Urban Transportation Planning Model Update - Phase II TASK E DEVELOP STANDARDIZED MODAL SPLIT MODELS

Transcription

1 Urban Transportation Planning Model Update - Phase II TASK E DEVELOP STANDARDIZED MODAL SPLIT MODELS Prepared for: The Florida Department of Transportation Prepared by: COMSIS Corporation Killearn Village Square 3644 Shamrock West Tallahassee, Florida (904) October 1981

2 Urban Transportation Planning Model Update - Phase II TASK E DEVELOP STANDARDIZED MODAL SPLIT MODELS Prepared for: The Florida Department of Transportation Prepared by: COMSIS Corporation Killearn Village Square 3644 Shamrock West Tallahassee, Florida (904) October 1981

3 FOREWORD This report is the third in a series of four reports documenting Phase II of the urban Transportation Planning Model Update Study. This report documents Task E. Task E has developed standardized modal split models for use in the urbanized areas of Florida. Phase I of the Urban Transportation Planning Model Update has been completed. Phase I consisted of Tasks A and B. Task A included the initial review of operating experience within the Urban Transportation Section (now the Bureau of Urbanized Area Systems Planning) of the Florida Department of Transportation and the identification of individual programs and procedures that would best serve all or groups of Florida's urbanized areas in the application mode. In Task B, a trip generation model set was developed for potential use in the urbanized areas of Florida. Phase II is divided into four tasks. Task E, which is documented herein, has developed standardized modal split models; Task C has developed standardized distribution and assignment models; Task D will develop a systems evaluation model; and in Task F, corridor analysis procedures were developed. This report is divided into four chapters which parallel the study progress in Task E. The first chapter presents a review of operational experience. In Chapter Two, alternative modal split procedures were recommended for further consideration. In Chapter Three, an evaluation of the alterna- i

4 tive modal split procedures was accomplished and specific recommendations were advanced as to which areas should use the recommended procedures. In Chapter Four, refinements of this recommendation are documented based upon Florida DOT and Model Task Force review. In addition, a procedure to standardize the application of the modal split models is presented. ii

5 TABLE OF CONTENTS Page FOREWORD... LIST OF FIGURES... LIST OF TABLES... CHAPTER ONE - REVIEW OF OPERATIONAL EXPERIENCE... Mode Choice Models - Background... Review of Florida Mode Choice Models... Auto Occupancy... CHAPTER TWO - FORMULATION OF ALTERNATIVE PROCEDURES.. Categorization of Existing Florida Models... Alternatives to be Considered... Recommendation for Modal Choice Models... Accuracy Evaluations... CHAPTER THREE - EVALUATION OF ALTERNATIVE MODAL SPLIT PROCEDURES... Review of Modal Split Model Categorization... Review of Recommended Modal Split Alternatives.. Area Sensitive Evaluation of Recommended Procedures... Recommendations... Summary... CHAPTER FOUR - SELECTION AND REFINEMENT OF RECOMMENDED PROCEDURES... Sensitivity Tests of Logit Formulation... Specific Model Recommendations... Procedures for Applying Standardized Modal Split Model... iv i V 1 : I ; APPENDIX I - TRIP TABLE DEVELOPMENT CODE CONTAINING POLK COUNTY MODAL SPLIT MODEL... I-l APPENDIX II - TRIP TABLE DEVELOPMENT CODE CONTAINING GAINESVILLE MODAL SPLIT MODEL... II-1 APPENDIX III - STANDARDIZED MODAL SPLIT PROCEDURES... III-1 iii

6 Figure LIST OF FIGURES 1 Mode Choice Model Alternatives 2 Mode Choice Model for Polk Study 3 Example of Transit Trip-Making Indices for Sarasota-Manatee 4 Summary of Mode Choice Models Page iv

7 Table LIST OF TABLES Page 1 Choice Models and Other Characteristics for Florida Study Areas 20

8 Chapter One REVIEW OF OPERATIONAL EXPERIENCE The purpose of this chapter is to accomplish a generalized review of the existing modal split models in use within the State of Florida as well as to review current Urban Mass Transportation Administration (UMTA) policy with regard to modal split modeling. Also, a review of the current stateof-the-art in disaggregate, behavioral modal split modeling techniques currently being researched is made. The findings of this task will be utilized to formulate alternative procedures in Chapter Two. Mode choice analysis is accomplished to determine the number of transit trips and auto person trips. Auto occu- pancy relationships are utilized to convert the auto person trips to auto drivers and auto passengers. Although auto occupancy relationships will be reviewed herein and alter- nates presented in Chapter Two, Task E will not evaluate and validate alternate relationships since this is beyond the scope of this project. MODE CHOICE MODELS - BACKGROUND There are numerous methods utilized for mode choice analysis. The four basic approaches have been: 1. Direct Generation: Trip generation rates are developed for auto driver trips, thereby eliminating any mode choice model procedure, as such. Rates can also be developed to generate transit trips directly. This procedure is most appropriate for smaller urban areas where transit use is very low. 1

