Improving ADA Complementary Paratransit Demand Estimation (2007)

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.

1. Factors Influencing the Demand for ADA Paratransit 2. Development of Recommendations for Tool Development Factors Influencing the Demand for ADA Paratransit As a prelude to proposing specific tools for estimating demand, the team has researched factors that influence the demand for ADA paratransit. This research has been conducted through an exten- sive literature review and by means of a survey of paratransit practitioners, advocates, and riders. This chapter presents the findings of the literature review followed by the findings of the survey. Literature Review The literature review includes: • Basic travel demand theory as it would apply to ADA paratransit. • Impacts of paratransit service design and delivery based mainly on prior work by the research team. • How personal characteristics affect travel by people with disabilities based on empirical and theoretical research. • Limited evidence in the literature about demand by specific subgroups of people with disabilities. • Limited evidence in the literature about mode choice by people with disabilities. • Evidence from FTA Compliance Reviews. • A summary of other literature concerning methodology, survey results, and trends. Travel Demand Theory A review of basic travel demand theory as described in Domencich and McFadden (1975 and 1996) and Glaister (1981), as well as theoretical work by one of the research team (Lewis, 1983), provides some insights as summarized in Figure 1. Only one of these works dealt specifically with people with disabilities. It finds that, in the most fundamental terms, people with disabilities behave according to the same rational travel demand principles as other people. Impacts of Service Design and Delivery There have been several studies that specifically addressed the impact of service design and delivery on the demand for ADA or other types of paratransit. These studies include several econometric analyses that used time-series data and one that used a cross-sectional data set. 78 A P P E N D I X C Excerpts from the First Interim Report (May 2005)

Excerpts from the First Interim Report (May 2005) 79 Results from these models, including several developed by members of the research team, have been summarized in Figure 2. For a number of factors, a range of elasticities is available from these models, while for others less precise statements are possible based on survey research. Personal and Trip Characteristics Several studies have specifically examined how personal characteristics affect travel, including specifically the travel of people with disabilities. Figure 3 summarizes these results. Demand by Specific Subgroups In the survey of paratransit practitioners, advocates, and riders, many respondents noted that it is important to distinguish among various types of disabilities. One group that is often noted as making large numbers of trips consists of people with developmental disabilities, including mental retardation. A paper from 1986 (Starks, 1986) describes how trends in ser- vices for mentally retarded people are affecting travel demand. Starks notes the following trends: • A marked increase in the quantity and availability of a great variety of services; • A strong programmatic emphasis on the delivery of these services in community rather than institutional settings; • Increasing decentralization of delivery of these services; • A propensity of mentally retarded persons to use these services extensively. The paper finds that these trends result in a demand for transportation that is particularly con- centrated and exceeds that of elderly or physically disabled persons. Starks provides some data about the demographic factors that underlie the travel demand of mentally retarded persons and data about their distinctive travel patterns. Factor Short run /long run Degree of influence Comments specific to this factor Utility Both Strong: Travel time and discomfort are a disutility that consumers seek to minimize. An important factor in making hypotheses about role of paratransit relative to car and relative to fixed-route transit in creating greater mobility. Socio- economic characteristics Both Strong Travel demand has been shown to stem from demand for activities away from home, the consumption of which is strongly linked to disposable income. Gender is tied to trip frequency, with females engaging in more trip chaining. Modal characteristics Both Strong Car ownership, a function of income, is strongly linked to mode choice. Bus travel has been found to be an “inferior good” (demand declines with increasing income). Tastes Both Strong Lewis reports that disabled peoples’ value of time spent in a segregated (i.e., paratransit) rather than an integrated (i.e., fixed-route) public transit vehicle does not outweigh the value of door-to-door convenience. Implies limited ability of accessible fixed-route to draw people with disabilities away from paratransit (assuming equal fares). Domencich and McFadden (1975 and 1996), Glaister (1981), Lewis (1983) Figure 1. Insights from travel demand theory.

