Characteristics of Premium Transit Services that Affect Choice of Mode (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

J-1 A p p e n d i x J Model Implementation and Calibration Contents J-1 Overview J-2 Existing Salt Lake City Models J-5 Incorporating Transit Amenities and Service Characteristics J-6 Transit Mode Definition and Path-Finding J-9 Transit Mode Choice Utility Expressions J-11 Transit Path Choice Model Calibration J-13 Comparative Results Overview The purpose of the model implementation and calibration was to apply the scaled marginal rates of substitution or values in equivalent minutes of in-vehicle travel time (presented in the previous section) in a standard practice travel model and modify the mode choice model structure to recognize path choices rather than technology mode choices.The estimated values of non-traditional transit service attributes from the MaxDiff models were incorporated into an existing travel model’s transit path-building and mode choice components to demonstrate their applicability in practice. This appendix describes in detail a process that was developed to implement and calibrate a mode choice model that accounts for the influence of non-traditional or premium transit service attributes. The travel model for the Salt Lake City region was chosen for this demonstration. Transit mode shares for the Salt Lake City region are provided in FIGURE J-1. The Salt Lake City region is encompassed by two MPOs, which are the Wasatch Front Regional Council (WFRC) and the Mountainland Association of Governments (MAG). The two MPO planning areas are adjacent, and the agencies utilize the same travel model. The WFRC/MAG model region encompasses four counties—Weber, Davis, Salt Lake, and Utah. Bounded by the Great Salt Lake on the west and Wasatch mountain range on the east, the region is relatively narrow. Hence, most of the transit travel is in the north-south direction to and from Salt Lake City which is centrally located in the region. The transit system mainly consists of five service types—local bus, express/fast bus, bus rapid transit (BRT), light rail transit (LRT), and commuter rail transit (CRT) which together service approximately 150,000 boardings on a typical weekday.

J-2 Characteristics of Premium Transit Services that Affect Choice of Mode SLC: Salt Lake City FIGURE J-1. Regional overview of transit system in Salt Lake City. Existing Salt Lake City Models Before discussing the specifics of how the research methods were deployed and tested in Salt Lake City, it is useful context to understand the current model, and specifically aspects related to transit modeling. The following is a brief introduction to the existing transit path- finding and transit mode choice modeling processes, prior to any enhancements as part of this research effort. The existing WFRC/MAG travel model has been implemented in Citilab’s “CUBE” software environment and “TRNBUILD” is used for transit path-building and assignment. TRNBUILD constructs transit paths using transit routes and various path-building parameters as inputs. TABLE J-1 shows the parameters used by the transit path-building process in the existing model. TABLE J-1. Path-building parameters: existing model. Path-Building Parameter Value *IVTT: in-vehicle travel time For each origin-destination (OD) pair, the path builder attempts to build up to 10 unique transit paths that are mode-specific, developing one path for each of the five “primary” modes—

Model Implementation and Calibration J-3 local bus, express bus, BRT, LRT, and commuter rail—by each of the two access types (walk and drive). For each of the 10 path types, the shortest path is skimmed for every OD pair. These mode-based paths are available as alternatives for transit trips in the mode choice model. There is no formal kiss-and-ride (KNR) access modeled. TABLE J-2 defines the mode hierarchy for primary mode identification for a specific transit path. To illustrate, a transit path involving LRT and BRT would be called an LRT path because light rail is highest in the hierarchy of modes in the transit path. In addition, the transit paths are built for two time periods—peak (6 a.m. to 9 a.m. and 3 p.m. to 6 p.m.) and off-peak (the rest of the day). A nested multinomial logit mode choice model is used to estimate the split among auto, walk/bike, and transit trips. The nesting structure and model parameters are shown in FIGURE J-2. The transit trips are further split into one of the 10 transit paths identified by the path builder. TABLE J-2. Mode hierarchy by transit path type Mode Allowed Local Bus BRT Express Bus Light Rail Commuter Rail P at h ty pe Local bus Required No No No No BRT Yes Required No No No Express bus Yes Yes Required Yes No Light rail Yes Yes No Required No Commuter rail Yes Yes No Yes Required Source: Wasatch Front Regional Council Mode Choice Model Documentation FIGURE J-2. Existing mode choice model for Salt Lake City.