9 2. Pre-Distribution A mode choice model that is applied after person trips are generated but before trip distribution. This technique results in trip distribution model applications for auto and for transit trips. Also, most approaches of this type do not consider system attributes directly. Basically, trip end characteristics such as auto ownership and population density, may be considered at the production end and employment density at the attraction end. Sys tern characteristics such as whether service is provided or not may be considered. Interchange characteristics related to specific production and attraction areas cannot be considered here. This type of approach is most appropriate in smaller to moderate size urban areas with a small bus system. 3. Post-Distribution: A mode choice model that is applied after trip distribution of person trips. Here the model is applied to the person trip table resulting in transit passenger trips and auto passenger trips. The advantage.of this technique is that it can reflect system operation between pairs of production and attraction zones based upon operating characteristics and cost of travel. This allows the ability to assess: competing highway and transit levels of service; auto operation cost; parking fees; and transit fares. This approach is most suitable for larger urban areas with a transit system large enough to support system level alternatives analysis. 4. Simultaneous: A simultaneous trip distribution and mode choice model which in one operation determines the interchange by mode between production and attraction zone pairs. The mode choice analysis element can reflect the same system parameters as the post-distribution mode choice model. At best what is saved is a computer step. What is lost is the ability to separately evaluate the two steps of distribution and mode choice. A pictorial representation of the place of each of the above approaches to mode choice analysis is shown in Figure 1.

10 TRIP GENERATION Direct Generation --_-_-~----_.o--~ of Trips by Mode I Pte-Dlstributlon * I I I I TRlP DISTRIBUTION ti Simultaneous --mm i MODE CHOICE MODEL 4-a _- I Post-Distribution MSw--w --w-b-- J I TRAFFIC ASSIGNMENT I FIGURE 1 MIDI CIiOICE MODEL ALTERNATIVES 3

11 REVIEW OF FLORIDA MODE CHOICE MODELS The Logit Model - Post-Distribution Basically, all of the above generic types of models have been used by Florida's urbanized areas - except for the simultaneous approach. The most utilized model is the post-distribution LOGIT formulation of the form: MSt=A+l-A-B 1 + ec(au +D) Where: MSt = Modal Split to Transit Mode A= Transit Captives Fraction B= Highway Captives Fraction c= Slope parameter which determines elasticity of the LOGIT curve D= Point of Symmetry of the LOCIT curve The calibration consists of determining the values for terms A, B, C and D. In most cases the Au, or disutility term, is expressed as: nu = Transit Impedance - Auto Impedance The values forming the transit and auto impedance are both expressed in terms of time in minutes. Impedances considered are line haul travel time, excess time and out-of-pocket cost converted to time. For the automobile, the line haul time is defined to be the time spent in the auto during travel. The excess time is composed of access, parking and egress time. The out-of- pocket costs consist of parking costs, tolls and a perceived automobile operating cost per mile. 4

12 For transit, the line haul time is the actual time spent on the transit mode. The excess time consists of access time (auto or walk), transfer time, waiting time and egress time. Cost consists of transit fare and station parking fees. The costs mentioned above are converted to equivalent minutes through the function: ACostmin = ACOStcnts X 60 X 2000 X 1 Income Where: 2000 hrs./year is assumed number of working hours to convert annual income to hourly income. 60 min./hr is used to convert hourly cost to costs in minutes. Annual income in cents/year. The function thereby allows income to be an input parameter. In most Florida applications, a single income value was used for an entire urbanized area. To allow for proper sensi- tivity this value should be estimated by zone. The total disutility Au is then the sum of the (transit - auto) value differences for line haul time, excess time and cost, or: Au = (Line Haul Time A ) + 2.5fExcess Time A) (Cost A) The coefficients of 2.5 and 3.0 are a result of research on the value of time. Indications are that these parameters are reasonably stable for urban areas. The model coefficients A, B, C and D are calibrated by urban area by trip purpose. This usually considers the trip purposes of Home-Based Work (HBW), Home-Based Non-Work (HBNW) and Non-Home-Based (NHB), but can be derived for other 5

13 purposes. For example, in Jacksonville, Home-Based School was an "other" purpose. Input data requirements for the LOGIT model used in several Florida urbanized areas include: Transit Network Highway Network Person Trip Table by Purpose Cost Functions: Operating Cost/Mile - Auto Parking Cost by Purpose Transit Fares Income The LOGIT formulation has been calibrated for the fol- lowing Florida areas: Miami Broward Pinellas Tampa Jacksonville Orlando Pensacola Tallahassee Basically, all the model forms are the same but values for the parameters A,. B, C and D vary. There is a very slight difference in weighting of parameters for the disutility function in Orlando. Other differences include use of dif- fering numbers of purposes and special situations such as differing model parameters for Miami's beach and non-beach areas. West Palm Beach utilizes a modal share model of the multinomial logit formulation, which considers parameters similar to that described above. The form of the model and variables used are shown below for non-work trips: 6

14 MSP = 1 l+e(a*apa+b) + e(c*dtt+d*difc+e) e(a*apa+b) MSD = l+e(a+apa+b) + e(c*dtt+d*difc+e) e(c*dtt+d*difc+e) MST = l+e(a+apa+b) + e(c*dtt+d*difc+e) Where: MSP = Probability of Making an Auto Passenger Trip; MSD = Probability of Making an Auto Driver Trip: MST = Probability of Making a Transit Passenger Trip; APA = Autos Per Adult; A B = Coefficient of APA; = Constant; DTT = Difference in Total Time; C = Coefficient of DTT; DIFC = Difference in Cost; D E = Coefficient of DIFC; and = Constant. 7

