dependence characteristics of the two points, and design features of the two points. The degree to which transit connects two points is the control for transit level of service. Density, urban design, and socio-economic variables control for various aspects of urban structure. A useful finding of this study is that it is possible to distinguish between the potential of an area, such as a census tract, for generating transit traffic and the traffic that actually is generated. Because of this distinction, we are able to address the following four questions: 5

PAGE 16

1 What is the range o f potentials of suburban census tracts for producing or attracting transit traffic? 2. How well does transit tap the potential that exists? 3. How much transit traffic could be generated from t r ansit oriented developments and typical post-World War II suburbs at different levels of transit connectivity to the rest of the region? 4. What census tract characteristics are related with transit potential? The method that we developed is based on data that t ransit systems can collect; namely, transit origin-destination trip tables derived from onboard surveys The survey used here was performed for Sacramento Regional Transit by Evaluation and Training Institute in May and June 1989 (Evaluation and Training Institute, 1989). The method is a direct demand model that distinguishes between potential and observed transit trips in various parts of the Sacramento regional area. The study has important policy implications. Many suburban areas produce and attract few transit trips, but it is unclear whether the small number reflects lack of potential demand or lack of service. Our study attempts to show which it is and to highlight suburban areas where potential transit demand exists but where there currently is little transit ridership. A suburban tract may lack potential demand, even if a high level of transit service were offered. Improvements to transit service i n such areas would waste resources. On the other hand, potential may exist for transit traffic, but the service offered to the tract is inappropriate to demand, and t raffic does not materialize. Service improvement might be warranted. The key to making such a determination is to distinguish between potential and observed transit traffic. This study does so by analyzing observed flows of transit traffic between pairs of census tracts in relation to the quality of transit and road service linking them. Where transit service is bad, but moderate flows of transit trips occur anyway, there probably is a large potential for additional transit traffic. Where service is relatively good, but little traffic materializes, potential is probably lacking. Through a systematic analysis of transit traffic in relation to service, this study Infers the potential of most of the tracts in Sacramento County for producing and attracting transit trips. This method contrasts to three other approaches for determining transit demand. Disaggregate models of destination and modal choice often are used to infer what qualities of the choices influence the probability of persons of different socio-economic characteristics for making a choice. While such models are soundly based in utility maximizing theory, they by themselves cannot predict flows of people from one area to another. Being able to determine what causes such flows is a major concern of transit 6

PAGE 17

managers and policy analysts attempting to determine the welfare consequences of transit investments. I n order to obtain such insights from disaggregate choice models the user must combine them with models that the frequency of trip making from a given area. Another approach, the Urban Transportation Model ing System (UTMSI simulates regional road and transit systems for urban regions throughout the U.S. and has the ability to forecast transit traffic tor assumed transit systems I n suburban regions. We do not use the UTMS for three reasons. First, it assumes that trips are produced and attracted as a function of l and uses but not as a function of the transportation system. We argue later for a different assumpt i on. Second, it assumes that trips pJoduced and attracted are split between all available modes. We argue that total tri ps produced and attracted are the sum of modal trip productions and attracti ons and that each of the modes has a degree of autonomy in trip generation. Third, the UTMS lacks the flexibility to allow transit managers or other policy analysts to translate infor mation that they might have produced, such a on-board rid ership surveys, into insights about how the systems might be restructured to better serve Our method shows how ridership survey information can be used to the benefit of ttansit policy analysts. A third approach is to infer what soc i odemographic or land use characteristics produce or attract transit trips by regressing observed transit trip productions and attraction against such zonal variables. We do nol use this approach, because observed trip productions and attractions are Influenced as much by service offered as by potential demand that is derived from zo nal variables. Such an approach can y i eld little insight about what land uses i nfluence transit t r affic, unless the quality of transit service is controlled. The approach that we use, f irst inferring transit potential of zon es, does control for the qualitY of transit service. LITERATURE REVIEW Since the 1960s most theoretical wo
PAGE 18

doubts the ability to develop advanced theories of spatial human behavior and proposes instead mechanistic models of individual decision-making. Typically choice models are made useful for policy analysis by integrating them as one step into the urban transportation modeling system, or UTMS. Stopher and Meyburg (1977) provide a comprehensive overview The UTMS first predicts how many trips are produced and attracted in each zone of a system of zones blanketing a region This is done on the basis of regression or classification models for each of several trip types. The models are estimated on the basis of observed trips produced and attracted under controlled situations In relation to hypothesized explanatory variables. Trip productions of each type then are distributed to all possible destination zones using singly-constrained gravity models, with trip attractions as the measure of attraction. Choice models then are used to split trips from one zone to another by mode of travel based on relative generalized costs of each mode. Finally, trips for each mode are assigned to specific routes. The UTMS generally is not used for transit analysis except in the systems planning stage for r e gional rapid transit systems. The modeling system is cumbersome and requires huge data sets for estimation. Because of the cost of data collection. parameters for the various models in the UTMS often are borrowed from past time periods or from other regions. Transit managers often view it as a black box and do not understood where parameters for models in each of the four steps came from o r whether they are valid for use in a given applicat i on. The model that they are most interested in, the modal split model, generally has been estimated for some other city and is made to replicate observed transit traffic by manipulating the modal constants. Often times the modal constants are so large that any variations in service variables affecting generalized costs of transit or autos will have no impact on changing predicted t r ans i t traffic, causing further mistrust of the models There also are theoretical p rob lems with two underlying assumptions in the UTMS. One is that trip productions and attractions arise independently of the transportation system The other is that one must analyze transit demand as a share of total transportation demand. _Both are questionable based on empirical evidence. Pushkarev and Zupan (1977}, for example, show that as the auto was adopted in urban areas, its main impact was to increase dramatically the number of trips made in the urban areas rather than to merely substitute for transit trips (which it also did). Analysts of the decline in demand for U.S. intercity passenger trains came to similar conclusions (U.S., 1935). In 1920 average Americans used autos 50 miles per year for intercity travel, while they used intercity trains 450 miles per year. In 1930 the average American drove 1,691 miles per year in intercity travel but rode only 219 miles per year on trains. In addition. some also used intercity buses and a few used air services. Very few of the bus passengers came from trains. Most of the lost rail passengers had switched to autos. During the 1920s real per capita income remained static, suggesting that increased intercity travel did not come from increased personal budgets. It appears that the 8