J-4 Characteristics of premium Transit Services that Affect Choice of Mode The mode choice model has been estimated separately for four trip purposes: (1) home- based work (HBW), (2) home-based other (HBO), (3) home-based college (HBC), and (4) non- home-based (NHB). All trips are segmented into three classes based on access to transit in the origin zone. The three access-to-transit segments are “no access to transit,” “must drive to transit” (no walk access transit available), and “can walk to transit.” Based on the access-to- transit segment of a particular trip, the available transit choices are determined. For example, the “must drive to transit” segment does not have a walk access transit mode available. There is further demographic segmentation applied to HBW and HBO trips. Prior to applying the mode choice model, HBW and HBO trips are segmented based on three vehicle-ownership classes and two income classes. The three vehicle-ownership classes are zero-, one-, and two- or more vehicle households for which separate transit alternative specific constants have been calibrated. The two income categories distinguish households in the lowest income quartile from households in the three higher income quartiles. The income category affects the cost coefficient in the mode choice model. The model coefficients were originally estimated and calibrated using an estimation dataset blended from home interview survey data and transit on-board survey data, with appropriate model skims appended to each survey. In some cases, parameters have been adjusted based on professional judgment and experience with the model. TABLE J-3 shows the mode choice model coefficients used in the existing model. TABLE J-3. Existing mode choice model coefficients for Salt Lake City. Variable HBW HBO NHB HBC In-vehicle time (minutes) -0.0221 -0.0160 -0.0233 -0.0221 Initial wait (minutes) -0.0442 -0.0320 -0.0466 -0.0442 Transfer wait (minutes) -0.0500 -0.0480 -0.0663 -0.0500 Drive access time (minutes) -0.0332 -0.0240 -0.0350 -0.0332 Walk time (1st mile) (minutes) -0.0442 -0.0320 -0.0466 -0.0442 Walk time (> 1 mile) (minutes) -0.0663 -0.0480 -0.0699 -0.0663 Cost—low income (cents) -0.0099 -0.0120 -0.0049 -0.0060 Cost—higher income (cents) -0.0023 -0.0040 -0.0049 -0.0060 Premium direct walk—bus 0.1105 0.0800 0.1165 0.1105 Premium direct walk—rail 0.2210 0.1600 0.2330 0.2210 Drive access distance divided by auto path distance -0.3315 -0.24 -0.3495 -0.3315 Transfers -0.265 -0.192 -0.280 -0.265 1/(transit distance)—walk -1 -1 -1 0 1/(transit distance)—drive -3 -3 -3 -1 Urbanization (density) at attraction end 0.0044 0.0032 0.0047 0

Model implementation and Calibration J-5 For the purpose of this project, the current mode choice model was recalibrated to a more recent transit on-board survey. The transit survey was conducted system-wide (including all the five transit modes) in the spring of 2011. It contained about 7,100 valid records. The calibration involved adjusting the alternative specific constants in the existing model to match boarding counts from the on-board survey by primary and access modes. In addition, boarding counts by transit route were also validated during this process. The calibration results of both the existing model and the new transit path choice model are provided in the comparative results section at the end of this chapter. Incorporating Transit Amenities and Service Characteristics The implementation of the methods explored in this research to incorporate non- traditional attributes into a standard practice travel model involved considerations around how to revise the transit path builder, the definition of transit modes, and the mode choice model utility expressions. As a first step in this model refinement process, an assessment of the availability of non- traditional or premium transit service characteristics for the transit system in the Salt Lake City region was made. Data pertaining to park-and-ride lots, station/stop shelter and seating, and route level on-time performance information were obtained from the local agencies. Other service information about stations/stops such as lighting/safety, security, and proximity to services was not available or was deemed too anecdotal and approximate to be useful. In the Salt Lake City region the on-board amenities were not available at a route level, but the perception among local transit agency staff was that variation in amenities and service characteristics among services was more obvious at the “mode” level (or between service types), than it was at the route level. Ideally, the path builder should account for all the information on premium transit characteristics at the appropriate level. For example, stop shelter and seating would ideally vary by stop; on-board amenities may vary by route, etc. The way in which Salt Lake City represented their supply system within TRNBUILD does not allow the user to easily apply node- or transit- route-specific penalties or benefits. It is possible to implement stop-specific penalties through careful access link coding leading up to each specific transit stop, but that was not needed for this proof of concept. Further, while route level path-building rules aren’t feasible, “similar” routes could be combined into more “modes” to incorporate more variation in path-building rules, parameters, and weights. The parameters that can affect path-building in TRNBUILD are initial/transfer/access/egress time weights, boarding penalties, and in-vehicle time weights. Most of these parameters can be directly applied by mode. Hence, as a work-around for this project, mode-specific composite premium transit characteristic benefits were computed and applied as boarding penalties. TABLE J-4 shows the asserted premium transit attributes at the mode level based on knowledge of transit system of the region. For each premium transit attribute, the values in terms of IVTT minutes were first obtained by averaging the scaled values from Chicago and Charlotte surveys in Phase 2 for commute trips for both bus and train. The values from the Salt Lake City survey were not used because the survey had changed from Phase 1 to Phase 2, and it was felt that survey data obtained in Phase 2 had better information from a methodological standpoint. Not all the premium transit service attributes measured in the MaxDiff models were available for