15 Other Post-Distribution Model The Polk study utilizes a cross-classification post-dis- tribution model. Three purposes are considered: HBW, HBNW, and NHB. A transit coverage variable is used considering: trip interchanges not covered by transit service; trip interchanges covered by transit service where one end is in a major transportation attraction zone; trip interchanges covered by transit service where neither end is in a major transportation attraction zone. A third variable considers the number of autos per dwelling unit (D.U.) at the production end of resident D.U.'S. Transient D.U.*s are also considered. An example cross-classification matrix is shown in Figure 2. Cross-Clasification Pre-Distribution The Sarasota-Manatee area utilizes a mode choice model based upon use of a cross-classification of transit trips per household based on auto ownership and income range by zone at the production end. Rates are developed and controlled based on an areawide rate of transit system users. The general indices are shown in Figure 3. These are multiplied by an areawide transit trip rate to obtain the final rate table. For example, if 25,000 users are to be served per day, and there are 100,000 dwelling units in the transit service area, the areawide factor for the rate table would be 0.25 (25,000 t 100,000). Trip attractions are handled based on a rate equation consisting of a coefficient to be multiplied by zonal (total) employment used to calculate an index for zone-to-zone 8

16 TRANSI TRIP FACTORS2 PURPOSE YAfWU TRIP INTERCHANGE PRODUCTION ZONE VARIABLE TRIP RESIOENT 0. U.. TRANSIENT PURPOSE TRANSIT SERVICE COVERAGE NO-AUTO D.U. l-auto 0.U. 2.AUTO D.U. i!s(j%je DJJ. Trip Interchanges Not Covered By Transit Service.ooq.ooo.ooo.ooo,000 H&WORK Trip ht. Covered By Trnn* Service One End in a Maj. Trans, Att. zone Neither End in I Maj. Trans. Att. Zn..45&t ootl ,091.ooo,059 Trip Interchanges Not Covered By Transit Service.a00.ooo.OOO.ooo.ooo HB=NONWORK One End in a MaL Trans. Att. fn ,000 Tr!p Covered by Neither End in e Tran. Service Maj. Tranr Att. fn ooo Trip Interchanges Not Covered By Transit Service.oob.OOO,000.ooo.a00 NON-He Trip Interchanges Covered.007 BV Transit Service1 I Note: (1) Factors for Non.HB interchanges cennot be related to D.U. characteristics of the production zone, hence factors do not vary by no. of autos in the zone of production. (2) The above factors were input into UMOOEL modal split program along with trip production rates for each D.U. category which in effect weighted the factors by the share of trips produced by each D.U. category. (3) H8 Nonwork factors were the same for both categories of trip interchanges covered by transit service. FIGURE 2 : MODE CHOICE MODEL FOR POLK STUDY

17 INCOME LEVEL AUTO OWNERSH IF 0 1 Zor rr0re k-. LOW MIDDLE HIGH *To be multiplied by the areawide transit trip rate. FIGURE 3 Em@LE OF TRANSIT TRIP-MAKING INDICES FCR SABASOTA-MANATEE 10

18 transit trip attractions. The coefficient is determined by setting total transit trip attractions equal to total transit trip productions and thereby dividing transit trip productions from the cross-classification model by total employment from the transit service area. Direct Generation of Auto Travel Gainesville, Lee and Brevard generate auto trips direct- ly, so that no modal choice model is used. Summary of Florida Models Figure 4 presents a summary of the mode choice model types that have been calibrated in Florida. Urban Mass Transportation Administration (UMTA) Input Basically, UMTA does not recommend a specific mode choice model to be used. UMODEL has been developed to allow user coded subroutines to be developed for mode choice models. The "Default" model included within UMODEL is a simultaneous mode choice/trip distribution model of the form: Tijm -_ Pi. Aj - e_bii'm Aj e-ekjrn SE Where: Tijm = trips from i to j by mode m; Pi Aj = productions at i (home end); = attractions at j (destination end); e = ; 0 = a calibrated constant that varies by trip purpose; Iijm = a measure of impedance by mode m equal to: (1.0 x in-vehicle time) + (2.5 x excess time) + (trip cost)/ (.035 x. income per minute). 11

19 POST DISTRIBUTION PRE- DISTRIBUTION OTHER MIAMI BROWARD PINELLAS TAMPA JACKSONVILLE WEST PALM BEACH ORLANDO POLK BREVARD PENSACOLA VOLUSIA SARASOTA-MANATEE LEE TALLAHASSEE GAINESVILLE LOGIT LOGIT LOGIT LOGIT LOGIT PROPORTIONAL LOGIT CROSS CLASS LOGIT LOGIT LOGIT CROSS CLASS GENERATE AUTOS GENERATE AUTOS GENERATE AUTOS FIGURE 4 SUMMARY OF MODE CHOICE MODELS 12

