USF Libraries
USF Digital Collections

A market-based approach to cost-effective trip reduction program design

MISSING IMAGE

Material Information

Title:
A market-based approach to cost-effective trip reduction program design final report, results of survey and conclusions
Physical Description:
i, 23 leaves : ill. ; 28 cm.
Language:
English
Creator:
Winters, Philip L
Cleland, Francis A
Florida -- Dept. of Transportation
University of South Florida -- Center for Urban Transportation Research
Publisher:
Center for Urban Transportation Research, University of South Florida
Available through the National Technical Information Service
Place of Publication:
Tampa, Fla.
Springfield, VA
Publication Date:

Subjects

Subjects / Keywords:
Choice of transportation -- Mathematical models -- Florida   ( lcsh )
Ridesharing -- Mathematical models -- Florida   ( lcsh )
Commuting -- Mathematical models -- Florida   ( lcsh )
travel demand management   ( trt )
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
technical report   ( marcgt )
non-fiction   ( marcgt )

Notes

Bibliography:
Includes bibliographical reference (leaf 23).
Additional Physical Form:
Also available online.
Funding:
Performed in cooperation with the U.S. Dept. of Transportation and the Federal Highway Administration, and sponsored by the Florida Dept. of Transportation under contract no.
Statement of Responsibility:
Philip L. Winters, Francis A. Cleland
General Note:
"April 1998."
General Note:
Final 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 - 001927303
oclc - 39935646
usfldc doi - C01-00016
usfldc handle - c1.16
System ID:
SFS0032139:00001


This item is only available as the following downloads:


Full Text

PAGE 1

A M ARKETBAS E D APPROACH TO C O ST-EFFECTIVE TRIP REDUCT I O N PROGRAM D E S I G N F in a l Report Results of Survey and Conclusions Prepared f or: Department of Transportation Stale of Florida By : Center for Urban Transportation Research CoUege of Engineering University of South Florida A pril 1 998

PAGE 2

Florida Depart ment of Transportation 605 Suwannee Street Tallahassee Florida 32399-0 450 (904) 488 -7774 Fax (904) 922 4 942 Project Manager: Elizabeth Stutts Center for Urban Transportation Research University of South Florida 4202 E Fowler Avenue CUT 100 Tampa F l orida 33620-5350 (813) 974 -3120 Suncom 57 4-3120 Fax (813) 974-5168 Principal Investigators: Francis C l eland Phi li p L. Winters The opin ions, findings and conclusions expressed in this publication are tho se of the authors and not necessarily those of the S t a t e of F lorida Department of Transportati on Prepared in cooperation with the S tat e of F l orida Departmen t of Transportation

PAGE 3

TABLE O F CONTENTS I. PROJECT OBJECTIVES ........ ................... ... ............... .. .. .. .............................. ... ..... .. .. ................ I Il REVIEW OF SURVEY D ESIGN AND RECOMMENDED ANALYTICAL APPROACH ............ 2 III. RESULT S ................ ............................. ............................ ................. ...... ............ ................ 6 Recruitment & Rerum of surveys ............... ...... ........... ........ .................................. ........... 6 Current Commut e Habits .. ................................ ..................... ...... . .......... .... ...................... 6 Initial Model Structure ..... ......... ......... . ................................... .. ....... . .................. ....... ..... 7 Process for Building Second Models ................................................................................ 9 Tests of IJA assumptions .......... ......... ......... . .......... ......... . .............. .... .... ............... I 0 Test of Existence of Multipl e Models for Metropolitan areas ................ ...................... 13 Final Model Structures ................................. ....... .................... .......... ......... ........... ..... ... l 4 Impact of ln.centives .... ..... .... ....................................... ............. ..... ............. ....... .......... l 6 Comparisons of Coeffici e nts between cities ...... ........... ...................... .......... . ....... ....... 1 8 Tampa ................................... ........ .... .......................................... ............ .............. I9 M iami/Fort Lauderdale . . .... ......................... ........ ...... .......... ... ....... ....... ................. 20 Jacksonville ................................................ ........... ............... .. .. ...... .......................... 21 Conclusions ..... ................................. ............................................... ....... ............... 22 Reconunendations .................. ............... ... ........... ... ....... ................... ......... ............. . ...... 22 Bibliography ............................................. .... ......... ............. .................................... .... ........ ............... 23 LIST OF TABLES Table I : Survey Disposition for Market-Based Incentives Project ..................................................... . 6 Table 2 : Current Mode S pli t ......... .................................................................. ............ ... ...... ....... ....... ... 6 Table 3: IJA Test for Final Model Structure (All Cities) .............. .... . . ................................. ............ II Table 4: !!A Test for F inal Model Structure (Miami/Fort Lauderdale) ...................................... .... . . I I Ta bl e s: IJA Test for Final Model Structure (Jacksonville) ....................... ... .... . ........................ .... . l2 T a ble 6: J!A Test for Final Model Structure (Tampa) .......... ............ ........................... .... ........... ......... 12 T able 7: Comparison of S ingle Versus Metropolitan-Area-Specific Mod els ... ......................... ......... 1 3 Table 8: Final Model Statistics . . ... ......... .......... ................................. . . . .... .... .......... ......................... 1 4 Table 9: Test O f Model Structure ...... . . ............... ... .... ...................................................... ....... ........ 1 5 Table 1 0 : Impact of Incentives by Metropoli tan Area ................ . ........................ ........ ........... ............ 16 Tabl e II: Model Comparisons Using Tamp a Data ................... ....... ......... .................... ...... .............. 19 Ta bl e 12: Model Comparisons U sing Miami/Fort Lauderdale D at a . . . . . . . ...... ..... ......................... 20 Table 13: M o d el Comparis ons U sing Jacksonville Data ....... . .... ....... ................................. . . .... . .... 21 1