J-6 Characteristics of premium Transit Services that Affect Choice of Mode the existing model. Hence, the values of attributes that were available were scaled by each bundle of premium attributes to reflect the full benefit that could potentially be gained from premium transit characteristics. For example, if only shelter and bench information was available in the “station amenities” bundle, which add up to 1.13 minutes, those values were scaled up to the full value of the bundle (when all other attributes were included), which is 4.33 minutes (see the “Value” and “Scaled Value” columns of TABLE J-4). The benefits were then converted to mode-specific relative penalties that could be applied at each boarding by the path builder. In addition, the estimated value of perceived reduction in the in-vehicle time in a premium mode was used for path-building and mode choice. It was applied to all modes except local bus again based on local knowledge of the transit system. Transit Mode Definition and Path-Finding An objective of this research was to explore the switch from mode-specific path-building to service preference related path-building. A separate travel time analysis conducted with the Chicago and Charlotte model networks indicated that there were inaccuracies in comparing reported and network-based travel times and paths. This travel time analysis is documented in Appendix I. To achieve this objective, mode labels on paths were removed from the existing model. However, access mode distinction (walk/drive) was considered important and retained in path- building. To obtain an optimal set of path-building parameters, an exercise was conducted in which path-building parameters were systematically varied. Two additional parameters to incorporate values of premium transit service attributes were introduced into path-building. One, called “non-premium service boarding penalty factor,” was used to weight the relative boarding penalty (shown in TABLE J-4) that is applied based on a specific mode boarding. The other, called “premium service in-vehicle travel time factor,” was used to weight the premium service IVTT percent reduction (21% as shown in the table above). The systematic variation of path- building parameters produced 243 paths each for walk and drive access segments. The specific number of paths generated was to keep the run times reasonable. Judgment was used in choosing the parameters to be varied and their values. TABLE J-5 shows the path-building parameters that were varied systematically and their values. TABLE J-4. Mode level values of premium transit service attributes. Bundled Aribute Premium Service Aribute CRT LRT LOCAL EXP BRT Value (min. of IVTT) Scaled Value (min. of IVTT) Staon amenies Shelter x 0.75 2.88 Bench x 0.38 1.45 Lot count x x 0.00 0 On board amenies On board seang availability x x 1.81 2.90 Producvity features x x x 0.82 1.32 Vehicle cleanliness x x 0.62 0.99

Model implementation and Calibration J-7 Bundled Aribute Premium Service Aribute CRT LRT LOCAL EXP BRT Value (min. of IVTT) Scaled Value (min. of IVTT) Other service features Reliability x x 5.12 7.79 Mid day schedule span x 0.32 0.49 Evening schedule span x 0.32 0.49 Vehicle ease of boarding x x 0.14 0.22 Fare machines x x 0.69 1.06 IVTT with premium (percent reducon in IVT) 21% 21% 0 21% 21% Premium Benefit (minutes) 11.0 9.5 2.5 2.6 8.3 Scaled Premium Benefit (minutes) 19.6 17.3 3.9 6.6 15.4 Relave Non premium service boarding penalty 0 2.3 15.7 13 4.2 TABLE J-5. Path-building parameter ranges for systematic variation. TABLE J-4. (Continued). Path Building Parameter Value Initial/transfer wait time weight 1, 1.5, 2 x IVTT* Access/egress time weight 1, 1.5, 2 x IVTT Transfer penalty 0, 5, 10 min. Non-premium service boarding penalty weight 0.5, 1, 1.5 x boarding penalty Premium service IVTT weight 0.5, 0.75, 1 x perceived IVTT reduction *IVTT: in-vehicle travel time A total of 486 transit paths (243 each for walk access and drive access) were analyzed to obtain a set of paths that matched best with the on-board survey data. A match in the modeled path was said to be found if the modes and routes in sequence involved in the path were the same as those reported in the survey between a specific OD TAZ pair. A combination of three sets of parameters that resulted in the highest match (about 76%) for both walk and drive access was derived by conducting a matching exercise. The choice of three paths was determined by judgment based on the matching analysis. The number of combinations to be analyzed increases exponentially with the number of parameter sets in a particular combination. For obtaining the best combination of three-parameter sets, 114 million combinations (C(486,3) = 114,083,640) were analyzed. TABLE J-6 shows the optimal three-parameter sets obtained for walk and drive access. Based on the weights in the parameter sets, labels on traveler preferences have been assigned. For example, if the weight for access/egress time is high, this may imply that travelers falling under this segment prefer shorter access times. Another example is if the premium service in-vehicle time weight is relatively low (say 0.5), this probably corresponds to a set of travelers who prefer premium transit service for longer trips. The choice of three paths was determined by judgment based on the matching analysis.