20 The mode choice model can be extracted from the above and is of the form: Where: MSt = _@I!$q&ja x 100 MSt = fractional share of trips estimated to use the transit mode t = mode of transit a = mode of auto Other terms as explained previously. This is a logit formulation where impedances for auto and transit are treated individually by the n-ode choice model rather than as difference values. Some of the considerations expressed in discussions with UMTA and FHWA relate to the use of captive riders in the logit formulation and in the use of income in the impedance calculations. It would appear that better representation of conditions would be reflected if these two characteristics were zonal input rather than regional parameters. Also, with regard to disaggregate behavioral models, the current thinking appears to be that logit-type formulations are the most appropriate type formulations for current applications and that they do reflect the behavioral characteristics of the traveler. Logit type models reflect system characteristics such as headway and speed, cost factors such as parking cost and transit fare and user variables such as income. All these factors are related to the traveler's decision making regarding choice of mode for any particular trip purpose. 13

21 Some other thoughts from Federal personnel include: Perhaps only work trips should be considered in mode choice analysis using a logit model formulation. For non-work trips a trip end model may be most appropriate. This is based on the thought that non-work trips often involve family travel of 2, 3, and 4 people traveling together. The logit formulation considers individuals, whereas households or groups of individuals should be considered for non-work trips. One possible solution is the use of an auto occupancy division of the cost function in the disutility portion of auto costs. There is perhaps some reason to stratify work travelers by some economic factor(s) and calibrate separate models by classification. For example, auto ownership or auto ownership combined with workers per household could be utilized to stratify households for the development of separate models. Other possible stratifiers might include population within walking distance to transit versus those not within walking distance. The logit model will not give reasonable results if walk and auto access times are not dealt with carefully (i.e. as zones get larger, average walk time cannot be used for the entire zone). The discussion relative to thoughts from Federal personnel indicate there are differences in approach to mode choice between individuals. There is no consensus relative to approach. AUTO OCCUPANCY There are two basic approaches to auto occupancy estimation that have been used in Florida. The one used most frequently is the development of a factor by trip purpose to be applied to auto person trips. This considers one factor for all trip movements of a particular purpose. For example, auto occupancy factors might look like: HB Work 0.86 HB Shop 0.57 HBSoc-Ret 0.50 HBMisc 0.58 NHB 0.61 All Purposes

22 The second approach uses the "default" model in the pro- gram UMODEL. This model is based on a model used by the Metropolitan Washington Council of Governments. This app- roach considers a cross-classification of occupancy rates considering: three classes of trip purpose; income (1st Quartile, 2nd and 3rd Quartile; 4th Quartile), five classes of destination parking cost; and five classes of trip interchange density. Trip interchange density is equal to highway person trips between two zones divided by the product of the residence zone area and the non-residence zone area. 15

23 Chapter Two FORMULATION OF ALTERNATIVE The purpose of this chapter is to Florida Modal Choice Models to assist PROCEDURES categorize the current in determining appro- priate modal choice models for application. Using the findings of Chapter One and a categorization of Florida models, a determination will be made of the alternative modal choice models to be developed. Additionally, specifications will be developed to allow accuracy evaluations to be made in comparison of model results with ridership counts made by corridor, cutline, trip end areas and other measures. CATEGORIZATION OF EXISTING FLORIDA MODELS An evaluation of modal choice model application as utilized in general throughout the United States indicates that large urbanized areas utilize post distribution models that consider system characteristics as well as rider characteristics. The logit formulation is currently the model form that is most useful for such analysis. A description of the logit formulation has been previously presented in Chapter One. Variations in the usage of the logit formulation were also noted in Chapter One. In addition to the proportional logit formulation calibrated in West Palm Beach, Orlando uses a model that is sensitive to some zonal variables. A Summary of the models used in the state is shown in Table 1. 17

24 TABLE 1 CHOICE MODELS AND OTHER CHARACTERISTICS FOR FLORIDA STUDY AREAS Urban Area Est % Growth Population (000'S) Per Year Number Buses Model Used1 Miami 1503 Broward 921 Pinellas 710 Tampa 615 Jacksonville 583 West Palm Beach 536 Orlando 624 Folk 285 Brevard 261 Pensacola 231 volusia 190 Sarasota-Manatee 185 Lee 181 Tallahassee 139 Gainesville 137 < 2% > 4% 2-4% > 4% < 2% > 4% 2-4% 2-4% < 2% 2-4% 2-4% > 4% > 4% 2-4% 2-4% 810 Logit 101 Logit 120 Logit 80 Logit 200 Logit 47 prop. Logit 54 Logit 3 Cross-Class2 * Gen. Auto 27 Logit * Logit 36 Cross-Class3 * Gen. Auto 35 Logit 31 Gen. Auto FOOTNOTES: IAll Logit models shown are post distribution models. 2This is a post distribution model. 3This is a pre-distribution model, 4Tallahassee calibrated a logit model but used a procedure based on policy determinations. *Data not available. 18

25 In addition to the logit formulation, which is presently used in most of the larger urbanized areas in Florida, the table indicates that a variety of other techniques have been used in Florida. A few areas have not developed modal split models. These areas for the most part generate auto or auto-person trips. In Gainesville, a procedure has been adopted which permits policy decision to provide a "modal split" technique. This technique identifies subareas within the urbanized area that should attract high transit usage. Transit usage for these areas is estimated and then subtracted from the input trip tables. Tallahassee also adopted a similar policy sensitive modal a target split percentage for lished, and the model determined ted to result in this target. ALTERNATIVES TO BE CONSIDERED split model. In this the entire area was which trips would be model, estab- selec- An evaluation of the models utilized in Florida ized areas, consideration of UMTA and FHWA comments, evaluation of logical need for systemwide modal choice sis has resulted in a determination of alternatives urbanand an analyto be considered. Also among the issues considered in the determination of the recommended modal choice alternatives was Florida's experience with the logit formulation. The fact that this expression has been used in several urban areas provides strong impetus to utilizing it as an alternative. Also, con- sidering the data variables used for highway and transit 19