PAGE 4

I. PROJECT OBJECTIVES The purpose of this project is to quantitatively estimate the impacts of various mixes ofTDM strategies on ridesharing tendencies. A major component of this project was t o develop est .imates of impacts under different conditions using identical methodologies and to test whether projected impacts were the same across all situations tested. This project proposes to accomplish the following three objectives: I. To determine if the impacts of selected TOM strategies are similar in different areas within Florida itself, 2 If so, to determine if these impacts are also similar to impacts measured in other areas of the nation from other SP discrete choice studies 3 To provide a mechanism for the development of effective TOM strategies for the areas surveyed, which included Miami/Fort Lauderdale Jacksonville, and TampaSt. Petersburg. I

PAGE 5

O. REVIEW OF SURVEY DESIGN AND RECOMMENDED ANALYTICAL APPROACH The literature is dominated by examples of the use of logit-based models to estimate traveler mode choices. In order to utilize a binomial or multinomiallogit model in the estimation of the impacts ofTDM strategies, it is necessary to collect data in the form of discrete choices (such as preferring alternative A to alternative B) or to interpret ranked data as a series of discrete choices. If one accepts the notion of the interpretation of rankings as a series of d i screte choices a ranking of, say, 16 alternatives provides effectively (15+ 1 4 + 13+ ... + 3+2+ I = ) 120 separate discrete choices from each respondent far more than one could reasonably ask any respondent to complete in a standard format. The main drawback of this approach is that it does not allow for the respondent to choose part time use of any given mode. The authors experience in a stated preference experiment conducted with employees of the City of Orlando was that in one in six choice tasks (16%) the respondents indicated part-time use of alternative modes, and in total thirty percent of respondents indicated part-time use of alternative modes in at least one of the choice sets presented. Because the partial use of mode s has been demonstrated in earlier projects and becaus e such use would need to be an integral part of the goals of implementing TDM strategies, it was necessary to design the survey in such a way that partial use could be recorded. For that reason, the approach used in the Orlando study, where the respondent indicates how many days per week they will use each mode, was retained and used in this study. The SAS system was originally considered to create a logit model based on the responses, but technical limitations of the software prevented this approach. Inste ad the ALOGIT system developed by the Hague Consulting Group was used to build the models. Because of the parameters of the study design (see Technical memorandum #I from this pFoject), ideaUy four tbree-Jevel variables needed to be identified for tbis study. The literature revie'Y (again see Technical Memorandum#!) Jed to the conclusion that four useful variables to test would include: Vanpool pricing (subsidies) Tran .sit pricing (subsidies) Use of compressed work week/telecommuting as rewards for ridesharing Vanpool pick-up point distance. 2

PAGE 6

Given the objective of efficiency particularly in the data collection side, it was appropriate to study each of the variables as three-level variables in a 9-choice design Given the variab l es chosen for inclusion, the survey took the following appearance: Matrix for survey Sequence Variable I Variable 2 Variable 3 Variable 4 (Rewards) (VP price) (TR Price) (pick-up point) 1 0 0 0 0 2 0 I I 2 0 2 2 l 4 1 0 I I 5 I I 2 0 6 I 2 0 2 7 2 0 2 2 8 2 I 0 I 9 2 2 I 0 A sample survey form appears on the following page. 3

PAGE 7

COMMUTER CHOICE 1: Carpool Vanpool, Bus, Drive alone Given the conditions described below, write in the box how many days per week you would use each option. -Option1.W DAYS PER WEEK Carpool (2 riders) Split your usual parking costs among the riders Allow 5 minutes extra commuting time compared to driving alone Option 2.teJ Vanpoo/ (8 riders) DAYS PER WEEK Your vanpool ride is provided at no cost t o you Vanpool picks you up at your home No parking costs Allow 10 minutes extra commuting time compared to driving alone mm Option 3.1;.;.1 Transit Bus DAYS PER WEEK Your employer provides a transit pass (or toke ns) at no cost to you. Option 4.!!! Drive A/one DAYS PER WEEK Pay your usual parking costs 4

PAGE 8

Additional demographic data were also collected with other materials i n cluded in the l ayout Respondents for the sUJvey w e re se l ected t hrough a randomdigit-dialing process in each of the three metropo l itan areas se l ected. For further details on the survey design, the reader is directed to Technical Memorandum # I prepared for this project. 5