J-8 Characteristics of Premium Transit Services that Affect Choice of Mode TABLE J-6. Path-building parameters for the transit path choice model. W al k Pa th Dr iv e Pa th Traveler Preferences Transfer Penalty Access/ Egress Time Wait Time Non-Premium Service Boarding Penalty Premium Service In- Vehicle Time The nesting structure of the existing mode choice model was changed as shown in FIGURE J-3. Instead of five mode-based paths each under the walk and drive access transit nests, there are now three paths that are more generic and based on traveler preferences (see TABLE J-3) in the new transit path choice model. FIGURE J-3. Transit path choice and mode choice model for Salt Lake City.

Model implementation and Calibration J-9 An issue of interest prior to implementing the transit path and mode choice model was to assess the number of “unique” paths that were being produced as a result of the modified path- generating process. Given that an overlap in paths implies a possible correlation among the attributes of the path choices, this may lead to the violation of the independence of irrelevant alternative (IIA) assumption in multinomial logit models. TABLE J-7 shows the proportion of paths overlapping in the survey dataset based on the paths built. An overlap here is same as the matching criteria specified earlier. Two paths are said to be overlapping/duplicates if the modes and routes involved in one path are the same as the modes involved in another path. TABLE J-7. Overlapping path choices in the transit path choice model. Number of Overlapping paths Number of Survey Records (%) Number of Overlapping Paths Number of Survey Records (%) W al k Ac ce ss 3 53 (2%) D riv e Ac ce ss 3 72 (2%) 2 1,363 (42%) 2 1,062 (36%) 0 1,862 (57%) 0 1,826 (62%) Total 3,278 (100%) Total 2,960 (100%) TABLE J-7 shows that there is an overlap in the path choice alternatives for about 40% of the records. The definition of a duplicate path was revised from matching modes to matching exact transit routes. In other words, two paths would be called overlapping/duplicates only if they both have the same transit routes in the same order of boarding. If two paths involved the same set of modes but different routes, they were to be considered as separate choices. The original method of matching modes was too general and many duplicate paths were identified that were effectively different. This method was not implemented as a result. The process to match transit routes involved modifying the path-building script to print out detailed transit paths (along with the routes involved in the path) in log files for all origin destination pairs. An offline process involving a Python script (originally created by MTC staff) to parse the log files for the transit route information and create files that could be imported back into the TP+ environment was developed. The route information is then processed to check for and remove duplicate paths by additional TP+ scripts before mode choice model is run. Transit Mode Choice Utility Expressions After generating a set of unique path choices for each OD pair, the next step was the modification of transit mode choice utility expressions. The existing mode choice model involved calculation of 10 utilities corresponding to the 10 transit paths generated (five paths each for walk access and drive access). The research mode choice model involved calculation of only six utilities (three paths each for walk access and drive access) in comparison to the existing model. Other changes to the utility expressions were to incorporate the effects of boarding non- premium transit modes (non-premium service boarding penalty) and traveling in a premium mode (premium service IVTT), similar to the way they were incorporated in the path-building process. The non-premium service boarding penalties were measured in terms of IVTT minutes and were added to the utility expression after they were factored by the IVTT coefficient. For this part of the utility expression, the number of boardings by mode for each transit path was

J-10 Characteristics of premium Transit Services that Affect Choice of Mode needed. The number of boardings by mode for each transit path was calculated during the process of identifying and removing duplicate paths. The “direct walk premium” component (see TABLE J-3) was dropped from the utility expression in the transit path choice model. This component adds to the utility of a transit path if it involves direct walk access to a premium service (all modes except local bus are considered as premium). It was felt that this component was no longer necessary in the transit path choice model because the benefits of premium boarding premium services were being incorporated using the non-premium service boarding penalties. The premium service IVTT factor was directly introduced in the utility expression as a part of the IVTT of a transit path. A factor of 0.8 was used to represent the perceived reduction in IVTT as a result of using a premium service. There was an interest in exploring the influence of sociodemographic characteristics on the choice of transit paths to reflect the differences in the survey of travelers who chose different paths. To this end, an analysis was done to examine path choice behavior along several demographic dimensions, such as income group, age, gender, auto ownership, licensed drivers, etc. Specifically, path choice behavior included critical aspects of walk access and drive access choices, such as access and egress times, transfer times, premium service characteristics, etc. Because the reported path choice components from the survey may not always be accurate, modeled components from transit skims were also used. The average path characteristic values were compared across various categories of each demographic attribute. For example, the average number of transfers was compared across the different income categories available. As a result of this analysis, it was found in the data that younger people have a higher average walk time (combined access and egress). Specifically, the data imply that: Persons under 18 years of age are only willing (allowed) to walk up to 1 mile Persons ages 18 to 44 years are willing to walk up to 2 miles Persons ages 45 to 64 years are willing to walk up to 1¼ miles Persons ages 65 and older are only willing to walk up to 0.5 miles TABLE J-8 shows the walk time coefficients and scaling factors used. This suggested that an additional demographic segmentation in the mode choice model based on age might help improve the model. The segmentation based on age was added to the existing mode choice model for HBW and HBO trip purposes. The distribution of age segments was obtained from Census 2010 data for the relevant counties. The influence of age on walk times was accounted for by scaling the walk time coefficient up or down based on the age category. The factors for scaling were obtained from mode choice models estimated using Salt Lake City survey data in Phase I of this project. The scaling factor was the ratio between the walk time coefficient in a model with segmentation and one in a model with no age segmentation.