80 Improving ADA Complementary Paratransit Demand Estimation Noland et al. (2004) examined differences in total trip making for various types of disabilities using multivariate analysis of travel diary data and found significant lower total trip making for people with difficulty walking, people with difficulty understanding directions, and especially for those who use a wheelchair. However, this analysis was not limited to people with disabilities that prevent use of public transportation. Mode Choice There have been a few attempts to apply mode choice modeling techniques to travel by peo- ple with disabilities. For example Stern (1993) used a correlated multinomial logit model and a Poisson regression model to measure the factors affecting demand for different types of trans- portation by elderly and disabled people in rural Virginia. The major results were: • A paratransit system providing door-to-door service is highly valued by transportation- disabled people. • Taxis are probably a potential but inferior alternative even when subsidized. Buses are a poor alternative, especially in rural areas where distances to bus stops may be long. • Making buses accessible would have a statistically significant but small effect on mode choice. Factor Short run/long run Degree of relative influence Comments specific to this factor Paratransit Eligibility Long term Strongest among paratransit factors if choosing between ADA and broader (non-ADA) eligibility. Analysis that includes eligibility is over 20 years old. Limited evidence indicates a strong impact of variations in ADA eligibility. Paratransit Advance request time Short- medium term The impact of offering same- day or advance reservations has been show to be very strong (20% to 30%). Has not been addressed in post- ADA studies. No evidence concerning one-day vs. longer advance reservations. Paratransit Fare Short- Medium Estimated fare elasticity = -0.2 to -0.8. Some evidence of elasticity over –1.0 when fare levels are high (i.e., ridership impact is greater at higher fares). Good evidence from recent time series studies. Impact of fares appears to be greater when systems have no denials. Secular Trend (Annual growth in trip requests after controlling for influence of fare and level of service) Short to Medium Term Ranges from 0.5% to 1.0% per year in monthly and quarterly time series studies over five to ten years. Measured secular trend similar to expectations based on population growth. Cross- sectional data would give more meaningful long-run impact of population (holding other factors constant). Paratransit Reliability/Predictability Short- Medium Strong in surveys and complaints data but as yet unquantified in multivariate analysis. Measures may include missed trips, late pick-ups, late arrivals. Paratransit Comfort Short- Medium Difficult to measure. Fixed-route accessibility / level of fixed-route service Short- Medium Weak but statistically signifi- cant in one multivariate study. Sources: Hickling Corporation (1991), HLB Decision Economics (1998 – 2004), Lewis (1998, 1992), Crain & Associates and HLB (1995). Moderate in surveys but as yet unquantified in multivariate analysis. Figure 2. Impacts of service design and delivery.

Excerpts from the First Interim Report (May 2005) 81 Factor Short run/long run Degree of relative influence Comments specific to this factor Income Evidence available pertains to long run. No short/medium term evidence of paratransit income elasticity available Measured impact of income about 1.0. Low income almost certainly interacts with disability and low car ownership to drive ADA paratransit demand in the long run. Unclear how disability interacts with income in influencing travel demand and paratransit demand. Disability Evidence available pertains to long run. No short/medium term evidence of paratransit income elasticity available Measured impact of disability on demand is distinct and additive to impact of age. Whether or not disability is distinct and additive to impact of low income of depressing travel demand is key outstanding question Age Evidence available pertains to long run. Strong. Travel demand declines with age, especially after age 75. Car ownership Long run Very strong Car ownership and availability found to be key factor in ADA paratransit demand, but poorly measured due to focus on short term time series data Employment Long run Strong. Those working or studying travel more than those who are retired, who travel more than those unable to work. Activity type Long run Strong in surveys and complaints data but as yet unquantified in multivariate analysis. ADA systems cannot prioritize trip purpose, but individual riders make travel and mode decisions based on purpose. Principal source: Noland et. al. (2004). Also, Lewis (1979, 1983), Hickling Corp. (1991), McFadden (1985). Figure 3. Impact of personal and trip characteristics. • Demand is price inelastic. • The total number of trips taken is insensitive to mode availability and characteristics. These results, while intriguing, cannot be conclusive since they deal with a door-to-door ser- vice that is not limited to ADA-eligible people and that most likely did not comply with ADA capacity constraint requirements. Levine (1997) describes the impact of efforts to manage demand for ADA paratransit, princi- pally by offering free fares on accessible fixed-route transit. The study concludes that elimina- tion of fares on fixed-route transit for people qualifying for ADA paratransit had a moderating impact on the growth of demand for paratransit. Neither information dissemination nor appeals to good citizenship appeared to have any effect on ridership patterns. A more comprehensive review of the impact of free fares on transit on ADA paratransit ridership is presented in Multi- systems and Crain (1997). Work in Sacramento (Franklin and Niemeier, 1998) illustrates mode choice analysis using choice-based samples—an on-board survey of riders with disabilities on Sacramento fixed-route service and ridership records of the paratransit system. The results, however, appear mainly to reflect limitations of the data. The survey on the fixed-route system had no way to identify ADA paratransit eligible riders. The data about paratransit riders did not indicate degree of type of disability, in particular disabilities that entirely prevent use of transit.