26 networks, and the trip purposes and cross-class variables used for trip generation leads to a selection of a modal choice alternatives which would not put additional burdens on local study staffs if possible in data forecasting. This means that some consideration must be given to expressions which do not require forecasting income. Bus transit studies or the long range planning of systems where fixed capital investments are not required usually do not warrant ultra-sophistication in the choice of mode share modeling techniques. The modal choice modeling, where non-fixed rolling stock is the alternative to be considered in systems planning is basically a mechanism to determine fleet size and physical facilities and the impact on highway needs. In systems such as this, modal split modeling tech- niques that are not system sensitive can be used. These techniques are most appropriate for use in the smaller urbanized areas. RECOMMENDATION FOf! MODAL CHOICE MODELS The standard trip generation model will produce internal person trips by five purposes. It is possible using the standard generation model set to utilize a modal choice model of either the post distribution or pre-distribution type. It is recommended that three modal choice models alternatives be tested. One model would be a logit model of the form currently in use in Florida, but with some recommended changes. The second model would be a cross-classification model that would consider the type of variables used in the trip 20

27 generation model. These include autos per dwelling unit, persons per dwelling unit and type of unit at the production end and employment by category at the destination end. A model of the type used in the Folk Study, where supply of transit is also a variable is a good example of a cross-classification rate procedure utilizing the trip generation para- meters plus some measure of transit service. This second choice of a modeling technique would be non-system sensitive and as such would not require a coded transit network. This leads to the recommendation of two procedures. One would require explicit network coding regarding the transit system; thereby allowing the testing and analysis of policy type variables such as level of service and fare structures. This model is of course system sensitive. The logit model formulation is recommended based on its previous use in the state, the evaluation of the alternatives available, and dis- cussions with Federal personnel. Some recommended modifica- tions are described later. The second procedure would not require a coded transit network and would thus be non-system sensitive. This procedure would consider population related characteristics such as considered for trip generation and a measure of the availability of transit service to determine split characteristics. The cross-classification post distribution model utilized in Polk County is recommended as an alternative to be tested. As an additional alternative in the areas where a nonsystem sensitive model can be used, the technique used in 21

28 Gainesville will be studied to determine if it has use in other areas. The logit formulation would most appropriately be applied in areas with greater than 500,000 population and/or with more than 50 buses in the forecast year. In the smaller areas and/or with smaller bus fleets the non-system sensitive model would be used. The logit modal choice models already calibrated will allow evaluation of the A, B, C and D parameters to determine possible groupings and allow and evaluation of accuracies now being achieved. It is recommended that three general im- provements to the process should be considered for the standard model. First, zonal income rather than an areawide income should be used to convert transit fares, parking costs and other costs to time. Secondly, auto occupancy by trip purpose should be used in the consideration of auto costs to determine a cost per person. Thirdly, the captive riders for auto and transit should be a zonal characteristic rather than one value for the entire area. This project was not intended to support the type of research required to re-calibrate and meters based on these considerations. posed that an already calibrated model determine new However, it is for one study para- pro- area, whose areawide parameters were used, be applied with esti- mates of zone level income, auto occupancies and estimates of captives by zone. The results will be compared with base data and results of previous calibration runs to determine if 22

29 improvements in results are achieved. If so, this will be the recommended procedure. If not, the existing modal choice models for the larger areas will be incorporated in the stan- dard model set. It is important to note that there is a great need to recalibrate the model split models in all of the areas. TO further test the sensitivity of zonal as opposed to areawide estimates of model inputs, the opposite test will also be conducted. In an area, such as Orlando, where zonal inputs are currently used, the modal split model will be applied using areawide estimates of the model inputs. The results of using these inputs will be compared with the results obtained from zonal inputs. The major effort will be to develop the procedure for the smaller urban areas utilizing the procedures of the Folk Study (See Figure 2 on page 9). It is recommended that tables similar to that shown in Figure 2 be developed for Gainesville and the results of the two areas (Polk and Gainesville) compared for consistency and possible transferability to other areas which would not use the LOGIT model. Polk and Gainesville provide a good population spread for areas under 500,000 population (See Table 1). Also, data for Gainesville were used for trip generation and should be in good condition for this use. The effort for the smaller areas will also attempt to determine if the policy sensitive models developed in Gainesville and Tallahassee are usable in other areas. 23

30 ACCURACY EVALUATIONS The mode choice estimates produced by the standard model set must be evaluated to determine their ability to reflect reality. This will be accomplished as part of the validation work effort. Basically, the validation involves comparisons of model results with field observations. So mode choice estimates would be compared with patronage information. This patronage information is scheduled to be gathered as has been presented in the CENVAL effort.* The validation for 1980 conditions to be accomplished as part of the CENVAL effort will not be possible within this work effort since data will not be available until sometime after completion of this contract. Calibration will be accomplished as part of this effort in the sense of comparing model application results with some base year conditions. The comparisons are made utilizing travel survey and field information, if available, for the base year. The comparison process will make use of available data documenting current or base year transit usage and compare these figures with the results obtained from the appli- cation of the modal split model. These comparisons will be made at the following levels: Areawide Transit Usage Cutline Comparisons Screenline Comparisons Route Comparisons * Utilization of 1980 CeIlSUS Data in Urban Travel EStimation,l Final Report, COMSIS Corporation, September,