Model implementation and Calibration J-11 TABLE J-8. Age segmentation in transit path choice model. Variable HBW HBO NHB HBC Walk time (1st mile)—age < 45 years 0.96*-0.0442 0.96*-0.0320 -0.0466 -0.0442 Walk time (> 1 mile)—age < 45 years 0.96*-0.0663 0.96*-0.0480 -0.0699 -0.0663 Walk time (1st mile)—age 45 to 64 years 1.04*-0.0442 1.04*-0.0320 -0.0466 -0.0442 Walk time (> 1 mile)—age 45 to 64 years 1.04*-0.0663 1.04*-0.0480 -0.0699 -0.0663 Walk time (1st mile)—age > 65 years 1.71*-0.0442 1.71*-0.0320 -0.0466 -0.0442 Walk time (> 1 mile)—age > 65 years 1.71*-0.0663 1.71*-0.0480 -0.0699 -0.0663 Addition of age segmentation to the model was found to have a significant impact on transit travel from TAZs with high proportions of people in the higher age categories, as shown in TABLE J-9. For example, it was found that in TAZs which had between 50% and 60% of people over the age of 65 years, there was a reduction of 13% to 29% in the number of walk access transit trips when compared to the model without age segmentation. However, at a more aggregate level, it was found that including age segmentation did not affect the overall model results significantly. This was probably due to the lower proportion of people falling in the higher age categories in most of the TAZs in the region. Due to the insignificant impact of age segmentation on the overall model results and also due to the fact that the age segmented model has a 50% higher runtime, it was decided to drop the age segmentation for calibration purposes. Hence, calibration efforts were focused on a model without age segmentation. Transit Path Choice Model Calibration The ultimate goal of the calibration of transit path choice model was to compare the constants with those in the calibrated existing model and analyze the impacts of involving premium service characteristics on the constants. The new transit path choice model (with modified path and mode choice process) used the same demand and supply inputs as the existing model. The calibration targets also were the same and had been obtained from the on-board survey conducted in 2011. At the beginning of the calibration process, the mode level alternative specific constants in the transit path choice model were set to zero. The motive behind this was to analyze the results from the transit path choice model which incorporated the effects of premium service attributes without any modal biases that were used to calibrate the existing model. Subsequently, there would be modal biases or constants introduced if required. It should be noted that the non- premium service boarding penalties are at a mode level and may themselves be interpreted as modal constants, albeit of a different kind. The difference is that these constants can be explained based on the finding from the maximum difference models. However, the mode level non- premium service boarding penalties will be accounted for while comparing the constants between existing and transit path choice models.

J-12 Characteristics of premium Transit Services that Affect Choice of Mode The calibration primarily involved adjusting overall transit constant. The number of transit paths built changed from 10 in the existing model to six in the transit path choice model. As can be expected, this reduced the transit logsum which in turn reduced the overall transit share of trips when the new research mode choice model was run. Appropriate adjustments were made to the overall transit constant to match the total transit trips target. The same adjustments were made to all trip purposes and the two time periods. There was an imbalance found between the walk access versus drive access transit trips in transit path choice model. It was theorized that the non-premium service boarding penalties were quite significant and additional transfer penalties were overkill for walk access transit trips. Hence, transfer penalties were dropped from the utility expression. For drive access transit trips, it is conceivable that transfers in the transit path are still quite onerous due to the burden of driving, parking, and transferring being already involved. Therefore, the transfer disutility for drive access trips was retained in the utility expression. In addition to this, the premium IVTT factor was adjusted from 0.8 to 0.9 to slightly lower the transit trips on premium services. Finally, a few minor adjustments were made to express bus and LRT mode-specific biases to match the mode level targets more closely. TABLE J-10 shows the coefficients used in the research mode choice model. TABLE J-9. Impact of age segmentation on select trips. P_TAZ A_TAZ % Age 65+ Transit Walk Trips 841 921 66% -29% 867 921 56% -13% 938 921 52% -13% 1352 921 61% -22% TABLE J-10. Research mode choice model coefficients for Salt Lake City. Variable HBW HBO NHB HBC In-vehicle me (IVT) (local) -0.0221 -0.0160 -0.0233 -0.0221 In-vehicle me (IVT) (premium) 0.9HBW IVT coeff. 0.9HBO IVT coeff. 0.9NHB IVT coeff. 0.9HBC IVT coeff. Walk me (1st mile) -0.0442 -0.0320 -0.0466 -0.0442 Walk me (> 1 mile) -0.0663 -0.0480 -0.0699 -0.0663 Transfers—walk access 0 0 0 0 Transfers—drive access -0.265 -0.192 -0.280 -0.265 Premium service characteriscs—local 16HBW IVT coeff. 16HBO IVT coeff. 16NHB IVT coeff. 16HBC IVT coeff. Premium service characteriscs—BRT 4HBW IVT coeff. 4HBO IVT coeff. 4NHB IVT coeff. 4HBC IVT coeff. Premium service characteriscs—express 13HBW IVT coeff. 13HBO IVT coeff. 13NHB IVT coeff. 13HBC IVT coeff. Premium service characteriscs—LRT 2HBW IVT coeff. 2HBO IVT coeff. 2NHB IVT coeff. 2HBC IVT coeff.