82 Improving ADA Complementary Paratransit Demand Estimation Evidence from FTA Compliance Reviews The ADA paratransit compliance reviews conducted for the FTA Office of Civil Rights often address factors that contribute to changes in paratransit riders, or that transit systems have used in an attempt to estimate future changes. Figure 4 provides a brief summary of factors noted in 30 of these reviews as having had an influence on demand at the system under review. Other Evidence Schmoecker et al. (2002) analyzed usage data for a paratransit pilot program in London to determine how user characteristics, fares, trip type, etc., influence the choice between short advance-notice trips and longer advance-notice trips at a lower fare. The analysis demon- strates how travel data by people with disabilities can be analyzed by disaggregate methods. However, the particular choices analyzed are ones typically not available in ADA paratran- sit. The specific results appear to be highly dependent on local circumstances. The biggest influence was found to be users’ remaining monthly trip allowance. This result supports the idea that, in paratransit systems that still have significant numbers of trip denials, these trip denials will be the dominating influence on demand and will mask the influence of other factors. In 2003 the Bureau of Transportation Statistics reported the results of a national survey of 5,000 people, of which 2,241 had disabilities, about use of transportation modes for local and long-distance travel and problems with transportation (BTS, 2003). In principle, the survey data could be used to analyze travel patterns and choices of people with disabilities. However, the Figure 4. Factors influencing demand identified in FTA compliance reviews. Factor Transit system MARTA (Atlanta)High fares compared with other systems Long term growth in population with disabilities due to growth in elderly population ASI (Los Angeles) VIA (San Antonio)Free fare on fixed route Alternative countywide service, more reliable but higher fare. Birmingham Hampton RoadsReduced service area to ADA required area Increased awareness among people with disabilities Other agencies discontinuing service (“shedding”) GHTD (Hartford) MATA (Memphis)Eligibility recertification Availability of statewide Transportation Disadvantaged program (Mentioned by reviewers in assessment of R-GRTA.) Florida WichitaCounty discontinued rural program GainsvilleShift from state TD program More thorough eligibility process Shift to state-funded Shared-Ride program for seniors SEPTA (Philadelphia) San FranciscoStraight line projection )dleifgnirpS(ATVPHistorical market growth Implementation of supplemental taxi program Cutback in state support of transportation for developmentally disabled Stricter ADA eligibility process CTA (Chicago)

Excerpts from the First Interim Report (May 2005) 83 survey was not limited to ADA paratransit eligible people; instead it used the Census definitions of disability, which are much broader. The sample includes only 137 people who used paratran- sit (not necessarily ADA paratransit) in the past month and 76 who used human service agency transportation. In the published report, the most striking result is the very strong influence of driving on paratransit use: 2% of drivers use paratransit compared with 13% of non-drivers. The raw survey data are available for download from BTS. TCRP research conducted by SG Associates (1995) provides an example of a tool for prac- titioners to estimate demand, in this case for rural passenger transportation. Although the tool is applied to a different mode, there is at least one aspect of the method used that may be appropriate for ADA paratransit. In the published demand estimation tool, a distinction is made between “program demand” and “non-program demand.” Non-program demand is estimated entirely on the basis of population in various age and income categories and vehicle miles of service available. Program demand is estimated based on enrollment in 13 different types of programs and demand factors (similar to trip generation factors) for each program type. Previous work by the research team for King County Metro (Crain & Associates, Inc. and HLB, 1995; Koffman and Lewis, 1997) illustrates the combined use of econometric analysis and other methods. A time series, econometric model was successful in estimating the impact on paratransit demand of reducing denial rate, changing fares, and long term growth, but only for individual weekday trips within the service zones that existed at that time. Group trips, weekend trips, and trips between service zones were estimated by other means. King County Metro found that the results matched experience for several years. Recent work for the Orange County Transportation Authority (Menninger-Mayeda, et al., 2005) illustrates both the power and limitations of time-series econometric analysis. The model is very successful at explaining fluctuations in demand due to day of the week, seasons, and hol- idays. This result could be quite useful to systems in planning capacity to accommodate what often appear to be random fluctuations. However, the model does not provide any means of pre- dicting the impact of changes in system policies such as fares, service area, on-time performance, eligibility methods, and so forth. The 1990 Bay Area Regional Paratransit Plan (Crain & Associates, 1990) provides an early example of how data from “exemplary systems” can be used to predict paratransit demand. By comparing trip rates per “transportation disabled” person from a sample of exemplary paratransit systems, the paper developed an estimated per-capital trip rate that would occur on a paratransit system without capacity restraints. The analysis used the counts of “trans- portation disabled” people that were included in the 1980 Census (but not in the 1990 or 2000 Censuses). In order to find a reasonable relationship between paratransit use and trans- portation disabled population, the research separated “program trips” (those generated by human service programs) and “general trips.” Only the general trips proved clearly related to population. Bearse et al. (2003) used National Transit Database ridership reports to examine trends in paratransit ridership between 1980 and 1995. They found about 10.5% annual growth and determined that the rate of growth was 3.6 times as fast as growth in the elderly population and 4.7 times as fast as growth in the general population. An analysis of detailed data from the para- transit system in Charlottesville, Virginia, found that most of the growth there was due to growth in the number of riders rather than increased trip rates per passenger. The analysis found higher than average trip rates for paratransit users with vision disabilities or mental retar- dation and lower than average trip rates for users who live in nursing homes or have kidney problems.