PAGE 19

approximately 400 percent increase in intercity travel during the 1920s derived from the introduction of new modes of travel which also substituted in pait but not entirely for the pre-existing modes. Thompson ( 1 993) also shows that during the 1930s in California improvements to intercity bus service through lower fares and faster service had no impact on the growth rate of rail passenger traffic, except for the case of one train. During this period rail service was not improved but patronage grew because of the recovering economy. He also shows that later when rail service was improved, its patronage growth rate escalated substantially, but not at the expense of bus patronage. Virtually none of the incr eased bus passengers came from autos; some of the increased rail passengers may have represented diverted auto trips but it appears that most of the rail traffic growth was induced. Such anecdotal information suggests that the introduction of a new mode that lowers cost or times creates new traffic. There a l so may be traffic that is unique to inferior modes. Aschauer (1991) for the nati on as a whol e and Thompson (1994) for Florida both show that the presence of trans i t service has substantial positive impact on the growth rates of real per capita incomes. The effect i s much greater than the magnitude of the transit subsidies transferred into the reg i ons. For the nation and particularly for Florida the transit modal share is negligible, so it is difficult to infer why transit might have such an effect. A possible explanation other than an unindent i fied confounding variable is the exi stence of a unique market even for such an inferior mode as transit. The presence of transit for example may allow the existence of enterprises that employ transit-dependent people. All of this suggests substant i al modal independence with a limited amount of substitution between modes. It also suggests that trip generation is highly dependent on the qua lity of transportation service offered. Other non-modeling approaches have been followed in attempts to determine transit potential. In path-breaking work, Pushkarev and Zupan (1977) studied the impact of residential suburban hous i ng unit density, the absolute magnitude of CBD floor space as a surrogate for jobs, distance, and quality of transit service on the patronage of sing l e transit routes connecting suburbs and CBDs. They concluded that direct and moderately frequent transit routes could not attract significant patronage unless both residential densities on the origin end of the trip and job densities on the destination end of the trips were at least as high as those found in streetcar-era neighborhoods and CBDs of mid sized U.S. cities. They a l so f ound that patronage fell off quickly as distance between CBDs and suburbs increased, primarily because with increasing distance, suburban residents tended to travel by auto to destinat i ons other than CBDs. An important implication of their work was that transit could not attract much patronage in most suburb-to-suburb markets, because neither residential nor employment densities were sufficiently high. In such situations, fixed routes with acceptable directness and frequency would attract so few patrons as to be unaffordable to governments supplying service, or they would run up huge deficits. Jones and others attribute transit initiatives in the suburb-to-suburb market 9

PAGE 20

as one cause for escalating transit operating costs without commensurate gains in ridership (U.S. Department of Transportation, 1987; Jones, 1985). At the same time that Pushkareli Zupan published their work, Thompson (1977) proposed that the design of transit networks explained greater or lesser t r ansit system effectiveness in suburban environments. He demonstrated that grid and timed transfer system designs could serve dispersed trip origins and destinations, while they also could serve suburb-to-CBD travel. They also could be affordable, because they required levels o f service about the same as systems offering good suburban-to-CBD orientations typical of many transit systems Using a modal choice model from San Diego, he showed that multi-destination systems could be expected to attract much greater patronage than CBDoriented systems in similar environments. Finally, he argued that transit systems that adopted such multi-destination route structures achieved much greater patronage than those that did not. However, his latter argument did not constitute a rigorous statistical analysis, and in particular it did not take into consideration density, urban design, or socio economic variables that might have accounted for differences in transit patronage among transit systems in various urban areas. Several studies since 1977 have focused on Pushkarev's and Zupan's density arguments. They critize the analyses supporting the density arguments for their coarse spatial aggregation as well as for failing to consider variables. They offer counter arguments that density is a proxy for other variables that are more important influences on transit patronage (Handy 1992; Cervero, 1994}. Studies of individual travel behavior have shown that income, gender, household composition, and other socio economic variables are important determinants of travel behavior and transit usage in particular (Hanson and Schwab, 1986}. Low income and pedestrian-oriented neighborhood design in particular are thought to be the true determinants of transit ridership, and these variables tend to be correlated with higher densities. I f one designed studies to control for density, this point might become obvious. Cervero (1989) found that job densities and absolute magnitudes of jobs in evolving edge cities exceed thresholds established by Pushkarev and Zupan, but transit ridership to most edge cities is nil Free parking, poor transit service, but most importantly, inappropriate physical design characterized by isolated clusters of single uses separated by multi-lane arterials, freeways, and vast parking lots inhibited almost all travel other than that by single occupant autos, Cervera found. Cervero's work is consistent with that of Muller (1986} and the U.S. Department of Transportation (1987}, which suggest that transit can attract ridership only in areas that have characteristics similar to neighborhoods and business clusters that originally developed when walking or streetcars were the dominant modes. Areas that developed in the post-World War II era have inadequate design qualities in terms of pedestrian linkages and mixed uses to make transit attractive, even if densities are sufficiently large. 10

PAGE 21

Perhaps influenced by such findings, a movement has sprung up advocating street car-era neighborhood designs, characterized by mixed land uses and direct, non-auto intimidating pedestrian paths, in order to influence residents to travel by other than auto modes. One of rationales for what has become known as the nee-traditional neighborhood movement is to make transit viable again. Peter Calthorpe, a leading proponent of this view, has built a suburban development, Laguna West in Sacramento, based on nee traditional principles, which he calls transit-oriented design (Calthorpe, 1993}. Cervero (19941 subsequently conducted a study of the impact of neighborhood design on transit user behavior for work trips, divid ing his study into two areas. In each area he compared travel behavior in several pairs of neighborhoods that were alike in terms of income and population density, but which differed in terms of neighborhood design. In doing so, he tested the impact of neighborhood design on transit user behavior. He also tried to control for transit leve l of service by choosing pairs of neighborhoods that had similar transit service; the variable t hat he used was the number of transit bus or car miles per unit of area in the neighborhood. Somewhat contrary to his earlier hypothesis, Cervera found that design had only a weak influe nce on transit ridership i n one of his study areas but no statistically important influence on transit ridership in the other area. However, neighborhood design did influence the tendency for people to make work-oriented walking trips in both areas. In reviewing the studies critical of Pushkarev' s and Zupan's conclusion that density is an important determinant of transit user behavior, we note a striking gap. None of the studies proved their original hypotheses that variables other than density are important other than at a very marginal level. On the other hand, they did not test the counter hypotheses that either residential or job density are important, once design and income are controlled for. Thus, it still is possible that density, both residential and job, is an important determinant of transit user behavior. We also note that neither Pushkarev and Zupan nor the subsequent studies controlled for transit level of service adequately. Doing so is important, because as Thompson ( 1977} notes, a bus mile in a suburban context can be of greater or lesser relevance to a suburban resident, depending upon where the transit route goes and to what other transit routes it connects. A finding that n ee-traditional design has no impact on transit user behavior might merely reflect the presence of irrelevant transit service rather than the unimportance of design. Many of the traditional neighborhoods studied so far in fact have only limited transit service. While it has not been included in studies referenced, an extreme example is Calthorpe's Laguna West, the only example of new transit oriented development so far in the United States. Laguna West has no transit service at all. A test of the influence Laguna West's design on transit user behavior would be meaningless. Even in suburban environments where transit is present, transit typically runs to a CBD. In such situations measures of transit service, such as transit vehicle mi l e 11

PAGE 22

density are meaningless if most residents do not wish to travel to CBDs, as it is well known that they do not. Pushkarev and Zupan controlled for transit service more carefully, but their study focused on suburbto-CBD markets served by single routes. They did not look at transit potential of multi-destination route structures in suburb-to-suburb environments. This may be an oversight, in that Thompson 11g931 shows that the traffic-accumulating or s cope economy characteristics of multi-destination route structures compared to single destination route systems can produce profitable passenger densities in sparse markets where single-purpose systems cannot survive. To better test variables influencing transit ridership potential in suburban environments, a variable is needed which captures the im portant quality of transit service: how well it potentially could take peop le from a given zone to destinations to which they might want to travel. The new variable should be used in conjunction with variables describing residential and job density as well as urban design, so that the influen ces of all three types of variables can be measures. This study attempts such an approach. I n it we propose a theoretically sound, but simpler mode ling method whose parameters transit systems can esti mate themselves, giving them greater understanding of the modeling system that they use. The model incorporates the idea that transit trip generation is dependent on changes to the quality of transit and auto service. It does so using measures of transit connectivity to places where transit patrons want to go. It also is based on the idea of modal independence in trip generation, although it incorporates cross elasticities with respect to auto service. MODEL SPECIFICATION Our approach is dictated in part by the nature of dependent variab le that we have available to us. This is an on-board transit survey that yields counts of transit trips between census tracts. The counts are potentially segmentable into trip type and trip time, though we use aggregated 24hour weel<. d ay counts in the study reported here. We do this, because when we began estimating our model, the estimator that we used could not estimate data sets with a large number of zer os for the dependent variable. The aggregated 24-hour flows has a sufficiently large number of non-zero observations to be estimatable; lesser aggregations were not estimatable. New estimators now are available that can handle ultra-sparse dependent variables. The aggregated count data preclude the use of dlsaggregate models derived from utility maximizing c hoice theory. The estimation of choice models requires data sets of choices made by individuals between potentially usable modes. Because our survey data is not of this type, we may not infer how transit demand arises from individual decisions. 12