Model implementation and Calibration J-13 It should be noted here that because this is a research effort a full-scale recalibration of the model was not attempted. The focus was more on matching transit trips by mode and transit trips by purpose (i.e., marginal distributions) and less on matching transit trips by both mode and purpose (i.e., joint distributions) or boardings by route and mode. Comparative Results The transit path choice model was calibrated to approximately the same level of accuracy as the existing model within a reasonable amount of time (approximately 1 week). TABLE J-11 shows a comparison of the number of linked transit trips in the existing and transit path choice models to targets obtained from the on-board survey data by access and primary mode (based on mode hierarchy in the existing model). Linked transit trips represent a trip from the origin to the destination. Boardings represent a trip from the access station to the egress station. The comparison shows that the existing and transit path choice models are more or less equally close to the targets even though it appears that the results from the existing model are “closer.” It should be noted again that the existing model had been well calibrated prior to this project, whereas for the transit path choice model significantly less amount of time was spent on calibration. TABLE J-11. Comparison of linked transit trips by access and primary mode. Access Primary Mode Survey Existing Model Transit Path Choice Model Walk CRT 1,200 900 1,800 Express 2,900 2,800 3,700 LRT 25,500 25,000 22,400 BRT 1,500 1,000 1,000 Local 34,200 37,200 33,900 Drive CRT 4,700 5,000 4,000 Express 4,900 4,100 4,100 LRT 17,700 17,900 18,800 BRT 100 700 600 Local 6,200 4,200 3,800 TABLE J-12 shows the comparison of linked transit trips in further detail by separating the primary mode path based on whether or not it involves a local bus boarding. Similar conclusions could be drawn about the existing and transit path choice models as those drawn based on TABLE J-11.

J-14 Characteristics of premium Transit Services that Affect Choice of Mode TABLE J-12. Comparison of linked transit trips by access and detailed mode. Access Mode Survey Exisng Model Transit Path Choice Model Local 34,200 37,200 33,900 BRT 500 300 500 BRT local 1,000 700 500 Express 2,100 2,000 2,200 Walk Express local 800 800 1,500 LRT 12,800 14,900 15,900 LRT local 12,700 10,100 6,500 CRT 400 200 400 CRT local 800 700 1,400 Local 6,200 4,200 3,800 BRT 100 200 300 BRT local 0 500 300 Express 4,300 3,400 3,600 Drive Express local 600 700 500 LRT 16,000 14,800 15,300 LRT local 1,700 3,100 3,500 CRT 2,800 1,300 2,600 CRT local 1,900 3,700 1,400 TABLE J-13 provides a comparison of route boardings from the existing and transit path choice models with observed boarding counts. For conciseness, all the boardings were aggregated to a “route group” level from the individual route level. It appears that the existing and transit path choice models are both close to each other in terms of matching observed route level boarding counts. In some cases, it may happen that the existing model matches the target better on one set of routes (for example 33rd South, WE-SL Express, etc.) and the transit path choice model matches target better on another set of routes (for example Parleys/Millcreek, 17/21, 13th East, etc.). In other cases, the transit path choice model overestimates boardings by the same amount as the existing model underestimates or vice versa (for example, Kearns/WVC, LRT, etc.). Both models are underestimating the total number of transit boardings. Overall, at a higher level, both models are calibrated to more or less an equal extent. The details of path level benefits as a result of various components of utility expressions in both existing and transit path choice models are presented in TABLE J-14 and TABLE J-15 for the home-based-work trip purpose. The path level benefits are calculated by converting all the relevant coefficients applied in utility expressions of the paths to a common unit (IVTT minutes in this case). It is useful to compare both alternative specific constants and other fixed parameters between existing and transit path choice models because the effects of premium service attributes have been added at mode level in the transit path choice model. The other fixed parameters included were transfer penalty, direct walk benefit, and boarding penalty. The transfer penalty is a penalty applied to the utility of a transit path for each transfer made in the