84 Improving ADA Complementary Paratransit Demand Estimation Results from the Survey of Paratransit Practitioners, Advocates, and Riders Paratransit professionals, researchers, advocates, and riders were surveyed to obtain their views about the importance of various factors that affect demand for ADA paratransit. A list of people to invite to complete the survey or to be interviewed was compiled from membership lists of the APTA Access Committee, the TRB Committee on Paratransit, and the TRB Committee on Accessible Transportation and Committee and from lists of people who had attended ADA paratransit training provided by team members. The complete list contained 447 names, of which 410 included e-mail addresses. A subset of 20 names was chosen for telephone interviews that allowed extended, open-ended discussions. Those chosen included panel members; transit agency staff; and advocates representing large urban areas, medium to small urban areas, rural areas, and all geographic regions of the United States. The researchers received 144 usable responses from the web version of the survey and conducted 16 telephone interviews. (Some of those originally identified for phone interviews chose instead to use the web version or could not be contacted.) The largest category of participants was transit sys- tem staff, but there was significant representation from paratransit advocates and advisory com- mittee members and some paratransit riders. Figure 5 shows participants’ self-reported categories. Participants marked an average of 1.35 categories. Participation by riders was lower than desired. In fact many of those who described themselves as riders were paratransit staff people. Since par- ticipation by riders is considered important to the success of the project, a special effort was made to include riders in the investigation of exemplary systems described in Chapter 3. The survey/interview asked participants to rate 30 different factors on a scale from 1 to 5 to indicate how strongly each one influenced the demand for ADA complementary paratransit. The average scores for all the factors are given in Figure 6. The rectangular bars show the average score for each factor and the thick lines at the end of each bar show two standard errors for the aver- ages. (Since this is not a random sample, the bars do not represent a 95% confidence interval, but they do give some idea of the spread of responses.) The top scoring factors were: • On-time performance • Whether programs that provide services for people with disabilities also provide transportation • Ability to get through on the phone to reserve a ride • Denial rate • Some other factors with high scores that are notable include: – Accessible transit service combined with travel training Paratransit Role Number of Participants Percent of Participants Transit system staff planning, managing, or operating paratransit service 112 70% 15%24Consultant or university researcher Staff of a contractor managing or operating paratransit service 13 8% Staff of a planning agency with responsibility for paratransit issues 21 13% 14%22Paratransit advisory committee member 11%18Staff of an advocacy or policy organization 4%7Paratransit rider Figure 5. Survey participants.

Excerpts from the First Interim Report (May 2005) 85 – Accessibility of sidewalks, buildings, and other public facilities • Some factors notable for their relatively low scores include: – Paratransit fare – How far in advance reservations are taken – Length of the on-time pick-up window – Accessible taxicabs in the community Note that denial rate was included as a factor for the sake of completeness, even though denial rate is not an issue in systems that are in full compliance. The survey was not limited to systems How much does each of the following affect ADA paratransit ridership levels? 1 2 3 4 5 On-time performance Whether programs that provide services for people w ith disabilities also provide transportation Ability to get through on the phone to reserve a ride Denial rate Changes to the eligibility process Availability of other specialized transportation Accessible transit service combined w ith travel training Climate (severe hot or cold, snow , rain) Changes in availability of programs for people w ith disabilities Accessibility of sidew alks, buildings, and other public facilities Overall corporate attitude and philosophy Availability of subscription service Community transit services Trip-by-trip eligibility screening Ride duration Driver assistance betw een vehicle and building Improvement of bus or rail service Marketing, outreach, and public information Discounted taxi service Changes in location of programs for people w ith disabilities Paratransit fare Other factors in the community How far in advance reservations are taken Free or low transit fare service for ADA-eligible riders Availability of w ill-call returns Length of the on-time pick-up w indow General economic conditions (e.g., unemployment) Vehicle condition Accessible taxicabs in the community Reliable conventional taxicab service Effect on ADA paratransit ridership levels Figure 6. Summary of survey ratings of factors influencing demand.