PAGE 23

We will have to content ourse l ves w ith inferences about how aggregate zonal characteristics influence po t ential transit demand. A model that is consistent with the avail able data is the doubly -constrained g ravity model. Th is form can b e Inte rpreted i n a way that distinguishes between actual and potential trip production s a nd attractio ns It also can be e stimated aggregate count data. Its most general form derived from entropy theory follows (Stopher and M eybu r g, 1975, pp. 23746): where the constants E, and are defined as: ""=[L A :E e x ( -"" ')J '"' j P "1 I and where T 0"' trips between I and j by mode k and person type n; Po" = the numbe r of origins at i of people of type n; M{nl = the set of modes available to type n people. = attractions at j; and, E q 1 Eq. 2 Eq. 3 exp( fJ"*c !"l a generalized costs for persons of type n by mode k between i and j. This fonn rese mbles Williams' derivati o n of the j oint destination/mode sp lit mode l f r om pr i ncip l es of utility max imizat i on (Williams 1977) It differ s only in its less r estrictive cost specification 13

PAGE 24

Equations 1 through 3 may be interpreted in a way that distinguishes between actual and potential trips productions and attractions. P, equals actual trip productions from zone i and A1 equals actual trips attracted to zone j E is a constant for zone i. It is balancing factor that often is ignored but which is in reality an connectivity measure for zone i with respect to attractions in the rest of the city, as some theorists have noted (Stopher and Meyburg, 19751. Similarly B1 is a constant for zone j that measures zone j's connectivity to attractions in the entire system. The quantities P;"/E1 and AlB1 are constants that we will call respectively PDN." and ATN1 These quantities, we argue, may be interpreted as the trip production potential of zone i and the trip attraction potential of zone j. To see this, we substitute PDN," and ATN1 into Equation 1, 2, and 3, yielding Equations 4, 5, and 6: Tu'"=PDN, ATN1 exp( -/J" c,'). Eq. 4 Eq. 5 Eq. 6 By using dummy variables for PDN," and A TNp specifying the cost function, and obtaining data on the variables in the cost function and on travel between all zonal pairs in the system, one may estimate Equation 4. When using dummy variables, one must leave the variable for one zone at 0 to prevent over estimation. The estimated coefficients for the remaining zones indicate potential productions or attractions with respect to the reference zone. Equations 5 and 6 confirm the status of E, and A1 as connectivity indices to trip production or attraction for each zone in the system. 14

PAGE 25

We now consider characteristics of PDN;" and A TN1 that are useful for interpreting the model. We illustrate the argument with product ion potentials; the reader may construct analogous arguments for attraction potentials. Equat ion 7 summarizes the definition of production potentials : Eq. 7 We rewrite Equation 7 in terms of summations of trip interchanges and connectivity to all tracts in the system in Equation 8: Eq. 8 It can be inferred from Equation 8 that the model can be estimated with a representative sampl e of tracts drawn from the total tracts for a region without affecting the estimation of PDN1 A sample will reduce the magnitude of trips produced in tract i (shown in the numerator of the right hand side of Equation 8). but it also will reduc e the connectivity of tract i. The relative reduction of the numerator to the denominator will vary from one zonal pair to another, but a random sample of zones either chosen or deleted should insure that rati os higher than one will ba l ance ratios lower than one, leaving the estimates of potential unaffected. A corollary is that the estimation of PDN is i ndependen t of both connectivity of zone i to the reg ion as well as to total trips produced i n zone i. This is suggested by the observation that trip productions and connectivity In Equat i ons 7 and 8 are functions of sample size, whereas PDN, is not. This is an important point. In the usual app licati ons of transportation models P;" and A1 are considered as exogenous to the model and are functions of the socio-econom i c characteristics of people and activity in zones i and j. Better or worse connectivity from zones i and j has no bearing on trip generation. Such reasoning runs counter to our argument in the preceding paragraph. It also runs counter to the underlying assumpt ion of transportation demand as derived from socio-econom i c interaction H igh cost travel should 15

PAGE 26

manifest itself in high cost socio-economic interaction, and less of It should happen as a consequence, manifesting itself in less travel. Less travel should manifest itself in fewer trip productions or attractions as well as in shorter trips. Certainly the historical examples that we exam ined suggest such to be the case. Such reasoning leads us to conclude that P," and A; should not be considered exogenous. Rather, PDN," and ATN1 should be considered exogenous and as functions of characteiistics of zones i and ;. Our argument i n the preceding paragraph supports this interpretation. If we consider PDN;" and A TN1 as exogenous and as functions of characteristics of zones i and j, we see that actual trip productions and attractions can be derived from them through the connectivity of each zone to the entire system of zones Equations 9 and 1 0 show the relations: Eq. 9 Eq. 10 If connectivity improves, these relations predict greater productions and attractions from each zone. If i t worsens, they predict fewer productions and attractions. Connectivity for a g iven zone could change by changing the quality of any of the modes of transportation linking the zone to the rest of the system. It also could change from changes to the production or attraction potentials of other zones Such model behavior is consistent with the underlying theory of transportation demand that Is derived from the demand for socio e conomic interaction. The model thus d istinguishes between, on the one hand the potential for trip productions a_nd attractions, and on the other hand actual trip productions and attractions. It also depicts actual productions or attractions as the product of potential productions and attractions and connectivity. Despite the desirable qualities of the model depicted in Equations 4, 5, 6, 7, and 8, which we will call Model I, we need to modify it to take into account our limited trip interchange data. We have only transit travel between all pairs of zones; to estimate Model I, we need trip interchanges by all modes. If we estimate Equation 4 with transit only trip interchange data, changes to Equations 5, 6, 7, and 8 are implied, and the result 16

PAGE 27

is what we call Model U Model U is shown in Equat i ons 11, 12, 13, 14, and 15, where the superscript t denotes tran sit productions and attractions either potent ial or actual: Eq. 11 Eq. 12 S.'J( PDN ,'" exp ( -{J" C.'), Eq. 13 where, P t n_PDN"' E I I I Eq. 14 Eq.15 In Equation 11 the PDN, of Equation 4 becomes PDN.'", which represents potential transit trip productions in zone i by person n. ATN1 becomes ATN1', which represents potential transit tri ps attracted to zone j. Although structura lly similar the equat i ons in Modal U depict a d ifferent assumpti o n about travel behavior than those i n Mode l I. Th i s can be seen by summing T." for bot h models across all modes contained i n the set of modes M(n ) Doing so achieves total tri ps between all i and j, as shown i n Equation s 16 (for Model I) and 17 (for Model II) : 17