PAGE 9

ID RESULTS Re cruitme nt & re t urn of surveys A separate random-digit-dialing sample was created for each of the three metropolitan areas to be surveyed: Tampa/St. Petersburg, Miami/Fort lauderdale, and Jacksonville. The disposition of the surveys is outlined in the table below : TA BLE 1 Survey Dispo sition For Market-Based In centives Project M etropoli tan Area Tot a l Recruit Total Surveys Tota l Mail Total Valid Surveys m ailed out Surveys Surveys completed Returned Tampa/St Petersburg 666 502 277 158 Miami/Fort Lauderdale 1,117 832 389 220 Jaclcsonville 750 569 290 176 The designation of approximately 60% of the returned surveys as "valid" requires some explanation If a respondent does not select an alternative mode in any of the scenarios presented, clearly the values of all coefficients in the logit model will be zero. This pattern of response while not without meaning, does not provide any useful information for the development of the model. While such responses must be considered in any forecasting estimates, they can be treat ed as a separate group of respondents for the purp oses of model-building and inte grated into the main model for any forecasting analysis. Also, it s hould be noted that tbe percentage of surveys that were withheld from analysis i s virtually the same percentage of the population in ea ch metropolitan area (about 58%) and thus should have a similar impact on the value of coefficients produced Curren t Co mmut e Habi t s Current mode use was measured in the three areas as part of the survey: TAB LE 2 -C urrent Mode Split JacksonviU e Miami/Ft La u derdale Tamp a Carpool 8.6% 8.5% 10 .5o/c Vanpool 0.9"/o 0.9% 0.5% Bus 1.0% 3.4% 1.1% Drive Alone 89.5% 87.2% 87 9% 6

PAGE 10

Initial Mod e l Structure The initial approac h to m odel structure s was to build a logit model which attempted to estimate the number of days each respondent chose each alternative. The e stim a te was based on t h e incentives available and the respondent's individual circumstances, as described by responses to the demographics and commuting characteristics questions While the response s t ructure in the survey does not technically meet the format of a discrete" choice, the responses were interpreted as a separ a te choice for each day of the week, and were analyzed as "repea ted measures" as desc ribed by Ben -Akiva and Lerman. (pp. I 19-120) The models were buil t by initially including all of the measured variables and several t ransformations of t h ose variable s (including speed, vehicles per house ho l d member interactions between bus s top distances and length of bus ride, interactions bet we e n incen tives and bus stop dis t ances, and so forth). The ALOGIT software o ut pu ts "t" ratio s for each parameter and the associated probability of significance. Parameters were eliminated and mode l s rebuilt in an iterative fashion until every variable met the 95% confidence criterion. I n som e modeling situations it is desirable to retain even those coefficients that are not stat is tically significant particular l y w h en a theory of travel choice is based upon the existence of a non-zero coefficient for that variable. The reason for insignificance may be due to the survey design, particularly sample size issues. In the case of this s urvey, however, the purpose was to determine if th e coefficients d iffered significantl y in d ifferent areas. Thus if the coefficients were not significantly different from zero it would be difficult to argue tbat they differ ed significantly from each o ther. The elimination of all non-significant coefficient s from the model thus strengthens the ability to draw conclusions about differences in model characteristics from one ar e a to a nother. A crucial te st to validate the correctness and appropriateness of the model st ruct u r e is the test of I ndepe ndenc e from Irrelevant Alternatives (IIA) The purpose of this t est is to e nsure that the alternat ive s presented to respon dents are i ndeed viewed as independent. The exampl e put forward in Ben-Akiva and Lerman's text is that a model could be built where alternatives included riding a red b u s and ridi n g a blue bus. (p. 52) If respondents viewed a lt ernatives as differing only along irrelevant dimensions (such as bus color), when the model was re-estimated it should show a significant diff erence in expl anatory power from the original model. The explanatory power of a log it model is measured by comparing the log-likelihood valueS (abbreviated as I) a t the initial and final i terations of the model building process. The rh o squared st at istic (analogous to the R-square statistic for regressions) is ca l culated by dividing t h e impr ovement in the loglikelihood value b y the initial l o g-likelihood val u e State in the form of an e q uation, this is: I ) / (t(O)) An adjusted statistic ( rhobar squared) is calculated by subtracting the number of parameters from the numerator in the above eq uation. 7

PAGE 11

The IIA test is conducted by removing all observations where one of the alternatives was cb.osen (for this test, the bus alternative was removed). The model is re-estimated using the same variables as the original model. However, the coefficient values are re-estimated. The original model (less any parameters that are specific to the removed alternative) is then re-applied and the difference in the ability of the models explain the results is examined. The resulting statistic is compared to chi-square tables. The number of degrees of freedom is equal to tbe number of parameters in the restricted model. This test was performed by removing the bus-riding alternati ve from the choice set and re estimating each of the models. The test statistic used is Small and Hsiao's corr ected approximate likelihood ratio test as described in Ben-Akiva and Lerman. (p. 185) The test statistic is calculated as follows: where ct is a scalar, generally assumed to be equal to I, the value of which is checked by examining the covariance matrices of tbe models being compared, Beta refers to the matrix of coefficients in the logit model, and N and N 1 are the numbers of observations in the restricted and unrestricted models, respectively. The models created with this method did not uniformly pass the IIA assumption The usual procedure to correct thi s problem is to eliminate variables whose covariances differ greatly between the restricted and unrestricted models However, even after numerous attempts to do so, the models would not pass this test. This indicated that there were proble ms with this model structure. With hindsight, this finding was not entirely unexpected. Many of the incentives were set up as rewards for using commute alternatives "at least two days per week", which might make the choice between carpooling and vaopooling indifferent. Or it could be t ha t the real choice was between whether to drive alone i n combination with an alternative or to use the altemative every day. In any case, tbe failure-of the models to pass the IIA test requ ired a new approach to tbe issue. 8