86 Improving ADA Complementary Paratransit Demand Estimation believed to be in full compliance, and open-ended comments make it clear that many of the par- ticipants still have significant denial rates. In the open-ended portions of the survey, many participants provided very thoughtful comments: • A frequent comment was that many factors have less impact on demand than they would if ADA paratransit riders had any other options. This view coincides with theoretical expectations and survey research that people with disabilities who do have other travel options, especially as a driver, choose not to use paratransit. • A number of respondents indicated that, as long as a system has capacity constraints, other factors will have a weak impact on demand. This view underlines the importance of focusing on exemplary systems in this project. The comments also indicate that many of the respon- dents based their comments on experience with systems that do have capacity constraints. This fact suggests that the impact of some factors might be understated in the survey. • A number of participants pointed out that different subgroups of people with disabilities will be impacted differently. One phrased this as follows: “There are different ‘markets’ within the group of people using ADA paratransit. The different markets are differentially affected by the factors.” For example the choice between door-to-door and curb-to-curb service could strongly affect very frail people and some people with mental disabilities while having a weaker impact on others. Ride duration could similarly affect certain subgroups more than others. • As with other modes, travel time and reliability are more important for some types of trips and some riders than others. A comparison of ratings between 40 participants who are advocates or riders and the remain- ing participants showed no statistically significant difference in ratings. Survey participants were also asked to rate long run impacts on demand in an open-ended for- mat. The instruction was: “Thinking over the long run (the next 10 to 20 years), what are the things you expect to have the biggest impact on ADA paratransit ridership levels? (Choose three of the above or something else. Please list in order of impact.)” The results are shown in Figure 7, in which participants’ responses have been grouped into categories. The top-rated item, “Increasing number of elderly” is obviously important, although it was not a factor provided in the numerical rating section of the survey. The second most frequently mentioned factor was “Quality/predictability of service provided,” which is consistent with the fact that participants gave “On-time performance” the highest quantitative rating and “Ability to get through on the phone to reserve a ride,” which had the third-highest quantitative rating. Two factors listed frequently concern funding and the cost of providing service. However, for our purposes, these factors can only be an issue indirectly. The purpose of this project is to esti- mate demand for ADA paratransit that fully complies with pertinent regulations. Cost and fund- ing cannot be a reason to avoid complying, but they can influence policy choices within the framework of the regulations, such as the choice of door-to-door and curb-to-curb service, pro- vision of feeder service, and whether to charge the maximum-permitted fare. The factor with the second-highest rating in Figure 6, “Whether programs that provide ser- vices for people with disabilities also provide transportation,” corresponds to “Human service transportation, including coordination,” in the open-ended responses, which was the ninth most-frequently mentioned item. This difference may indicate that participants recognize this as having strongly influenced current demand but believe there is little prospect for change in the future. In contrast, “Denial rate” was rated as a strong influence in Figure 6, possibly indi- cating the major impact of having recently eliminated denials, but the same factor was men- tioned by very few participants as an important future influence, which is consistent with the fact that denial rate is not a future policy choice.

Excerpts from the First Interim Report (May 2005) 87 Development of Recommendations for Tool Development The team has examined potential demand estimation tools based on guidance from the proj- ect Panel, priorities expressed by practitioners and advocates in the survey, and feasibility for development within the scope of this project. Feasibility will depend on the availability of data, the expense of obtaining data, and the level of effort needed to conduct data analysis. Panel Guidance The project Panel has provided guidance in the project statement. Some of the same points cited earlier for selecting exemplary systems apply to the selection of appropriate demand esti- mation tools. In other words, the tools should estimate demand: • Only by those persons who are truly eligible for service, as determined by eligibility process that use best practices in the transit industry. • Only for those trips that these eligible individuals are unable to make by fixed-route service when the fixed-route system complies with the ADA. • For service operated during the same hours as the fixed-route system and operated without capacity constraints. Factor Listed 1st Listed 2nd Listed 3rd Total 88111760Increasing number of elderly Quality / predictability of service provided 12 20 14 46 Availability/level of paratransit operations funding 13 16 13 42 37101611Eligibility criteria and process Funding levels to support convenient fixed-route service 5 7 12 24 Accessibility of sidewalks, buildings and other public facilities 6 10 5 21 201226The cost of providing ADA service 14545Travel training Human service transportation, including coordination 5 3 6 14 14473Urban growth 13175Quantity of service provided 12543Fares 9252Technology improvements 99General health of population 5122Denials 5212Awareness of service 422Increased service areas Availability of other specialized transportation 4 4 Revision of the Federal Regulations on eligibility 1 2 3 211Hybridization of service design 211Curb to curb vs. door to door 11Local agency involvement Number of responses (in order of impact) Figure 7. Long run factors having the biggest impact on ADA paratransit ridership levels.