31 Care must be taken that the modal split model is being evaluated and not the results of the transit assignment which might be accomplished in a subsequent, step. Some urban areas may not accomplish a transit assignment since they do not necessarily require a transit network as input. Therefore, route comparisons would actually be a check of both the modal split and transit assignment, where this latter step is accomplished. The areawide transit comparison is basic and would be accomplished in both validation (against field data) and in calibration (usually against survey data). Screenline comparisons are again basic for validation and calibration and would check major directional movements across the entire study area. Cutline comparisons are most valid as an assignment check where routes are grouped into major corridors and comparisons made against ground counts. However, where survey data is available, the survey data as well as the model results may be compared across screenlines. What must be carefully evaluated, however, is the fact that transit use is very low in smaller urban areas and is not accurate at this level. On-board survey results are most appropriate for these type comparisons when available. Route checks can be made for calibration by assigning survey data and model data and comparing results on a routeby-route basis. The same consideration made above for cutlines is even more significant here. For the areas using the non-system sensitive approach, there generally will not be a network for assignment purposes. 25

32 In these cases comparisons will be on an areawide basis and across screenlines. For those using the logit formulation, the four types of checks can be made, realizing, however, that the individual route and cutline checks will also be a check of assignment results. There are no standards of accuracy available for modal choice model estimates, as is the case for highway traffic assignments. Therefore, the accuracy standards that have been established for use in validating highway assignments have been transferred to the application of modal split models. The standards established below are also based upon experience, review of both highway and transit results and intuition. Basically, the same types of accuracy obtained in person trip distribution should be obtained in mode choice estimates while keeping in mind the differences in the volume sizes between the two types of travel. Also, the use of the estimates is taken into account. For non-capital intensive systems, long range planning generally is accomplished to "size" the bus system in terms of maintenance and storage facilities as well as number of buses. The individual route demands except as it may help size the facility needs are not of primary importance. Likewise, where transit assignments are not made, the major reason for acceptable mode choice estimation is to insure accurate highway trips and assignments. 26

33 For calibration purposes, the range of error that should not be exceeded for transit trips are presented below: Total Area Transit Trips: Trips Entering the Central Area: Sectors of the Central Area Boundary: (perhaps 4-5 corridors) Total Area Transit Average Trip Length: Cutlines and/or Routes: (1000 Passengers/Day: Passengers/Day: Passengers/Day: SOOO-10,000 Passengers/Day: lo,ooo-20,000 Passengers/Day: >20,000 Passengers/Day: < 3% error < 5% error <20% error < 5% error <150% error <90% error (70% error (45% error (35% error (30% error For validation in a year other than used for calibration the following is recommended: Total Area Transit Trips: Trips Entering the Central Area: Sectors of the Central Area Boundary: Total Area Transit Average Trip Length: Cutlines and/or Routes: < 6% error (10% error (25% error < 8% error Same as above 27

34 CHAPTER THREE EVALUATION OF ALTERNATIVE MODAL SPLIT PROCEDURES The purpose of Chapter Three is to evaluate the alternative modal split procedures advanced in Chapter Two, and to determine which of these alternatives is most appropriate for use in each of the four test areas determined in Phase I of this effort. This evaluation will develop a recommended modal split procedure for each of the urbanized areas in the state. The evaluation documented in this chapter was designed to determine the ability of the alternative modal split procedures advanced in Chapter Two to respond to the issue requirements of the four test areas. This evaluation and analysis will permit the determination of a recommended model set for each of the urbanized areas in the state. REVIEW OF MODAL SPLIT MODEL CATEGORIZATION In Chapter Two, a categorization of the modal split models currently used in Florida was accomplished. The results of the categorization indicates that in the fifteen urbanized areas in the state, a large variety of modal split models are used, This large variety of model choices has been utilized as a source of models from which models appro- priate for standardization can be drawn. As can be seen from Chapter Two, a large variety of modal split techniques are in use in the state. This indi- cates that a large amount of effort has been applied in Flor- ida to addressing a problem of making logical decisions as to the future course of public transportation in the state. It 29

35 also indicates that available experience in Florida is not being fully utilized as new modal split techniques are de- vised. REVIEW OF RECOMMENDED MODAL SPLIT ALTERNATIVES Three general recommendations of modal split models to be evaluated further were advanced as a part of the effort documented in Chapter Two. These are: 1) The logit model with some modifications to permit more standardization in its application and use and to permit policy sensitivity; 2) A non-system sensitive post distribution Cross Classification modal split model similar to that used in the Polk study; and 3) A non-system sensitive Policy Modal Split Model that would permit simulation of high transit usage scenarios. Logit Model Of these three alternatives, the logit model is current- ly used in the majority of the urbanized areas in the state. A thorough technical discussion of the logit model and its functions has previously been presented in Chapter One. Non-System Sensitive Non-Policy Sensitive Post Distribution Cross Classification Model A table provided in Chapter One illustrates the concept behind the post-distribution cross classification model cur- rently used in Polk. This is the cross classification model previously recommended for further evaluation in this effort. As can be seen from the table which was presented as Figure 2, modal split ratios or transit trip factors are determined in Polk based upon whether or not transit service is avail- 30