PAGE 12

P r oce ss f o r B ui ld ing Seco n d M o d el s Following the failure of the initial model t o p rovide a stable set of results, a second structure was applied. In this structure it was assumed that the respondent made a series of decisions instead of allocating days directl y to each of t h e individual modes. T h e deci s i o n was modeled as a "tree" with the following structur e ......... .............. .... 1 I ........... II ..,.....,,......, I 1 ............ .-1 I .,...,_, J 1-..,..1 1""'-'1 I """"'..,..I I ...,,.,., I I I$N .. -1""-1 I """'"""" I 1, ............ 1 .......... 1 I tl ....... II-"'"' II .......... H....,"'" I I I ......... ............. "'""'*'" I I'""'"''"' I 1""'"''""'1 I ........ I I I t1 .,. ..... 1 H ,..,._ I .......... 1 ........... 1 H .. ..,. ...... I 1 ... -I H w.,._, I H ,,..,. .. ,.. I I ,,..,._ I 1 ... I y.......... I y ........... I 4 ........... .......... 1 ........... 1 H..,...,.l 1 .... ,. .... I ......... I y ........ I i'""""'"' I ""-I ........... 1 -j ........... l -j ......... I -j .......... 1 ; .......... I .......... 1 .. .... I H .., ...... l H ,...,..,..l y .......... l ...,,..,.. I -1 ............ 1 -1 .... ..,..1 9

PAGE 13

Each level of the decision tree detailed above was modeled as a separate decision among multiple choices, with its own set of explanatory variables. The initial variables to be used for the model were determined by running a SAS stepwise discriminant ana l ys i s at an 85% confidence level of significance. Thi s was done to simplifY the process of variable selection from a standard trial-and error procedure. In the initial decisions in the tree, variables selected by the discriminant procedure mirrored those deemed significant by the logit estimation process. The selected variables were tbeo run in the ALOGIT modeling software to select those variables that were significant at a 95% confidence level. This process required sequential partitioning of the responses in the data as well. T he modeling process was run for each of the three metropolitan areas as well as for all three areas combined in a single model. Tests of //A assumption Following successful completion of the model building, the models were again tested for violation of the IIA assumption. In this case, however, the alternative removed was not a single alternative (such as carpool or vanpool, or bus) but rather a unique combination of the alternatives (such as drive alone/carpool), which constituted a "choice" as modeled in the revised structure. In total, there are 15 possible combinations. The test statistic for each is reported below. With these results, and given that there at a minimum 75 degrees of freedom for each model, this is well under the bounds required for the result to be random at a 95% level of confidence. Rejection of the IIA statistic as explained in Ben-Akiva and Lerman (pp. 185-194) (i.e., that the upper-tail probability of the chi-square distribution be below the .05 level) would occur at test statistic values of96. 2 or less. All oftbe values reported below pass this test. Thus the IIA assumption cannot be rejected, and the mode l structure is accepted 10

PAGE 14

" "' 3 DA Test For r '"'" Val u e JTest 'Only :5/4) 30. < I Only -1 I I I -10275.' ml vruy -1 :G 13 ( "' vruy -11031.6 -11031 4801 24' + 1 i Q -10844.8 4812 6. +Bus -IORQI Q -1089U -6.o -11?1? R 4881 J.2:c -113310 1 J110 Q -73RO Q r 3273 7:( AIM + -106 _, r 9 . AIM + B u s -107 1 -I -4781 2f:i + -103 4 7.( 471S Alone+ +Bus u 18< Alone+ +Bus -IV:>IO, -10577 ( '" +BUs ( 491; 38 TABLE 4 DA T es t For Final Mode l Structure Miami/For t Lau de rdal e Start Value Finis h Observations Test Va lue Statistic Drive Alone Only -3451.0 -3447.4 1482 10.3 Carpool Only -3938.4 -3935.5 1810 9.1 Vanpool Only -4 123.4 -4120.8 1870 8.3 Bus Only -4224 0 -4223.2 1899 21 Carpool+ Vanoool -4079 8 -4079.4 1884 L3 Carpool + Bus -4126.2 -4 125.7 1888 1.6 Vanoool +Bus -4193. 7 -4193.2 1908 1.6 Carooo l + Vanoool +Bus -4274 8 -4274.7 1921 -0.3 Drive Alone + Carooo1 -2769.4 -2762 5 1262 18.5 Drive Alone+ Vanoool -4 050.0 -4049 6 !866 L3 Drive Alone+ Bus 4100 5 -4100 1 1879 L3 Drive Alone+ Carpool + Vanooo 1 -3897 3 -3895.3 1855 6.4 Drive Alone + C ool +Bus -3686.1 -3684.8 1780 4.1 Drive Alone + V anoool + Bus -4042 1 -4041.8 1868 LO Drive Alone + Caroool + V anoool + Bus -4314 0 -4313.9 1932 0.3 11