PAGE 28

Eq. 16 Eq. 17 Equation 17 differs from Equation 16 in that its total potential trip productions for a zone are the sum of the modal potential t rip productions. Model I is consistent with the theory that demand for total travel first arises, in part influenced by the connectivity offered by all modes of travel, and then total travel potential manifests itself in actual travel on each mode in proportion to that mode's imped ance. Model II suggests the idea that demand for travel on each mode arises independently of each other in response to a potentia l trave l for that mode and the connectivity that the mode offers throughout the system. Whether or not the idea in Model II can rep l icate empirical evidence on the behavior of transportation systems depends upon whether its impedance function can be specified to reflect modal substitutability. As Stopher and Meyburg (1975) show, the impe dance function is der i ved from time and cost constraints of people living in zones where tri ps are produced. Impl icit in the latter model is the idea that there is a unique cost constraint for eac h mode. Such modal uniqueness appears to prec l ude modal substitutability. In contrast, Model I contains a cost constraint for travel in general, allowing for substitutability between modes Empirical results suggest that some modal substituti on occurs implying that the former specification is superior. Howeve r, it is superio r only if date are available to estimate it. Since such data are not available to us and generally are not available to transit systems the question is whether we can specify the i mpedance function of Model II, which can be estimated with the data we do have, to approximate modal substitutability. We do so with the following reasoning. The total cost and time constraint for a given mode, such as transit, represents the maximum resources that actually are expended on transit from a given zone The better the transit connectivity and the worse connectivity offered by other modes from all zones, the greater are the resources expended on transit, presuma bly at the expense of other modes Such behavior can be reflected i n the impedanc e function for transit by including variables for competing modes as well as for transit describing time and cost expenditu res between pairs of zones. I f flows of travel are availab le from only transit to estimate a model, the second mol:le i with 18

PAGE 29

this interpretation of the impedance function shoul d replicated empirical transit travel behavior satisfactorily. In fact, the Model II may represent reality just as well, or even better than Model I, even if system-wide flow data were available for estimation. If such data were available, one could estimate a separate model for each mode as in Equation 17. Each model would have an impedance function reflecting the service offered by all modes. If a priori expectations were met in the estimation of the models, the self elasticities for each mode would be negative and the cross elasticities would be positive. The total cost constraint would be the sum of those from each of the i ndividual modes. An improvement in the service offered by one mode would be reflected by the model in greater trip productions and attractions by that mode and lesser productions and attractions for competing modes. As an example, we consider the case of densification. It is well known that denser development produces lower trip generation rates per capita, even if socio-economic va riab les such as income and household size are controlled for. This result most likely derives from lower rates of auto usage as driving becomes more difficult in restricted environments and as higher auto storage and parking costs depress levels of auto ownership for given levels of income. Traditional transportation modeling cannot account for this phenomenon endogenously, but Model II can. We can imagine several zones in a system that are allowed or made to density. According to Model II, densification would have the effect of increasing the trip production and attraction potential for each of the zones for all modes. We will consider only auto and transit. In the case of auto, increasing potential will manifest itself in greater auto productions and attractions, but greater auto traffic will increase congestion, which will decrease the auto connectivity of the zones to the rest of the system. lesser connectivity will depress auto productions and attractions per capita, even as they increase in absolute numbers. At the same time, because it is a component of the transit impedance function, lower auto connectivity will manifest itself in greater transit connectivity, which will stimulate transit productions and attractions. (Of course, greater congestion also will worsen transit service in the absence of transit priority, but typically auto level of service falls more precipitously i n congested environments than does transit.) The rates of transit trip productions and attractions will increase with densification, and the absolute number of transit productions and attractions will increase rapidly. The net result of Model II' s treatment of densification would be forecasts of lower per capita auto productions and attractions and greater per capita transit productions and attractions, a significant increase in transit modal split, but probably a decrease i n overall trip productio n and attraction rates All of this is what one would expect in reality. On the other hand, if road capacity were expanded in this environment, the model would predict falling transit and incr easing auto trip p roductions and attractions. This again meets a priori expectations. 19

PAGE 30

MODEL ESTI M ATION We now turn to the task of estimating Model II (Equation 11) with Sacramento data The dependent variable is 24 hour transit travel between census tracts. As discussed earlier, we do not distingu i sh between socio economic categories in each census t ract, and thus we sum over all n soc i o economic categories in the earlier equations. T he dummy variables representing PDN and A TN1 a r e treated as shown i n Equations 18 and 19: Eq. 18 Eq. 19 where there are p production zones and q attraction zones in the sample, and U and v are 1 if p = i or q = j and 0 otherwise The dummy production parameter to be es t imated is p,, and in a given zone i, PDN1 = exp(p1). The dummy attraction parameter to be estimated in a . For a given zone j, ATN1=exp(aJ. The variables that we specify for the friction expression in Equation 11 are those typically used in general cost specifications: doorto-door t r avel time of both trans i t and autos. Transit fares also would be desirable, but because the system has a flat fare with free transfers, a fare variable lacks the variance to show its influence on transit patronage. We a l so include a variab l e to denote whether the existence of rail trans i t influenced the magn i tude of transi t usage between a pair of tracts, all else equal Sacramento Reg i onal Transit operates mostly buses but does have one light rail route The friction function that we used is shown in Equation 20: Eq. 20 where HTIM. i s door to door highway time from tract i to tract j; TTIM; is door to door transit time from tract ito tract j; 20

PAGE 31

and RPAXij describes the proportion of transit passengers from i to j who make use of rail for at least part of their journey. The terms t,, t2 and t, are coefficients to be estimated To obtain transit and auto times and highway distances between census tracts, we created computer models of both the transit and road systems in Sacramento County and adjacent parts of Yolo County. We used QRSII, because it is suitable for use with large zones For transit, It allowed us designate which parts of census tracts are served by different transit routes (and those parts that have no service at all). When computing the transit travel time between a given pair of census tracts, the computer finds the shortest transit paths from one of the designated pieces of the first census tract to each of the pieces served by transit in the second census tract. The transit travel times are door, including walk time, wait time for the first bus or tra in, transfer time to another bus or train, and walk time to the destination. 2 Once the computer has ident ified shortest paths from one piece of the f irst census trance to each of the pieces of the second census tract, it then goes to the second piece of the first census tract and repeats the process. It does this until all pieces of the first census tract are covered. The transit travel time between a given pair of census tracts is a weighted average of the times over all of the paths between the two tracts. We modeled the freeway and arterial road system with centroid connectors to yield free-flow auto times between all census tract centroids. The transit system that we modeled consists mostly of radial local bus routes operating with base headways of 30 or 60 minutes to the CBD. There also is a peak period only freeway and arterial express bus service focused on the CBD. Several years ago system planners improved transit service in the southern part of the city by focusing several local and express routes on a timed transfer center at Florin Mall. Over the two years prior to the survey, the system introduced i n stages a light rail line running from the northeast to downtown and then back to the east. Planners restructured bus service in the territories served by the light rail to focus on timed transfer centers at several of the suburban r ai l stations. They a lso introduced a long local bus route connecting the two outer rail terminals. This runs on a 15 minute base headway via important strip commercial areas and the reg ion's largest mall, Sunrise. Finally, two north-south cross-town bus routes run on 30 minute headways and connect into timed transfer centers on the north and south. 2Transit travel time is equal to 1.3*WALK TIME + 0.95*WAIT TIME plus 8 : 4 minutes wait time penalty + O.S*TRANSFER HEADWAY (omitted for timed transfers) + a transfer penalty of 23 minutes for an untimed transfer or 1 2 minutes for timed transfer + INVEHICLE TRAVEL TIME. We coded the transit network with untimed transfers with the following exceptions, made In accordance with the system's planners: Florin Mall in the southern part of the city (a timed transfer center for several local and express bus routes) and several major suburban bus/rail stations 21