36 able to the trip interchange and also based upon certain density definitions of the attraction and production ends of the trip and also based upon trip purpose. The computer set-up used to apply the Polk modal split procedure is shown in Appendix I. This set-up is labeled PL931PO.DISCOMOD.2430.CNTL. This set-up is used to build form trip distribution, trip There is also a 2552 version. and update highway skims, per- purpose, combinations and modal split. There are a number of steps in this procedure as detailed below. 1) STEP0 2) s2 3) AGM 4) STEP1 5) STEP2 6) STEP3 7) STEP4 8) STEP5 9) STEP6 UROAD to build time and distance skims. UMODEL to update time skims. AGM to distribute person trips. UMCON to convert tables to FHWA format. TRPTAB to create HB work trip tables for zones covered and not covered by transit service. TRPTAB to separate HB work trips covered into those with and without major attractions. TRPTAB to create HB Non-work trip tables for zones covered and not covered by transit service. TRPTAB to separate HB Non-work trips covered into those with and without major attraction. TRPTAB to create one dataset with 6 tables - HB work - covered, not covered and covered with major attraction; HB Non-work - covered, not covered and covered with major attraction. 10) TESTMCON- UMCON to convert from FHWA to UTPS format. 11) USRCD - USERCODE of split procedures. 12) MODSPLT - UMODEL to apply previous step. 31

37 13) STEP7-14) STEP8 - TRPTAB to separate NHB trips into those covered and not covered by transit. Also mod-split percentages and auto occupancy factors are applied. UMCON to convert NHB and IE trip tables to UTPS format. 15) Tl 16) T2 - UMATRIX to balance (P-A to O-D) and combine NHB, truck, IE and EE trips. UMATRIX to balance (P-A to O-D) and combine HB trips. Non-System Sensitive - Policy Sensitive Modal Split Model The third modal split modeling scheme recqmmended for further evaluation in Chapter Two was the policy sensitive model currently under development in the Gainesville area. This model functions by establishing "desired" modal split fractions at certain trip ends. Trip interchanges involving the targeted trip ends are modified to result in the desired modal split or transit usage at that trip end. This process is heavily dependent upon "hand manipulation" to achieve its desired results. This type of modeling effort is extremely important to the transportation planning effort in a number of urbanized areas in the state where the populace and the decision making bodies have become aware of the need to increase transit ridership to alleviate several of the socio/ economic problems that will confront them in the coming decades. The computer set-up used to apply this procedure is shown in Appendix II. This set-up is located on PL931RD.GA. DISCOMOD.CNTL. This set-up also builds and updates highway skims, performs trip distribution, trip purpose combination, 32

38 and modal split. The function of each of the program steps is detailed below. 1) UROAD - 2) s2 3) AGM UROAD to build time and distance skims. UMODEL to update skims AGM to distribute trips 4) UMATRIXl - UMATRIX to add EE trips and balance tables (P-A to O-D). 5) UMCON - UMCON to perform modal split. Split factors are read in from dataset of the form PL93lGA.UMCON.SPLIT-.DATA 6) UMATRIXZ - UMATRIX to create an auto trip table and a transit passenger trip table AREA SENSITIVE EVALUATION OF RECOMMENDED PROCEDURES The three recommended modal split procedures must be evaluated to determine their usefulness as a standardized mo- dal split procedure in each of the various Florida urbanized areas or in a selected sub-set of these urbanized areas. Based on this evaluation, and subject to the intent to test each of the various procedures that might be recommended, test areas for each of the various procedures will be established. It is important to note that the tests accomplished as a part of this effort will in all probability not provide the final answer as to the appropriateness of the standardized modal split procedures to be recommended. The lack of data available to completely evaluate the models' functions will most likely not be alleviated until the Census Validation Phase III activities now scheduled for 1982 are accomplished. The tests accomplished as a part of this effort will attempt to define the limits of the proposed modal split schemes and 33

39 to develop standardized procedures that can be further defined following collection of adequate data in the Census Validation effort. The first step in the evaluation of each of the proposed standardized modal split models was to determine if the proposed models are able to respond to the issue requirements in the various urbanized areas in the state. This determination was made in each of the test areas used in Phase I of this study. (These areas are Broward, Orlando, Tallahassee, and Tampa Bay.) The evaluation of the various models' ability to respond to the issues in these test areas was based upon discussions with DOT personnel and local officials in the various test areas. Broward County The Broward Urban Area Transportation Study currently uses a logit model to perform modal split modeling. The particular model calibrated in Broward County was developed from the Miami model with some modification to simulate transit travel in Broward County. The resultant model simulated 23,682 daily transit trips, which was 8.3 percent lower than the 1976 daily patronage. BY purpose# the model assigned 47.3 percent of the transit trips to homebased work, 46.5 percent to homebased other, and 6.2 percent to non-homebased. Local officials in Broward County and officials of the DOT agree that the logit formulation has been and in all liklihood will continue to be an adequate tool for performing modal split in this area. The post-distribution cross classifi- 34