Model implementation and Calibration J-15 TABLE J-13. Comparison of transit boardings by route group. Route Group Counts Existing Model Transit Path Choice Model Local bus 13th East 2,983 3,462 2,619 17/21 2,773 3,208 2,819 2 to U 2,323 2,900 1,268 33rd South 1,815 1,617 1,135 45th & 39th 7,836 5,853 4,526 Avenues 2,267 894 742 Kearns/WVC 4,702 5,087 4,590 Magna 1,491 1,046 522 Misc SLC 813 1,888 2,169 Parleys/Millcreek 1,005 1,419 1,013 Redwood 4,434 4,536 3,432 Rose Park 3,378 828 757 S. Davis 316 389 308 Sandy/Midvale 4,215 5,434 5,554 SL 3rd-5th East 3,653 2,127 1,951 State 5,472 3,440 2,080 UT_local 11,617 10,627 10,456 WE/DA intercity 6,388 6,324 5,927 WE/N Davis 8,977 5,781 5,884 West Jordan EW 617 846 689 BRT S35MAX 3,358 2,517 4,554 Express/Fast SL Fast 2,000 1,941 1,656 Tooele 813 185 125 UT-SL 3,997 3,086 4,984 WE-SL Express 1,342 1,342 975 LRT 47,923 45,205 49,399 CRT 5,300 5,898 5,946 Total 141,808 127,880 126,079

J-16 Characteristics of premium Transit Services that Affect Choice of Mode TABLE J-14. Path benefits (IVTT minutes) in existing model for work trips. Path Composion Relave ASC Transfer Penalty Direct Walk Total Relave ASC Transfer Penalty Total W al k Ac ce ss Local 0 0 0 0 D riv e Ac ce ss 0 0 0 BRT 17 0 5 22 17 0 17 LRT 33 0 10 43 33 0 33 Express 33 0 5 38 33 0 33 CRT 43 0 10 53 43 0 43 Local-local 0 -12 0 -12 0 -12 -12 Local-BRT 17 -12 0 5 17 -12 5 Local-LRT 33 -12 0 21 33 -12 21 Local-express 33 -12 0 21 33 -12 21 Local-CRT 43 -12 0 31 43 -12 31 BRT-local 17 -12 5 10 17 -12 5 LRT-local 33 -12 10 31 33 -12 21 Express-local 33 -12 5 26 33 -12 21 CRT-local 43 -12 10 41 43 -12 31 Local-express-LRT 33 -24 0 9 33 -24 9 TABLE J-15. Path benefits (IVTT minutes) in transit path choice model for work trips. Path Composion Relave ASC Transfer Penalty Boarding Penalty Total Total Shied Relave ASC Transfer Penalty Boarding Penalty Total Total Shied W al k Ac ce ss Local 0 0 -16 -16 0 D riv e Ac ce ss 0 0 -16 -16 0 BRT 0 0 -4 -4 12 0 0 -4 -4 12 LRT 14 0 -2 11 27 0 0 -2 -2 14 Express 0 0 -13 -13 0 9 0 -13 -4 12 CRT 0 0 0 0 16 0 0 0 0 16 Local-local 0 0 -31 -31 -15 0 -12 -31 -43 -27 Local-BRT 0 0 -20 -20 -4 0 -12 -20 -32 -16 Local-LRT 14 0 -18 -4 12 0 -12 -18 -30 -14 Local-express 0 0 -29 -29 -13 9 -12 -29 -32 -16 Local-CRT 0 0 -16 -16 0 0 -12 -16 -28 -12 BRT-local 0 0 -20 -20 -4 0 -12 -20 -32 -16 LRT-local 14 0 -18 -4 12 0 -12 -18 -30 -14 Express-local 0 0 -29 -29 -13 9 -12 -29 -32 -16 CRT-local 0 0 -16 -16 0 0 -12 -16 -28 -12 Local-express-LRT 0 0 -31 -31 -15 0 -24 -31 -55 -39