PAGE 15

TABLE 5-DA Test For Final Model Structure Jacksonville Start Value Finish Observati ons Test Value Statistic Drive Alone Onlv -2919.5 -2916.6 1212 7.7 CSl"pool Onlv -3272.7 -3269.3 1452 9.6 IV an pool Onlv -3461.6 -3460.3 1513 3 7 !Bus Onlv -3552.1 -3551.8 1547 0.9 CSl"!>.OO! + Vanpool -3523 6 -3523 0 1551 1.7 CS(poo1 + Bus -3514.8 -3514 .6 1547 0.6 Vanpoo 1 +Bus -3609 9 -3609 8 1571 0.3 CSl"pool + Vanpool +Bus -3642 8 -3642.5 1578 0.9 Drive Alone + Carpool 2378.2 -2371.7 1071 16 .6 Drive Alone+ Vanooo1 -3402.7 -3 402.2 1519 1.4 Dri ve Alone+ B u s -3428.4 -34 28.1 1527 Drive Alone + + Vanoool -3248.7 -3245.9 1 4 98 8.( Drive Alone + +Bus -3191.7 31 9 1.2 1470 1.4 Drive Alone + Vanooo1 +Bus 3309.7 -3308 8 1502 2.6 Drive Alone+ Car ooo1 + Vanooo1 +Bus -3635.1 -3635.1 1576 0 0 TABLE 6-DA Test For Final Model Structure Tamoa Start Value Finish Observations Test Val ue Statistic Dri v e Alone Onl y -2 292.2 -2287.2 1051 12.7 Carpoo l Onlv -2658.3 -2653.4 1301 13.3 Vanpoo1 Onlv -2 824.9 -2823.4 1351 4.1 Bus Onlv -289 0 3 2889 3 135 5 2 8 Carpool+ Vanoo ol -2830.8 -2829.9 1377 2 5 Carpool + Bus 2827.9 -2826.9 136 7 2 8 Vanpool +Bus -2 9 48.7 -2948 6 1 402 0 .3 CS(pool + Vanoool +Bu s -1859.4 -1854.1 940 13.1 lOri ve Alone + Carpo o l -2790.4 -2789.6 1361 2.2 Drive Alone + Vanoool -283 6.3 -2835.8 1375 1.4 Driv e Alone + Bus -2735.0 -2734.3 1366 1.9 Drive Alone + Carpool + V an pool -2681. 5 2679.6 1345 5 2 Dri ve Alone+ CSl"pool + Bus -28 1 1.6 -2811.4 1372 0.6 Drive Alone + V anooo1 + Bus -2955 .8 2955 7 1404 0 3 IDrive Alone+ CSl"POO! + Vanpool + Bus -29 54 .9 2954 .7 1401 0.1 12

PAGE 16

Test of Exis t e n c e o f Multip l e Models f o r Metrop o li t an are a s The next task in the model building was to determine if it were in fact necessary to create a separate model for each metr o p olitan area, or if one model would serve t o describe the results for all three areas This pr oc ed u re does not however, determine the t r ansf erab ility o f t h e policy coefficie n ts on i t s own. This proced ure is t est ing all of the paramet ers including n on-po licy variab les such as income and vehicle ownership to determine whether these parameters d o i n fact differ b e tween m etr o politan areas or if th e differences have b een accounted for by the dem ogra phic da ta collected The procedure for con ducting this type o f analysis is described in Ben-Alciva and L er m an's text on discrete choice. (pp 194 -2 0 4 ) M o dels are built for the three area s and th e n a not h er model is built that covers all of t h e results. The log likelihood values for the two type s of models are compared The resulting model s tatistics were : TABLE 7-Comparison Of S ingle V e rsus M e trop o litan Area-specific M odels M o d e l Statistic Single mod e l Metro area specifi c model s Initial Log Likeli hoo d -17,063.9 17,063.9 F inal Log Likeli hood -11, 48 2 6 -11,026 0 Tes t Statistic 456 6 2 = 913 2 Probability t bat 99.9+% difference is significant (p2 with 78 degrees of freedom) p1 value 327 354 -p1 value.323 .336 -This test proves conclusively that the inde p endent model s for Tampa/St. Petersburg, Miami/Fort Lauderda le, a n d Jack sonville are superior to the single m odel tha t covers all three metrop o lita n ar eas S u bseque n t analy ses were perform ed using the three area-s pecific models. 13

PAGE 17

Final Model Structu res The model structure is completely defined in the appendil{. Each model has 32 equations each of which predicts utility for one of the 4 modes in each ofit.s potential combinations as described in tbe tables above. The final statistics on each model are shown below: TABLE 8-Final Model Statistics Model Statistic Tampa I St. Miami/Ft. Jacksonville Petersburg Lauderdale Initial Log Likelihood -4893 6 6696.1 -5474.15 Final Log Likelihood -3004 2 4347 6 -3674.21 Base p 2 value 0.386096 0.350727 0.3 28807 Number of parameters 99 102 )08 -0.365866 0.335494 0.309078 Base p2 value 14