PAGE 32

Combining Equations 18, 19, and 20 with Equation 11 summed over all n socioeconomic categories, we obtain: Eq. 21 Equation 21 is the model that we estimated The estimating technique, Poisson regression, was appropriate to the nature of the dependent variable. T;;' is greater or equal to 0, with a great many of its observations being 0. It has a low mean, but some values are high. Poisson regression often is used to estimate models predicting variables exhibiting such skewed distributions (Fiowerdew, 1991 ). While Poisson regression is valid only if the mean and the standard deviation of the dependent variable are equal, Davies and Guy 11987) modified the technique for situations where the standard deviation of the dependent variable greatly exceeds its mean, which often is the case in transportation and migration applications. Our study area contains 88 census tracts numbered from 1 to 1 02. Missing numbers are tracts not in the urbanized region. We created 88 production dummy variables labeled for the numbers of the census tracts lP1 to P102l.' We also created 88 attraction dummy variables lA 1 to A 102 with some intermediate numbers not in the urban region). These added to the transportation variables resulted in 179 variables. Unfortunately, the LJMDEP estimating package that we used could accommodate a maximum of 150 variables, so we eliminated 17 census tracts (accounting for about 12,000 out of the system's 44,406 daily linked trips) from the analysis, resulting in the elimination of 34 variables. 4 Our final data set includes 145 variables and 5041 cases. Multicolinearity was not high enough between any of the included variables to cause difficulty In interpretation of results. However, HTIM has a correlation coefficient with highway distance of .973, which does pose problems in the interpretation of the coefficient for HTIM. RESULTS "There are no census tracts 83, 84, 85, 86, 88, 92, 93, 94, 95, 96, 97, 98, 99, or 1 00 in the metropolitan area. 4The deleted census tracts are 24, 41, 43, 46, 51, 58, 70, 72, 74, 77, 79, 80, 82, 87, 90, 101, 102. 22

PAGE 33

With Poisson regression, one estimates goodness of fit by first r unning the model with only the constant included and then running it a second time with the desired explanatory variables. The percentage reduction in the likelihood ratio caused by the addition of explanatory variables is an indicator of the model's explanatory power; in this case 47 percent. The overall fit is not bad, considering the lumpiness of the dependent variable. The lumpiness derives from Tij's creation through expansion factors from the survey sample. Expansion factors were as high as 50. This means that only one passenger may have been observed traveling between a pair of census tracts, but he or she may have been expanded up to 50 trips. On the other hand, if 0 passengers were observed traveling between a pair of census tracts, 0 is what is shown for that observation in the transit flow variable. This comparison implies that one observation in the flow variable may show 0 passengers, and another may show 50 passengers, but the difference in the survey may have been only between 0 and 1 passengers. As discussed earlier, Poisson regression under-estimates standard errors when the variance of the dependant variable exceeds the mean. For Til' the variance exceeds the mean by about three-fold. To correct for this, we ca1culated a factor of 5.85 for each reported standard error in the manner recommended by Davies and Guy (1987). In other words, we obtain a reasonable approximation of the !statistic for each coefficient by dividing the reported t-statistlc by 5.85. After modifying the t-statistics with the reduction factor explained above, we found that the coefficients estimated for the transit variables are statistically significant and have the expected signs (Table 1 ). The highway time variable is insignificant and has the wrong sign. 23

PAGE 34

Table 1 Transportation System Variables in the Equation (the dummy variables for the census tracts are not shown) Variable B Modified tstatistic ---------------TTIM -.0161 -5.45 HTIM -.0046 -0.45 RPAX .5794 5.32 constant 3.4331 7.86 Table 2 presents the elasticities of transit traffic with respect to each of the explanatory variables at mean values. Maps 1 through 6 show the distribution of transit trip productions, attractions, transit connectivity to product ion potentials (Ei), transit connectivity to attractions potentials (Bj), and the ratios of actual trip productions and attractions to potential productions and attractions. Table 2 Elasticities of Tij with Respect to the Transportation Variables Variable Elasticity at Means TTIM -1.807 HTIM .061 24

PAGE 35

DISCUSSION OF RESULTS Tables 1 and 2 The parameter for transit travel time has the correct sign and is highly significant. Transit traffic also i s highly elastic with respect to this variable. A one percent reduction in transit travel between a pair of census tracts having average travel time will increase transit traffic between the affected census tracts by 1 .8 percent. These results suggest that policies to reduce transit travel time should yield subs tantial increases in transit traffic for those pairs of census tracts where there is substantial transit production and attraction potential. The parameter for RPAX a lso is highly significant. Its sign shows that to the extent that passengers traveling from one census tract to another need to use rail in whole or part, they will be more l ikely to use transit, all else equal. The magnitude is substantial: the presence of rail increases transit traffic by 78 percent over an all-bus path. This result contrasts with observations made in many policy analyses that find that rail transit investments have had little o r no impact on transit ridership A possible explanation that is consistent with our results is that the expense of rail requires reductions in bus service in other parts of the service territory, the loss of which reduces system-wide patronage more than the public appeal of rail increases it. The real issue here is the expense of rail; while its appeal is 78 percent greater than buses, its cost typically is much greater than that. These considerations suggest that rail should be supported only where its cost, particularly operating cost, can be kept low in comparison to buses providing similar service. The parameter for the highway travel is insignificant and has the wrong sign based on our a priori expectations. There may be two explanations for this counter-intuitive result. One is that highway travel time is collinear with road distance. It may be that transit travel is negat ively affected by distance. To test this possibility, we respecified the transit impedance function to separate times into their constituent parts : speeds and distance. We included transit speed, highway speed, highway distance, and RPAX. Speeds were door-to-door travel times divided into highway distance. In the respecified model, which eliminated the problem of multicolinearity, transit speed, highway distance, and RPAX all were highly significant with expected signs (positive for transit speed and RPAX, and negative for distance) Highway speed also was significant, but barely so at the five percent level, and it had the expected sign (negative) We did not use this specification because a reviewer argued that the use of speeds and distances rather than times resulted in a non-linear impedance function that was not readily interpretable in terms of generalized costs upon which the entropy model is based. 25

PAGE 36

The counter-intuitive resu l t also may stem f r om our use of free-flow highway times that do not adequate l y take into account the di fficulty of driving i n congested areas. We suspect that had we coded a congested peak period highway network, auto travel t i mes would not have been as collinear with distance and would have been positively and significantly related to transit traffic. Because of this result, we i ntend to use a congested highway network for a similar project that we are working on in Orange County, Florida. Estimations of Trip Production and Attraction Potentials The trip production and attraction coefficients for each census tract are the natura l logs of each PDN1 and ATN 1 Thus, each PDN1 and A TN 1 is the exponential of the appropriate estimated coefficient. A coefficient which is statistically ins i gnificant means that the census tract that i t represents has transit production or attraction characteristics similar to Sacramento census tract 15, which is shown in Map 1 Census tract 15 lies east of the central area of Sacramento and has characteristics of traditional streetcar suburbs : most development dating from before 1940, relative l y narrow, tree-lined streets, a grid street pattern, substant i al population and employment densities, moderate i ncome, a mix of housing types, and a m i x of land uses Tract 15 also has very good bus service linking it to the Sacramento CBD (tracts 7, 10, 11 as defined by the Sacramento Council of Gove r nments). Production or attraction potentials between 0 and 1 i ndicate that the relevant tracts have potentials l ess than tract 15; tracts with potentials greater than 1 indicate potentials greater than those for tract 15. Maps 2 and 3 show the distributions of respectively transit production and attraction potenti als in Sacramento Production potentials are spread over the Sacramento region much more evenly than attraction potentials and appear to be 5 A light rail line a lso runs along the southern border of the tract, but it has no stations in the vicinity of the tract. 26

PAGE 37

Map 1 Census Tract Id e ntifiers (follows this page) 27

PAGE 38

Map 1 -Census Tract Identifiers 81 75 67 65 761-----. 59 7 8 62 39 55 56 I 57 91 52 32 34 4$"'"'""-+---t 33 48 42 49 50 0 I 2 3 4 5 6 1 apprx. 2.7 mi.