40 cation alternative and the policy-sensitive alternative both have features that are attractive to planners in the Broward area; however, the general feeling is that while these models would not be sufficient for systems planning, they might provide a valuable tool for supplementing the systems effort in special conditions. The concensus is that the logit formulation should be used in the area, but that the other expressions should be made available for use if they are needed. Broward County is now participating in a study being conducted by the Florida DOT to determine the need for a regional transit system. The general feeling in the area is that the logit expression is the only alternative under consideration that could be used to perform the modal split function in this area. Orlando The Orlando Urbanized Area Transportation Study (OUATS) utilizes a modal split model of the logit formula that was originally calibrated in This model was updated and minor changes to the model's functions were made adequately model certain conditions. The logit model Orlando is different than those used in other areas in 1978 to more used in in the state in that internal calculations in the model which calculate the disutility functions utilize zonal income data, and also because the highway and transit captive portions for a particular trip interchange are related to the income at the production end of the trip. Information on transit and auto captives were based on attitudinal surveys conducted in 1973 and repeated in

41 In Orlando, it appears that the logit model is adequate to respond to the systems planning issues. Some questions have been raised in the past as to the Orlando model's results, particularly in specific corridors, but generally, the reaction to the modal split results has been favorable. Orlando has a series of future alternative transit networks which have already been developed. These networks will reduce the transit network coding requirements in future updates of the study. Although Orlando has a number of "special attractors", it is felt that the existing logit model has adequately modeled modal split for these attractions. The area is in the process of studying at least one major future transit corridor to determine if fixed guideway rail service should be provided a- long this corridor. The last update of the OUATS did not show the level of transit service on this corridor that many thought it should have. The general feeling is that the determination of the need for more sophisticated service in this corridor should be made a part of a corridor analysis procedure and not as a direct result of a systems planning effort. The review in Orlando has led to the conclusions that a standardized expression of the logit formulation would be adequate to handle the systems planning requirements. The ability to make modifications to this process should not be eliminated and some desire to see a process similar to the Gainesville model available in unique areas was expressed. 36

42 Tallahassee The Tallahassee Urbanized Area Transportation Study has developed both a logit formulation and a policy-sensitive modal split process. Dissatisfaction with the results of the calibration of the logit model, a consultant's recommendation that the model should not be used for long-range planning, and local policy makers' desire to see what would happen if large modal split percentages could be obtained led to the development of a policy-sensitive model for the area. The policy-sensitive modal split model was designed to produce a target split percentage for the entire area. Specific trips were dependent upon trip purpose, zonal income, and the magnitude of the areawide split. The model was not sys- tem-sensitive, however transit trips were not allowed in zones which did not have transit service. Modal splits as high as 25 percent were tested in the study. In general, satisfaction was expressed with the results of this model and the plan adopted had a target modal split percentage of 5 percent. General feeling now in the area is that sophisticated system-sensitive modal split procedures such as the proposed logit formulation should not be used in Tallahassee. The ability to maintain policy maker participation in the process is thought to be important, and thus the area seems to favor an approach like the one previously used in the area or like the one now being developed in Gainesville. These models would adequately meet the technical needs of the area and would also permit policy participation in the effort. 37

43 Tampa Bay The Tampa Bay study effort has calibrated a logit model to perform the modal split function. This model is used in the regional study effort but different values of the elasticity and bias coefficients are used in the three transit service areas in the region. As with the other large study areas evaluated, local users of the logit formulation in the Tampa Bay area appear satisfied with its results and are hesitant to substitute other models, particularly a model that is non-system-sensitive. General interest in the ability to make modifications in the results of the logit model are expressed, usually in an effort to obtain results in special studies, Satisfaction in the models' ability to simulate HOV lane usage was noted in the past experiences in using the existing logit formulation. Some reservation in the logit formulation's ability to produce adequate results in special study efforts was expressed. Such a special study effort is currently underway in a Tampa Central Business District transit study. Generally, however, it was agreed that this probably was not a function that a standardized systems modal split model should perform. RECOMMENDATIONS Current use of the logit formulation to perform modal split is pervasive in Florida, particularly in the larger urbanized areas. The evaluation of the issues impacting modal split decisions generally tend to indicate that in these larger areas, some sort of modal split procedure which is system- 38

44 sensitive should be used. In the smaller areas in the state, the general feeling is that the more sophisticated modal split procedures are not needed and that other procedures such as the Polk County procedure would be adequate to perform the modal split function. In both the larger and in the smaller areas, some type of policy input to the modal split process was seen as desirable and an important element to be maintained as a part of the standardized modal split procedure. These considerations, the past experiences of the various urbanized areas in the state along with an appraisal of the possible future functions of a modal split model led to the preliminary recommendation of two separate models for use in the state. The first of these models would be the logit formulation. To permit standardization, different formulations of the logit expression such as that currently used in West Palm Beach and use of non-standard variables as is currently done in the Orlando effort would be eliminated. Calibrated values of the A, B, C, and D parameters that are currently used in the various areas should continue to be used until better data becomes available as a part of the Cen- sus Validation effort. West Palm Beach should borrow cali- brated A, B, C, and D parameters from another area, probably from the Broward area. In Orlando, survey results should be used to develop area-wide parameter values. The areas that should use the logit formulation are: Broward, Jacksonville, Miami, Orlando, Pinellas, Tampa, and West Palm Beach. These areas are a preliminary recommendation 39