PAGE 18

It should be note d that the rho -bar square v al ue s are well a b ove the reas o nableness test cited by Beaton which i s that m odel rho bar square values should be between 0 2 and 0. 3 A further test is c o nduc ted on the mod e l stru cture t o determine t h a t t he coefficients (rather than m erely applying con stants) add sigoillcant information The results of this test are shown below : Tabl e 9 T es t of Mo del Str ucture M ode l Stat isti c Tam p a / St M iami/Ft. Jackso n v ill e Pe ter sb ur g L auderdale Base T est Base Test Base Test Initial L o g Like lihood 4 893 6 -4,893.6 -6,696 1 6 696.1 -5 4 74.2 -5,474 2 Final Log Likelihood -3 004 2 -3,558 8 -4,347.6 4 ,984.3 -3 674.2 -3,964 Tes t Stat is tic 1,1 10 1,272 580 Probability that 99 .9+% 99 9 + % 99.9 + % difference is significant (p 2 with 99 degs of (p 2 wit h I 02 deg s (p 2 with I 08 deg s freedom) of fr e edom) of freedom) The mode l s t ructure is confirm e d. 15

PAGE 19

Impact oflntentives B ecause of the complexity of the model structures, it is perhaps best to vie w the model as a black box for tbe purposes of evaluating the impact of i ncentives on individual modes. Clearly it would be an extremely complex task to isolate out the effects of individual vari a b les on the individual modes In order to simplifY this analysis, CUTR is applying a sensitivi t y analysis to each of the in centives The impac t of offering each incentive is predicted for each potential mode. The impact of mul t iple incentives on any g ive n mode is nearly, but not exactly, additive. The "base lin e" figures in the survey indicate th e projected mode split if neither compre sse d work weeks nor t e l ecom muting is offered as in incentive, an d if transit and vanpooling both cost $ 50 per month Thes e figures indicate a mu ch higher l eve l of ridesbaring than the respondent s actually indicated in their survey returns when asked dir ectly how they commu ted to work. This suggests two things : Some respondents may not be using panicular options such as carpooling, van pooling or transit because they bad not considered using them or don' t know how to get staned, but would use them if this information were provided Some of the respondents responded unrealistically to the survey, indica ting that t hey would use commute alternatives in situations where they clearly would not This is a sort of Hawthorne effect where subjects are responding differently because they are being studied This situation probably results in overstating th e amount of ridesbaring that would occur given any of these scenar ios, but it does not dimini s h the ability o f the study to compare the impact of incentives re lativ e t o o ne another. These results are reported in percent of trips as opposed to percent of peo ple using a particular mode TABLE 10-Im1 act Of Incentives By Metropolitan Area Tam a Model Miami Model Jacksonville Model CP VP Bus Drive CP VP Bus Drive CP VP Bus Drive Alone Alone Alone (Basel ine 15% 4% 3% 78"/o 16% 3% 3% 78% 19% 4 % 4% 7 4 % CWW 22% 4% 4% 70"/o 22% 4% 4 % 70"/o 22% 4 % 4% 69% rr elecornrnute 15% 4% 3% 78% 17% 3% 3% 77% 19"/o 4 % 74o/c $25 Vanpool IS% 4 % 3% 78 % 16% 3% 3% 78% 19"/o 4 % 4% 74o/c Free Vanpool 14% 6% 3% 77% IS % 8% 3% 75% 17% 8% 3% 72% $25 Transit 14% 3% 5% 77% 17% 3% 3% 77% 19% 4% 4% 74% Free Transit 16% 5% 8% 71% 17% 4 % 7% 73% 17% 4% 7% 72% Pick-up 1/4 15% 4% 3% 78% 16% 3% 3% 78% 19% 4% 4 % 74% mile Pick-up at 14% 7% 3% 76% 15% 5% 3% 77% 18% 5% 4% 73% door 16

PAGE 20

From these data, it is apparent that the incentive with the most impact on reducing solo diver traffic is the reward of allowing compressed work weeks for using commute options at least 2 days/week. The tendency is for this reward to increase carpooling. The next largest decrease in drive alone traffic was provision of free transit service. Free vanpooling had some effect on solo driving in Miami/Fort Lauderdale, less in Jacksonville and very little irt Tampa Other incentives (pick up point for the vanpool providing transit or vanpools for $25/month) had minimal impact. The model structure did not allow for the analysis of interactions between incentives (such as free transit and compressed work week being offered together). A survey that would allow that type of analysis would have required far mor e respondent time and effort than could be reasonably expected. It is quite likely that combining an incentive that proved beneficial to users of any alternative with an incentive specific to one fortn would produce greater increases for that form but a separate study would have to be conducted to determine whether this is in fact the case. 17

PAGE 21

Comparisons of Coefficients between cities Due to the complex structure of the model, the best way to test for the equality of the impact of the incentives by mode is to examine the results of applying each of the three cities models to the same data set. This procedure accounts for differences in response to incentives due to demographic difference, as opposed to variations in preference (or "taste variations" as they are commonly referred to i n the discrete choice literature). Also, because of the rigorous structure of the models (i.e. that all coefficients were required to be different from zero at a 95% level of confidence to be included a t all) it is highly likely that the differences observed are in fact significant. It should be kept in mind that it was clearly demonstrated that the separate models do in fact perform significantly better than a single model covering all three areas, as discussed earlier. This fmdiog also increases the probability of significance of the differences observed. Each of the mode l s was applied in tum to the data collected from the Tampa area, the Miami/Fort Lauderdale area, and the Jacksonville area. The results for each will be examined separately. 18