PAGE 39

Map 2 PDN1 Distribution ( P e r cent) (follow s this page\ 28

PAGE 40

..... 1-o ...., (I) ..... Q 00000 "' "'--oo Q-Nft")'4'" qc::c::c::-: 00000 vvvvv 00000 o...o--o N oo-. Ntl"') 00000 00000 -

PAGE 41

Map 3. ATN1 Distribution (Percent) (follows this page) 29

PAGE 42

1:: 0 .... ..., ;:I ,D .... !l (/) .... Q N -E .... .... .. "" "" ..

PAGE 43

more influenced by the size of the census tracts. Higher income suburban tracts in the east, northeast, and southwest show considerab le potential as do some inner city tracts. To control for the effect of great si:ze in the census tracts, it is tempting to analyze production potentials in terms of p r oduction potential per unit of area i n each census tract. Unfortunately, doing so would introduce ecological corre lation fallacies, invalidat ing inferences that might be made. We thus must make what inferences we ca n from the aggregate results shown in Maps 2 and 3 and try in the future to obtain data from a more finely and evenly grained zone system. The attraction potentials are much more concentrated in central Sacramento, particularly in the core of the CBD. Despite this, certain areas of suburban transit potential exist. Suburban tracts 16, 18, 26, 40, 47, 52, 55, and 81 all have attraction poten tials considerably greater than tract 15. Tracts 55, 47, and 81 contain reg ional shopping malls and related office activity, and tract 52 contains California State University, while tract 16 adjoins the unive rsity. Tract 18 is a closein mixed use suburb with heavy office employment. Tracts 26 and 40 contain no obvious trip attractors. Tract 26 is a streetcar era closer in mixed use neighborhood; tract 40 is a large, higher income post World War II bedroom suburb. Several other mixed use inne r suburban tracts contain attraction potentials similar to those of tract 15. Maps 4 and 5 show how well Sacramento Regional Transit connects each census tract to opportun i ties I n the remainder of the region. Map 4 shows connectivity from each tract to trip attraction potentials; the higher the connectivity the easier it is to travel to places that are attractive to transit users. For people desiring to depend upon transit, these would be good places to live. The downtown tracts, inner suburban tracts, and outer suburban tracts served by the light rail line or connected to it stand out with relatively high connect ivity in this regard Tracts served directly by light rail include 91, 52, 16, 13, 19, 12, 9, 8, 10, 11, 5, 69, 63, and 74 (one of the tracts deleted from the study). Several of these zones contain timed t ransfe r centers with buses radiating to nearby trip attraction potentials. Such stations are in tracts 91, 16, 17, 52, 69, 63 and 74. Of these, only zone 63 shows poor connectivity possibly because we miscoded it by not showing it as a timed transfer or possibly because its connecting bus lines do not go to tracts with high transit attraction potential Zones with good trunk bus service to the CBD in addition to good cross town service to regional shopping centers and universities, and other bus services to older mixed use close in suburbs also have high connectivity. These tracts include those extending in the ray running to the southeast of the central area. Tracts with po or connectivity are those that rely on peak period exp r ess bus serv i ce to the CBD with poorly developed ancillary bus service. They generally have poorly developed local buses to other destinations that are not well integrated with the express buses, or the local buses take too long to reac h large potential tri p attractions outside of the CBD. Such tracts lie in the southwest, south, and far northeast of the re gion. One 30

PAGE 44

Map 4. E, Distribution (Relative to Average) (follows this page) 31

PAGE 45

Map 4 -Ei Distributio n ( R e l a tive Av e r age) 10 00 0 75 1 25 I a p pr x 2.7 mi

PAGE 46

Map 5 BJ Distribution (Relative to Average) (follows this page) 32

PAGE 47

Qi' b.O 1-< Q> :> < 0 ...., Q> :> ..... ...., ro ..... v o:< '-' d 0 ..... ...., _g ..... .... ...., CIJ .... Q

PAGE 48

tract has poor connectivity, contrary to a priori expectat i ons. This is the timed transfer center at Rorin Mall in tract 4 7. Several hourly local bus routes converge here, feeding a peak period express route to the CBD as well as trunk service to the downtown and cross towns running in two directi ons. It may be that potentia l trans i t trip attractions reachable by transit are just too far away by buses that are t o o slow to register high on the connectivity index. Map 5 shows transit connectivity from each tract to trip production potentials. The higher the connectivity, the easier it is to travel from a g iven tract to tracts with the potential for produc i ng large numbers of transit trips. Tracts with high potentia l s would be good locations for employers or retailers desiring to have employees or customers travel by transit. Downtown tracts, light rail timed transfer stati ons, the F l orin Mall timed transfe r center, othe r zones in the southeaster ray, and middledistant zones in the near northeast connected to light rail do well in this regard. Zones perfo r ming poorly again are those in the southwestern and far northeastern parts of the city. Maps 6 and 7 show whether a given zone's share of total productions or attractions exceeds, equals, or is less than the zone's share of total production or attr action potential. Here "total" refers to the sample of zones in the study. A z one's share of productions or attractions exceeds its share of potential when the zone has greater than average connectivity to the zones in the sample Conversely, its share of productions or attractions fall short of its share of potential when it has less than average connectivity. Thus, Maps 6 and 7 show the inverse of the information in Maps 4 and 5; however, t o better illustrate where improved transit service could yie l d substantial suburban traffic increases, we show only the more important zones in Maps 6 and 7. Only nacts with production potentials 9 or higher and attracti ons potemials of .7 or higher than tract 15's potential are shown. Maps 6 and 7 can be interpreted as showing Regional Transit's market penetrat i on relative to its average level of service. It achieves greater tl]an average o r average market penetration in the central area, in inner suburban zones, at timed transfer stations on the light rail line and at zones in the southeastern ray. It achieves less than average market penetration in the southwest, far north, and far northeast. Regi onal Transit could obtain greater traffic in the under-performing suburban zones by increasing transit c onnectivity in those zones to levels achieved in some of its other suburban zones. It also could i mprove its overall average level of connectivity, in which case it would obtain more traffic from all zones in the city. The degree to which transit traffic could be obtained from different types of suburban zones through improved connectivity is shown In Figures 1 and 2. Figure 1 shows trip productions; Figure 2 shows trip attractions. In Figure 1 the lower line represents trip production potential of a typical post-World War II suburb of about one mile square The upper line represents the trip production potential of tract 15, which also is 33

PAGE 49

about one mile square and which has characteristics of transit oriented development. The range of accessibilities cover those in 34

PAGE 50

Map 6. %P/%PDN1 (for those Tracts with PDN, Greater Than .9) (follows on next page) 35

PAGE 51

"' "' ... e .., N ,.-... 0) .. .. ... .. Ill N ..Q E-c .. 1-4 Q) 0 ...., Ill Q) "" 0 ..... z Q p.,. ..Q ...., ..... Ill ...., () Ill 1-4 E-c Q) Ill 0 ..Q ...., 1-4 0 .... '-' .... z Q p.,. '-..._ ..... A. I co Ill