88 Improving ADA Complementary Paratransit Demand Estimation These criteria can be met by basing tools on the demand observed at exemplary systems, while taking care to distinguish between demand at these systems for ADA paratransit and demand for any non-ADA services offered by these same systems. The problem statement noted that, under the ADA paratransit regulations, transit operators are free to tailor their ADA complementary paratransit operations in response to the communi- ties they serve. The problem statement gives the following examples: • In some systems, complementary paratransit service is provided as a door-to-door service, and, in other systems, it is curb-to-curb; • Systems have different rules regarding trip reservation policies and whether subscription ser- vice is to be provided; and • Systems have different policies and standards regarding on-time performance, on-board travel times, and other performance characteristics. These considerations suggest that the tools should be able to provide demand estimates that are sensitive to these choices. In other words, to the extent that it is possible to do so, the tools should provide a way to estimate the impact on demand of these policy choices. The problem statement notes that effective coordination with other transportation programs in a community can have a significant impact on demand for ADA paratransit services. This sug- gests that demand estimation tools should take these other services into account. Under the heading, “Improved Tools for ADA Complementary Paratransit Demand Estima- tion,” the problem statement identifies as needed: • A better understanding of riders who qualify for this service, their travel patterns, and the likely demand, given the level of service provided. • Efforts for ADA paratransit similar to those devoted to fixed-route transit and other trans- portation services, including research to understand trip-making needs and patterns; choices to use various transportation options (i.e., mode choice); and the effects of various service parameters (i.e., fares, frequency of service, days and hours of operation, and service quality and reliability) on demand. Under the heading, “Future Research on ADA Complementary Paratransit Demand Estima- tion,” the problem statement says that: “developing an accurate understanding of demand for ADA complementary paratransit ser- vices will likely require ongoing research beyond what will be accomplished in this project. A better understanding of the number and percentage of people who are eligible for service will need to be developed. The travel needs of this segment of the population will then need to be studied in more detail (e.g., types of trips needed and trip making rates). Factors that influ- ence potentially eligible individuals to apply for ADA paratransit eligibility and/or to use other transportation options will need to be better understood. The influence of service design parameters (e.g., fares, days and hours of service, on-time performance, and travel times) on demand and trip-making rates will also need to be researched.” The problem statement continues: “Tools also should be available to predict demand, when there are changes in system design or level of service variables change. For example, if the service area is expanded and the total pop- ulation served increases, what will be the impact on demand? If systems provide curb-to-curb rather than door-to-door service, how will that affect the number of riders and trips that will be requested? If the hours of fixed-route operation are extended later into the evening, what effect will that have on demand? If systems have varying levels of reliability and service quality (e.g., on-time performance and on-board travel times), what effect will these have on demand?”

Excerpts from the First Interim Report (May 2005) 89 Subsequent guidance from panel included a request to examine the potential for demand esti- mation tools that can lead toward eventually incorporating ADA paratransit into regional travel demand models and the regional transportation planning process, based on state-of-the- practice and emerging transportation models. The panel also requested consideration of potential applicability to estimating demand for more generalized paratransit. Guidance from the Survey of Paratransit Practitioners, Advocates, and Riders Survey participants were asked: “For your purposes (for planning, management or advocacy), what are the most important factors that should be included in the ridership estimation tools? (Please list up to four factors in order of importance.)” As shown in Figure 8, to a great degree responses mirrored those provided to the question regarding the factors that will have the great- est impact (shown in Figure 7 earlier). However some differences are notable: • “Availability of funding” is much lower-ranked, possibly recognizing that this is a political issue and not a logical input to a demand estimation tool, since availability of funding is not a consideration in meeting the mandate for ADA paratransit. Number of responses (in order of importance) Listed 1st Listed2nd Listed 3rd Listed 4th Total 906132150Population statistics/projections 5210161313Quality / reliability of service 3071193ADA eligibility/certification Accessibility of bus stops and fixed-route service 5 10 10 5 30 Availability of other transportation services 6 8 8 4 26 24699Service area designation 221714Historical data (past patterns etc.) 163472Fare elasticity 151563Cost of providing the service 1421011Travel training 134612Funding levels 1028Number of current eligible riders 6132Community needs 5122Day of week/time of day 5131Trip length 431Location of riders and facilities 4112Population health 44Traffic conditions/trends 413Weather 413Paratransit capacity constraints 312Coordination with area programs 312Average trips taken by each rider 211Availability of housing 211Population income 211Economic growth of the region Figure 8. Most important factors to include in ridership estimation tools.

90 Improving ADA Complementary Paratransit Demand Estimation • “Historical data (past patterns)” was mentioned by many participants, indicating a concern that the tools should take into account local conditions as reflected in this established history. • The mention of “service area designation” reflects a desire to be able see how adjustments in the area served can affect demand. • “Cost of providing the service” in this context presumably indicates a desire for tools to help in determine not just the number of trips but the cost of serving those trips, for example as a result of peaking or trip length. This would overlap with “Day of week/time of day” and “Trip length” which were also mentioned separately. Survey participants were asked, “For your purposes, how far into the future should a useful tool project ridership?” As shown in Figure 9, the majority of respondents (68%) would like tools to project ridership five to ten years into the future. Criteria for Selecting Demand Estimation Tools The research team has combined the guidance from the Panel and the survey participants’ with its own expertise and understanding of the issues to arrive at a proposed set of criteria for selecting the most appropriate methods to develop demand estimation tools in this project. These criteria are: 1. Feasibility within the scope of this project. The tools must be able to be developed using data that is available or can be collected within the schedule and budget of this project, and ana- lytical methods than can be implemented within the schedule and budget of this project. 2. High confidence that the methods will produce an immediately usable tool (or tools) and not just interesting research results. 3. Transferability among regions, taking into account highly varied local conditions and histo- ries. The tools should have widespread applicability in a variety of service areas: large and small cities and rural areas, severe and mild weather, high and low income. They should use local conditions as inputs. No. of respondents Percent of respondents < 1 year 3 2% 1 year 4 3% 2 years 4 3% 3 years 6 5% 3-5 years 12 10% 5 years 41 33% 6 years 4 3% 7 years 1 1% 5-10 years 18 14% 10 years 22 17% 10-20 years 2 2% 15-20 years 1 1% 20 years 6 5% 25 years 1 1% 30 years 1 1% 100%126 Figure 9. How far into the future should a useful tool project ridership?