PAGE 22

Tampa The result s for application of each of the models to the Tampa data is shown be low: Again, this data is presented as percent of work trips. TABLEll Model Comparisons Usi.W Tamna Data Tam1 a .Model Mia m i Model Jacksonville Model CP VP Bus Drive CP VP Bus Drive CP VP Bus Drive Alone Alone Alone !Baseline 15% 4% 3% 78% 16% 3% 3% 79% 18% 3% 3% 75% cww 22% 4% 4% 70% 21% 4% 4% 71% 22% 4% 4% 71% ITelecommute 15% 4% 3% 78% 16% 3% 3% 78% 18% 3% 3% 75% $25 Vanpool 15% 4% 3% 78% 16% 3% 3% 79% 18% 3% 3% 75% Free Vanpool 14% 6% 3% 77% 15% 8% 2% 75% 17% 7% 3% 73% $25 Transit 14% 3% 5% 77% 1 6% 3% 3% 78% 18% 3% 4% 75% Free Tra nsit 16% 5% 8% 71% 16% 3% 7% 74% 16% 4% 7% 73% !Pick-up 1/4 mile 15% 4% 3% 78% 16% 3% 3% 79% 18% 3% 3% 75'V< !Pick-up at door 14% 7% 3% 76% 15% 5% 2% 78% 18% 4% 3% 75o/c The major differences between the models were: Jacksonville residents responded l ess strongly to the compressed work week (C'WW) incentive. Dri v ing alone was reduced by only 4%, whil e the reduction was twice as large for Tampa and Miami/Fort Laude rdale residents. Tampa residents responded less s trongly to the Free Vanpool options than either Jacksonville or Miami/Fort Lauderdale residents Vanpool u se was predicted to account for 2% more trips (up from 4%) for Tampa residents, wher eas the increase was 5% in Miami/Fort Lauderdale and 4% in Jacksonville. Response to free transit was different for each ma rket. Tampa residents increased use of carpooling, vanpooling, and bus whe n Fre e Transit was available (indicating that many would ride transit !/week with free transit, but not much more) whereas carpooling actually decreased for Miami/Fort Lauderdale residents and stayed the same for Jacksonville res i dents. Vanpooling increased slightly for Jacksonville residents w h en free Transit was made available. Increase in vanpooling was slightly greater for Tampa residents in re sponse to door to-door p i ckup than for Miami/Fort Lauderdale or Jacksonville residents, although the increases were negligible in all a reas 19

PAGE 23

Miami/Fort Lauderdale Below are the results of application of the three models to Miami/Fort Lauderdale data: Most of the major differences seen in the initial comparison using Tampa data were confirmed in the application of the models to the Miami/Fort Lauderdale data Jacksonville residents responded less strongly to the compressed work week (CWW) incentive. Driving alone was reduced by only 4%, while the reduction was twice as large for Tampa and Miami/Fort Lauderdale residents. Tampa residents responded somewhat less strongly to the Free Vanpool options than either Jacksonville or Miami/Fort L auderdale residents, although this difference was not as marked as with the previous data set.. Vanpool use was predicted to account for 3% more trips (up from 5%) for Tampa residents, whereas the increase was 5% for Miami/Fort Lauderdale residents and 4% for Jacksonville residents Response to free transit was different for each market. Tampa residents increased use of carpooling, vanpooling, and bus when Free Transit was avai.lable (indicating that many would ride transit !/week with free transit, but not much more), whereas carpooling increased only slightly in Miami/Fort Lauderda le and decreased mar.kedly for Jack sonville residents. Increase i n vanpooling was substantially greater for Tampa residents in response to door-to-door pickup than for Miami/Fort Lauderdale or Jacksonville residents. 20