PAGE 52

Map 7. %Ay%ATN1 (for those Tracts with ATN1 Greater Than 7) (follows on next page) 36

PAGE 53

. ....., z E-< < .... 0

PAGE 54

Figure 1 : Transit Trip Productions as A Function of Zone Potential and Connectivity (follows this page) 37

PAGE 55

F i gure 1 Transit Productions from Zone Potential and Accessibility 1,200 -,----------------------'---------, 960 I ----,_ 1:-:!: t/) c: CIJ t= 720 +-----. -----= --CIJ ----. -0 g> 480 r---------.... CIJ c: ---------------1 C) 0 240 -1------------------------1 0 200 400 600 800 Transit Accessability to Attractions ,._ Suburb Production Potential TOO Production Potential

PAGE 56

Figure 2 : Transit Trip Attractions as A Function of Zone Potential and Connectivity (foll ows this page) 38

PAGE 57

Figure 2. Transit Destinations from Zone Potential and Accessibility 1,200 ..,.------------------------------, :g_ 960 c: 1--"ii) c: ------------------------!!! 720 1-g> 480 "" {!!. 240 0 0 , _______ ___ .. _____ ,_. -----' ---200 400 600 800 Transit Accessibility to Productions *"" Suburb Attraction Potential E8-TOO Attraction Potential .,_ Streetcar Suburb Attraction Pot.

PAGE 58

Sacramento for the 70-zone sample. It is important to keep in mind that transit connectiv ity in Sacramento is about 37 percent greater than shown, because the transit syste m connects to census tracts left out of the sample. Thi s means that actual transit trips that could be produced or attr acted would be about 37 percent higher for each land use configuration than what is shown in Figures 1 and 2. The highest of production potential is 704 and belongs to a tract in the CBD; the average for tracts in the sample is 338. Most suburban tracts have connectiv ity indices in the range o f 39 t o the low 400s. Those on the light rail line have somewhat higher indices. The highest Is subu r ban tract 69 with an connectivity index of 655. This tract has four light rail stations offering quick service to the CBD. It also has relatively high quality bus service running from a timed t r ansfer center on the rail line to a regional shopping mall a nd office emp loyment in adjoin i ng tract 55. Figure 1 indicates that transit connectivity similar to that in suburban zone 69 could produce roughly 200 trips p e r day in a typical post war tract but roughly 750 trips per day in a transit oriented development. Three lines are shown in F i gure 2 The upper line represents trip attraction potential of the mixed use tract 26, which is about a mile square and which has much greater employment than tract 15. The m i ddle line represents the trip attraction potential of tract 15. The bottom line repr e sents that of a one mile square post war suburb. The accessibilities are to trip production potentia l s. The h i ghest in Sacramento is 903 and is in the CBD. The average for the sample is 630. The highest suburban potentia ls are 733 for tract 52 and 726 for tract 17, both having timed transfer centers on the light rai l line. Tract 81 has an connectivity index to production potentials of 365. Figure 2 i ndicates that transit connectivity to productions similar to that of zone 52 could attract over 1,000 trips p e r day for a trans i t oriented development wit h dense employment. Tract Characteristics Associated with Trip Production and Attraction Potentials At this point we have abstract not i ons of trip production and attraction potentia l s for a sample of census tracts in Sacramento. In this section we regress the potentials against census tract characteristics to determine what characteristics are statistically related to the potentials. We specify PDN ; and A TN ; as linear functions of variab l es denoting dens ity, socio economic characteristics, and design qualities of the census tracts. Briefly, the socio econom i c variables that we chose include POP (tract population). EMP (tract emp l oyment), FIFTEEN (proport i on of tract households earning less than $15,000 per year), PERSERV (the proportion of tract workers employed in service j obs), POPROOMS (number of tract residents per room). SIXNOCAR (proportion of tract residents over 60 and not owning a car) SIXTEEN (proportion of tract residents under 16), and VAPOHU (vehicles per occupied housing unit) Density variables include POPDENS (population per one one thousandth of a square kilometer) and JOBDENS (employment per 39

PAGE 59

one one thousandth of a square k i lometer). To capture des i gn we used thre e variables. One is FORTY H U, which is the proportion of housing units i n the census tract built before 1940. We expect 'this variabl e to be correlated with narrow streets, small blocks, mixed uses, and generally agreeable pedestrian environments. We also created a dummy variable to denote what the Sacramento Council of Governments defines as the downtown, but without the business core of the downtown. 6 This variable is called CBD, and we expect it to be correla ted with many of the same features as FORTYHU, but to a greater extent and with greater parking restrictions. Finally, we defined another dummy variable CBD2 as the core of the central business dis t rict, as defined by the Sacramento Council of Governments. We expect this area to be relatively pedestrian-friendly with much higher parking rates. Included tracts are 7, 10, and 11. One variab l e remains. This is SPLIT, which is the percentage of a census tract within one quarter mile of a transit route, measured in stra ight line distance and not taking into account curvilinear road patterns that might hinder access to nearby transit stops Data sources include the Sacramento Council of Governments for 1991 employment, the 1990 United States Census STF3 disks for most other explanatory variables, and output from the primary model for the dependeot variables. The SPLIT variable is abstracted from a modeled network of Sacramento Regiona l Transit done i n QRSII. Most of the explanatory variables do not have high multicolinearity Those with correlations above .7 include VAPOHU with FIFTEEN (-.8481), and VAPOHU with SIXNOCAR (-.7483). Tab l e 3 presents the results for the model predicting PDN, ; Table 4 does likewise for the model predicting ATN1 6Th is includes the area bounded by the Sacramento and American Rivers, Broadway, and Alhambra and includes tracts 4, 5, 6, 8, 9, 12, 13, 14, 19, 20, 21. 40

PAGE 60

Variable CBD CBD2 EMP FIFTEEN FORTYHU JOBDENS PERSERV POP PO PO ENS POPROOMS SIXNOCAR SIXTEEN SPLIT VAPOHU constant Table 3 Explaining PON, (Adjusted R Square ... 731; with 15 and 55 degrees of freedom) Significance F a .0000 B T .099436 .304 -.114342 -.318 -.000017 -.462 1.394357 1.034 -.425748 -.99 4 .026104 1.389 .93871 5 .839 .000080 9.277 .086820 .691 -.184695 -.368 -4.500330 -1 .123 -2.707137 -1.711 .009762 1.841 .759421 .963 -1.228154 -.680 41 Sig T .7621 .7514 .6527 .3055 .3247 .1704 .4048 .0000 .4924 .721 5 .2662 .0927 .0710 .3396 .4994

PAGE 61

Variable CBD CBD2 EMP F IFTEEN FORTYHU JOBDENS PERSERV POP POPDENS POPROOMS SIXNOCAR SIXTEEN SPLIT VAPOH U constant Table 4 Explaining the ATNJ (Adjusted R Square a .836; F = 26.2 with 15 and 55 degrees of freedom) Significance F = .0000 B T .282309 .468 .547030 2.590 .000041 2.670 .464163 .585 .303989 1.206 .105388 9.534 .293831 .447 .000006 1.183 .071576 .969 .168339 .623 1.616235 .686 .521440 .560 .001506 .483 .452445 .976 .089822 .026 Sig T .1477 .0123 .0100 .5607 .2328 .0000 .6568 .2419 .3370 .5362 .4957 .5775 .6310 .3335 .3095 In explaining the production coefficients, only the magnitude of census tract population is significant. This result may derive from the great range in sizes of the census tracts in the sample. This condition remains even if we remove the two variables collinear 42