Excerpts from the First Interim Report (May 2005) 91 4. Level of effort and data requirements to use. The tools should be usable by practitioners with- out extensive additional data collection with reasonable investment of staff time. 5. Technical sophistication required to use. The tools should be usable by transit planning staff and other interested parties without highly specialized expertise. Results should be easily explainable and transparent to policy makers, advocates, and the general public. 6. Limitation to ADA paratransit. The tools should produce estimates of demand for ADA com- plementary paratransit complying with all required service criteria (including the requirement that there be no capacity constraints), consisting of eligible trips by ADA eligible individuals only. 7. Ability to addresses policy issues of interest. The tools should address as many as possible of these issues: • Total ADA paratransit demand in cities that do not currently have ADA-compliant para- transit service. • Growth in ADA paratransit demand based on population increase, other demographic trends, and changes in service coverage. • Impacts of changes in paratransit policies and performance within the bounds of ADA ser- vice criteria: e.g., on-time reliability, fares, door-to-door vs. curb-to-curb operation, tele- phone hold times, availability of subscription service, strictness of eligibility process. • Impact of improvements in fixed-route transit accessibility. • Impact of changes in the availability of specialized transportation services for people with disabilities. • Detail related to determining the cost of providing service (time of day patterns, trip length, type of disability). 8. Relevance to planning in the medium term, i.e., five to ten years in the future. Tools that meet the criteria listed up to this point should in general be relevant in this time frame. Ability to use the tools for exploratory, “what-if” analysis of longer-term trends is also useful. 9. Contribution to increased understanding of travel behavior of people with disabilities. The tools and the development process should provide insights sufficient to guide future research. Analysis of Options for Developing Tools The research team has considered a variety of possible methods for developing demand estima- tion tools in order to recommend one that is most suitable for this project. This section provides a description of these methods and reasons why they are appropriate or not. Some methods that are clearly not appropriate include: • Consumer surveys, • A compendium of experience, • Time series statistical analysis, and • Stated preference analysis. Descriptions of these methods and their uses are provided for completeness. Methods that are stronger candidates for this research include: • System-level demand modeling, and • Disaggregate travel demand modeling. These methods are described at greater length, with a sketch of how they could be applied to this research. Consumer Surveys Surveys of people with disabilities have been invaluable in understanding their needs and preferences. In the 1970s surveys of this type helped define the need for accessible public trans-

92 Improving ADA Complementary Paratransit Demand Estimation portation, including paratransit. The most ambitious of these was the National Survey of Transportation Handicapped People, conducted by Grey Advertising under contract to the Urban Mass Transportation Administration (Grey Advertising, 1978). Many transit agencies conducted similar surveys after passage of the ADA to help them plan how to come into com- pliance with the complementary paratransit requirements of the law. These surveys commonly obtained information about respondents’ disabilities, their travel, specific barriers to use of existing transportation options, and likely use of new options. Examples of such surveys include one conducted by members of the research team for King County Metro in Seattle in 1995 and one conducted by the Denver Regional Transit District in 1993. In combination with other types of analysis, these surveys produced useful information for planning a service that was very different from existing services. Surveys of people with disabilities are expensive to conduct, since it is typically necessary to call multiple households before locating qualified respondents. Further it is difficult to locate and survey people with disabilities living in group settings, and a significant minority of people with disabilities are not able to speak for themselves (for example older people with dementia and some people with developmental disabilities). Accommodation needs to be made for people with disabilities who cannot use a voice telephone. A further difficulty is that a brief series of ques- tions in a survey format cannot reliably determine whether respondents would be judged eligi- ble for ADA paratransit. Beyond all these practical concerns, experience has shown that con- sumers’ predictions about their travel (or other behavior) are not very accurate. Even if all these difficulties could be overcome, consumer surveys would still have limited suitability for this research. Consumer statements about travel obtained in a survey conducted in one metropolitan area would not necessarily apply in a different area. A national survey would provide interesting data for policy development, but would not be useful for local planning. These statements apply to surveys that directly ask consumers about their preferences and likely travel decisions. As will be discussed at more length in a later section, consumer surveys may in fact be appropriate as part of developing a travel demand model. However, the surveys used for travel demand modeling obtain data about actual travel behavior, not planned or intended travel behavior. In addition, more structured surveys can be useful in stated preference analysis as described more below. Compendium of Experience Where rigorous modeling is not practical, practitioners commonly rely on the experience of other systems, applying their professional judgment to determining which other systems are most comparable and to determining how to compensate for different situations. TCRP is currently engaged in a long-term project that has collected this type of information for many dif- ferent public transportation modes and issues. The results of this project, called “Traveler Response to Transportation System Changes,” is being published as individual chapters of TCRP Report 95. (Two previous editions were published in 1977 and 1981.) Chapter 6, dealing with “Demand Responsive/ADA” service, was published in interim form in March 2000 and in final form in May 2004. The usefulness of the results for ADA paratransit planning is limited by reliance on previously published material (most of it completed before many paratransit systems were in compliance with ADA requirements and much of it completed before the passage of the ADA) and by the fact that most of the analysis treats services other than ADA paratransit. However, it is entirely possible that an effort focused specifically on the exemplary systems identified for this research could produce more useful data. For example, the exemplary systems have probably experi-