Model implementation and Calibration J-17 transit path. The coefficient of the transfer penalty has been converted to equivalent IVTT minutes by dividing it by IVTT coefficient. The direct walk benefit is applied to utility of a path if it involves directly walking to one of the four premium transit modes (express bus, LRT, BRT, and CRT). This, too, has been converted to equivalent minutes of IVTT. Finally, the boarding penalty represents the non-premium service boarding penalties (see TABLE J-15) applied to the path utility based on the number of boardings for each mode (only in the transit path choice model). TABLE J-14 and TABLE J-15 show the total path benefits in terms of IVTT minutes for 15 paths each for walk access and drive access broken down by their contributing components. For the sake of brevity, paths comprising all mode combinations are not presented. In the case of the transit path choice model, the total path benefits were shifted to set the benefit of “Local” path to zero so that all of them can be compared to the path benefits in the existing model. TABLE J-16 provides a comparison of both the total path benefits and path level biases (which result from alternate/mode-specific constants added to the utilities): If only the alternative specific constants (in terms of IVTT minutes) in both models are considered, quite clearly the transit path choice model has far fewer of them both in number and magnitude. TABLE J-16. Comparison of path benefits (IVTT minutes) by detailed mode for work trips. Exisng Research Shi Exisng Research Exisng Research Shi Exisng Research W al k Ac ce ss Local 0 0 0 0 D riv e Ac ce ss 0 0 0 0 BRT 22 12 17 0 17 12 17 0 LRT 43 27 33 14 33 14 33 0 Express 38 0 33 0 33 12 33 9 CRT 53 16 43 0 43 16 43 0 Local-local -12 -15 0 0 -12 -27 0 0 Local-BRT 5 -4 17 0 5 -16 17 0 Local-LRT 21 12 33 14 21 -14 33 0 Local-express 21 -13 33 0 21 -16 33 9 Local-CRT 31 0 43 0 31 -12 43 0 BRT-local 10 -4 17 0 5 -16 17 0 LRT-local 31 12 33 14 21 -14 33 0 Express-local 26 -13 33 0 21 -16 33 9 CRT-local 41 0 43 0 31 -12 43 0 Local-express-LRT 9 -15 33 0 9 -39 33 0 When total path benefits (total fixed effects) are compared, it can be seen that for simple paths involving one service, the total effects (in terms of IVTT minutes) in the transit path choice model are significantly lower than those in the existing model.

J-18 Characteristics of premium Transit Services that Affect Choice of Mode In more complex paths (involving two or more services), the total effects in the existing model are positive because of the hierarchical nature of the transit paths. On the other hand, in the case of the transit path choice model, the total effects for complex paths are negative, primarily because of the inclusion of non-premium service boarding penalties wherein boarding a non-premium service is penalized irrespective of the involvement of a premium mode (or a mode higher in the hierarchy). In the existing model, this effect is hidden or subsumed by mode hierarchy. Overall, the transit path choice model appears to be a more intuitive and realistic representation of transit path choice behavior. These results are detailed for the HBW trip purpose, but were also compiled for non- work trip purposes. The path benefits calculated for other trip purposes showed similar patterns because all of the calibration adjustments were applied uniformly to all purposes. As a proof of concept, this was sufficient, but additional calibration specific to non-work purposes would be needed to draw further conclusions about the non-work purposes.

Abbreviations and acronyms used without definitions in TRB publications: A4A Airlines for America AAAE American Association of Airport Executives AASHO American Association of State Highway Officials AASHTO American Association of State Highway and Transportation Officials ACI–NA Airports Council International–North America ACRP Airport Cooperative Research Program ADA Americans with Disabilities Act APTA American Public Transportation Association ASCE American Society of Civil Engineers ASME American Society of Mechanical Engineers ASTM American Society for Testing and Materials ATA American Trucking Associations CTAA Community Transportation Association of America CTBSSP Commercial Truck and Bus Safety Synthesis Program DHS Department of Homeland Security DOE Department of Energy EPA Environmental Protection Agency FAA Federal Aviation Administration FHWA Federal Highway Administration FMCSA Federal Motor Carrier Safety Administration FRA Federal Railroad Administration FTA Federal Transit Administration HMCRP Hazardous Materials Cooperative Research Program IEEE Institute of Electrical and Electronics Engineers ISTEA Intermodal Surface Transportation Efficiency Act of 1991 ITE Institute of Transportation Engineers MAP-21 Moving Ahead for Progress in the 21st Century Act (2012) NASA National Aeronautics and Space Administration NASAO National Association of State Aviation Officials NCFRP National Cooperative Freight Research Program NCHRP National Cooperative Highway Research Program NHTSA National Highway Traffic Safety Administration NTSB National Transportation Safety Board PHMSA Pipeline and Hazardous Materials Safety Administration RITA Research and Innovative Technology Administration SAE Society of Automotive Engineers SAFETEA-LU Safe, Accountable, Flexible, Efficient Transportation Equity Act: A Legacy for Users (2005) TCRP Transit Cooperative Research Program TEA-21 Transportation Equity Act for the 21st Century (1998) TRB Transportation Research Board TSA Transportation Security Administration U.S.DOT United States Department of Transportation