PAGE 62

with VAPOHU. These results suggest that neither socio-economic nor design variables have much impact on transit ridership The fact that population density and SPLIT are insignificant suggests that walking access to trans i t is not important for ridership on the trip product io n end of the trip More vari ables are signif icantl y r elated to trip attraction potentials job density being by far the most important. CBD2, denoting the core of the centra l business district also is important though the outer ring of the CBD i s not and has even a depressant effect on transit trip attraction {though not statistically si g nificant) The fact that density stands out for the destination end suggests that walking may be more of a considerat i on than for the origin end of the trip, which is as expected, though the SPLIT variable, which measures the percent of a tract within walking distance of transit, still is not important. This finding strongly suggests that compact development on the destination end of the trip is important for inducing t r ansit t raffic If this finding holds up, it supports findings by Ewing {1993) that compact development is an important consideration in reducing automobile vehicle miles per capita, although his finding derived from shorter auto trip lengths rather than modal diversion. The importance of CBD2 probably derives from the presence of high parking fees in the core; the unimportance and even negative i nfluence of CBD is unclear but may be due to what is perceived as a seedy quality of the downtown area outside of the core. It may also be related to a fear of crime in the area CONCLUSIONS Results here are only preliminary They indicate that the demand analysis method based on product i on and attraction potentials rather than on observed trip productions and attractions has prom ise as a means for analyzing transit's performance and opportunities for improvement in suburban environments. The technique is potentially less cumbersome and theoretically more accurate than the four step modeling system, i n that a model of known statistical behavio r can be estimated from data that transit systems have the ability to collect. Significant improvements can be made to the method, however. The zone system should be smaller with less variation i n size. We expect that with such a change, socio economic and design variables might become important explanatory variables for production and attraction potentials. For a forthcoming study based on Orange County, Florida, we will use traffic analysis zones. The analysis should distinguish between trip types. When we began the study, we lacked the estimating technology for doing this A trip table o.f 24hour flows had sufficient numbers of observat i ons to serve as a dependent variab l e; any disaggregation of 43

PAGE 63

the flows resulted in a trip table with too many zeros to be estimated. We now have the estimating technology to estimate ultra-sparse data sets. The transit and road networks also should distinguish between peak and off-peak service, and the auto network should reflect congested speeds. For a forthcoming study based on Orange County, Florida, we are making this adjustment. We will use TRANPLAN networks developed by KPMG Peat Marwick for this purpose. It also would be desirable to base estimates of walking distance to transit stops based on actual walking paths rather than on the naive approach of straight line distance used here. The fact that our SPLIT variable proved to be insignificant may reflect poor measurement on our part. 44

PAGE 64

REFERENCES Aschauer, David Alan 1991 "Transportation Spending and Economic Growth: The Effects of Transit and Highway Expenditures." Washington, D C.: American Public Transportation Association. Center for Urban Transportation Research 1994 Demographic & Commuting Trends in Florida (Tampa: University of South Flor i da) :33. Cervero, Robert 1989 America's Suburban Centers: The Land Use-Transportation Link (Boston: Unwin Hyman). Cervera, Robert 1993 "Surviving in the Suburbs: T r ansit's Untapped Frontier," Access 2:29-35. Cervero, Robert 1994 "Commuting in Transit Versus Automobile Neighborhoods," Journal of the American Planning Association, forthcoming. Couclelis, Helen 1986 :A Theoretical Framewor k for Alternative Models of Spatial Decision and Behavior," Annals of the Association of American Geographers 76:95113. Davies, Richard B. and Clifford M. Guy 1987 "The Statistical Modeling of Flow Data When the Poisson Assumption is Violated," Geographical Analysis 19:300. Evaluation and Tra in ing Institute 1989 On-Board SuNey, Final Report, Vols. I, II, and Ill (Sacramento, CA: Regional Transit). Ewing, Reid 1993 "How Land Use Design Affects Travel Behavior: Lessons from the Wellington PUD, West Palm Beach, and Everything in Between," presentation to the Florida American Planning Association annual meeting, Amelia Island, FL, 5 November 1993. Flowerdew, R. 1991 "Poisson Regression Modelling of Migration," in J. Stillwell and P Congdon, eds., Migration Models : Macro and Micro Approaches (London: Belhaven):92 112. Haines, Kingsley E. and A. Stewart Fotheringham 1984 Gravity and Spatia/Interaction Models (Beverly Hills, CA: Sage Publications):24-29. 45

PAGE 65

i Handy, Susan L. 1992 "Regional Versus Local Accessibility: Nee-Traditional Development and its Implic ations for Non-work Travel, Built Environment 18:253-267. Hanson, Susan and Margo Schwab 1986 "Describing Disaggregate Flows: Individual and Household Activity Patterns," in Susan Hanson, ed. The Geography of Urban Transportation INew York: Guilford): 154-178. Harvey, G. 1985 "Research Directions in Travel Demand Analysis," Transportation Research A 19A:455-459. Jones, David 1985 Urban Transit Policy: An Economic end Political History (Englewood Cliffs, New Jersey: Prentice-Hall). Meyer, J.R., J. F. Kain, and M. Wohl 1966 The Urban Transportation Problem (Cambridge, MA: Harvard University P ressl:9-55. Muller, Peter 0. 1986 "Transportation and Urban Form: Stages in the Spatial Evolution of the American Metropolis," in Susan Hanson, ed., The Geography of Urban Transportation (New York: Guilford Press):24-48. Pisarski, Alan E. 1 987 Commuting in America: A National Report on Commuting Patterns and Trends (Washington: Eno Foundation for Transportation). Pisarski, Alan E. 1992 Travel Behavior Issues in the 90s (Washington: U.S. Department of Transportation, Federal Highway Administration): 17-28. Pushkarev, B.S. and J. M. Zupan 1977 Public Transportation and Land Use Policy (Bloomington, Indiana: Indiana University Press). Rossetti, Michael A. and Barbara S. Eversole 1 g93 Journey to Work Trends in the United States and Its Major Metropolitan Areas, 1960-1990 (Washington: U.S. Department of Transportation, Federal Highway Administration). Sheppard, Eric 1986 "Modeling and Predicting Aggregate Flows," in Susan Hanson, ed., The Geography of Urban Transportation (New York: Guilford): 91-118. Stopher, Peter R. and Arnim H. Meyburg 1975 Urban Transportation Modeling and Planning (Lexington. MA: Lexing ton Books). 46

PAGE 66

Thompson, Gregory L. 1977 "Planning Considerations for Alternative Transit Route Structures," Journal of the American Institute of Planners 43: 158-168. Thompson, Gregory L. 1993A "Planning Beats the Market: Pacific Greyhound Lines in the 1930s," Journal of Planning Education and Research 13: 33-49. Thompson, Gregory L. 1 993B The Passenger Train in the Motor Age: California's Rail and Bus Industries 1910 (Columbus, OH: Ohio State University Press). Thompson, Gregory L. 1994 "Transportation Investment, Economic Growth, and Social Cost," (Tallahassee: Florida Institute of Government). United States, Office of the Federal Coordinator of Transportation 1935 The Passenger Traffic Report (Washington, D.C : Office of the Federal Coordinator). U S. Department of Transportation 1987 The Status of the Nation's Local Mass Transportation: Performance and Conditions (Washington: USDOTl:46. Washington, Earl J. and Robert W. Stokes 1988 Planning Guidelines for Suburban Transit Services Final Report (Washington: U S. Department of Transportation Urban Mass Transit Administration). Williams, H. C. W L. 1977 "On the Formation of Travel Demand Models and Economic Evaluation Measures of User Benefit," Environment and Planning A 9:285. 47