PAGE 24

Jacksonvi lie Finally, each of the three models are compared using demographic data from Jacksonville. TABLE 13-Model Comparisons Usine Jacksonville Data Jacksonville Model Miami Model TamoaModel CP VP Bus Drive CP VP Bus Drive CP VP Bus Drive Alone Alone Alone Baseline 19% 4% 4% 74% 17% 3% 3% 77% 15% 5% 3% 77% cww 22% 4% 4% 69"/o 23% 4% 4% 69% 21% 6% 4% 69% Telecommute 19% 4% 4% 74% 18% 3% 3% 76% 15% 5% 3% 77% $25 Vanpoo1 19% 4% 4% 74% 17% 3% 3% 77% 15% 5% 3% 77% Free Vanpoo1 17% 8% 3% 72% 16% 8% 2% 73% 14% 7% 3% 76% $25 Transit 19% 4% 4% 74% 17% 3% 3% 76% 14% 4% 5% 76% Free Transit 17% 4% 7% 72% 17% 4% 7% 72% 16% 5% 9% 70% Pick-up Y. mil e 19% 4% 4% 74% 17% 3% 3% 77% 15% 5% 3% 77% Pick-up at door 18% 5% 4% 73% 16% 6% 3% 76% 14% 8% 3% 75%, The same major differences in response are observed: Jacksonville residents responde d less strongly to the Compressed Work week (CWW) incentive. Driving alone was reduced by only 5%, while the reduction was nearly twice as large for Tampa and Miami/Fort Lauderdale residents. Tampa residents responded somewhat less strongly to the Free Vanpool options than either Jacksonville or Miami/Fort Lauderdale residents, although this difference was not as marked as with the previous data set. Vanpool use was predicted to account for 2% more trips (up from 5%) for Tampa residents, whereas the increase was 5% for Miami/Fort Lauderdale residents and 4% for Jacksonville residents. Response to free transit was different for each market. Tampa residents increased use of carpooling and bus when Free Transi t was available (indicating that many would ride transit i/week with free transit but not much mor e ), whereas carpooling did not increase for Miami/Fort Lauderdale residents and decreased for Jacksonville residents Overall reductions in driving a lon e were greatest for Tampa residents, less large for Miami/Fort Lauderdale residents, and smallest for Jacksonville residents. Increase in vaopooling was greater for Tampa residents in response to door-to-doo r pickup than for Miami/Fort Lauderdale, and both responded more strongly than Jacksonville residents. 21

PAGE 25

Conclusions The results from this analysis of applying the three models to each of the three datasets are remarkably consistent. From this analysis, CUTR concludes: The impacts of these incentives are not identical in all cities. Customized, single city trip reduction plans should be emphasized and standa rdized regional or statewide plans should be de-emphasi zed This finding essentially validates most of the state TDM policies to date. A compressed work week reward would be most effective in Tampa and Miami/Fort Lauderdale, and should have a substantial effect on both use of commute alternatives and traffic reduction The added benefit is that this type of schedule typically moves people off of peak hours for their commutes. Vanpool p rice reductions will be more effective in Miami/Fort Lauderdale and Jacksonville than in Tampa. Transit price reductions, or promotions (such as free bus passes) will be more effective in Tampa, and markedly less so in Jacksonvill e Door-to-door pickup for vanpools has marginal effects in all cities, but appears to be most effective in Tampa. Continued promotion of alternatives and how to start using them should remain a mainstay of TDM efforts, as the survey results indicate a higher potential for their use even given no incentives than currently exists. It should also be noted that the compressed work week and telecommuting options were provided as incentives r ather than actual commute options. The failure of telecommuting to operate as an effective incentive for use of alternate commute modes should not in any way discourage its use as an alternate mode in it s own right. The success of compressed work weeks in promoting use of alternative modes in this survey is an extremely positive result as it sho ws that commute trip reduction techniques can actually be used as effective rewards for use of other commute trip reduction techniques. Recommendations Tam pa area TDM programs should begin work to implement compressed work weeks, particularly in conjunction with transit subsidies, to reduce traffic io the _Tampa area Miami/Fort Lauderdale area TDM programs should also focus on implementatien of a co mpresse d work week reward in conjunction with vanpool/transit subsidies. Both vanpooling and transit could be effective trip reduction strategies for the area. Jacksonville area TOM programs should also focus on the CWW program However, impacts of such a program in Jacksonville will probably be Jess than in Tampa of the Miami/Fort Lauderdale area. This should be kept in mind when allocating resources and evaluating the effectiveness of this type of program Promotion of use of alternatives, their availability, and how to start using them should continue in all areas. 22

PAGE 26

Bibliography Ben-Akiva, Moshe, and Steven Lennan Discrete Choice Analysis : Theory and Application to Travel Demand. Cam bridge MA: MIT P r ess 1985 th printing 1991. 23


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 2200325Ia 4500
controlfield tag 001 001927303
005 20080327123138.0
008 980925s1998 flua bt s000 0 eng d
datafield ind1 7 ind2 024
subfield code a C01-00016
2 local
035
(OCoLC)39935646
040
CBT
c CBT
d FHM
043
n-us-fl
049
FHMM
090
HE336.C5
b W56 1998
1 100
Winters, Philip L.
245
A market-based approach to cost-effective trip reduction program design :
final report, results of survey and conclusions /
[Philip L. Winters, Francis A. Cleland]
260
[Tampa, Fla.] :
Center for Urban Transportation Research, University of South Florida ;
[Springfield, VA :
Available through the National Technical Information Service,
1998]
300
i, 23 leaves :
ill. ;
28 cm.
513
Final report.
September 1996-April 1998.
536
Performed in cooperation with the U.S. Dept. of Transportation and the Federal Highway Administration, and sponsored by the Florida Dept. of Transportation under contract no.
BA498
500
"April 1998."
504
Includes bibliographical reference (leaf 23).
530
Also available online.
0 650
Choice of transportation
z Florida
x Mathematical models.
Ridesharing
Florida
Mathematical models.
Commuting
Florida
Mathematical models.
travel demand management
trt
700
Cleland, Francis A.
710
Florida.
Dept. of Transportation.
University of South Florida.
Center for Urban Transportation Research.
994
C0
FHM
8 773
t Center for Urban Transportation Research Publications [USF].
4 856
u http://digital.lib.usf.edu/?c1.16