Excerpts from the First Interim Report (May 2005) 93 mented with most of the policy options of interest (fares, door-to-door and curb-to-curb service, advance reservation policies, etc.) and many no doubt have experience with changes in eligibil- ity processes and planned or unplanned changes in service reliability. This information can be useful and instructive, as in the case studies often presented in research reports. However, a sim- ple presentation of experiences can be very misleading, since it does not provide any way to account for differences among service areas or to separate out the influence of multiple factors. It also does not provide an organized way to predict future demand based on population growth. For this reason, the experiences of individual systems will not be the central focus of this research, although they can be presented as supplementary material that will enrich the structured tools that will be developed. Time Series Econometric Analysis Time-series econometric analysis has been the most-commonly used type of analysis for para- transit demand in recent years. Several examples of time series models developed by the research team and others were cited in the literature review. These analyses have used data about actual demand on a daily, monthly, quarterly, or yearly basis along with data about fares, population, service reliability, etc. The statistical method applied to the data with varying degrees of sophis- tication is so-called “ordinary least squares regression.” The result is an equation that matches past experience and allows predictions about the near future. For example a time series analysis of monthly data for Access Services Inc. in Los Angeles (HLB Decision Economics, 2004) pro- duced the following model: log (Trip Requests) = −70.3 −0.43 log(Average Real Fare) +5.11 log(Population) −0.04 (Winter) +0.07 (October) −0.03 (PDL of Complaint Rate). Where: Trip Requests are trips requested by customers in any month of the analysis period. “log” represents the natural logarithm. Average Real Fare is the paratransit fare in a month adjusted for inflation. Winter is 1 in December, January, and February and 0 otherwise. October is 1 in October and 0 otherwise. PDL of Complaint Rate is a polynomial distributed lag of the natural log of complaint rate in the region.1 The equation was estimated using monthly data over a 42-month period. Applying the actual values of fare, population, and complaint rate during the analysis period, the equation produces estimated trip request that closely match actual trip requests. By applying projected future val- ues, the equation provides projections of future trip requests. The model has an “R-squared” of 1 Polynomial distributed lags (PDL) are used to reduce the effects of collinearity in distributed lag settings by imposing a particular shape on the lag coefficients. The specification of a polynomial distributed lag has three elements: the length of the lag (the number of time periods it covers), the degree of the polynomial (the highest power in the polynomial), and the constraints on the lag coefficients. A near end constraint says that the imme- diate effect of x on y is zero, whereas a far end constraint says that the effect of x on y dies off at the end. It is also possible to impose both constraints or no constraint at all.

94 Improving ADA Complementary Paratransit Demand Estimation 0.97, meaning that, in a statistical sense, it explains 97% of the observed variation of trip requests over the 42-month analysis period. All of the coefficients are statistically significant with 99% or better confidence. Other factors were also tested, including denial rate, on-time performance, and employment. These probably influence trip demand, but the strength of their effects was not statistically dis- cernible using the data available. Models like can be very useful for short-term planning. For example, in the Los Angeles model, future projected values of population, fare, and complaint rate were inserted in the model to produce estimates of future demand, as represented by trip requests. Models with a so-called log-log form like this one produce coefficients that can be interpreted as elasticities. In this example, computed real fare elasticity is -0.43: for every 1 percent increase in real fare, trip requests decline by 0.43 percent. The great limitation of time series models is that they cannot make predictions about any change that the paratransit system has not actually experienced in the past. For example, a time series model cannot predict the impact of changing advance reservation rules if the paratransit system has not experimented with a similar change before. Also, predictions that go significantly beyond the range of recent experience are unreliable. For example, a prediction about the impact of doubling a fare will not be very accurate if the paratransit system has only made very small fare changes in the past. For similar reasons, early attempts to predict what would happen when capacity denials were eliminated were not very reliable. A further limitation of time series analysis is that a model developed using data from one sys- tem may not be valid for a different system. For example, the estimated response to a fare change (expressed by the fare elasticity) may be quite different in two systems depending on differences in service reliability, differences in the availability of other services, and differences in economic conditions. By way of illustration, the estimated fare elasticity for Los Angeles is at the high-impact end of the spectrum defined by experience in other paratransit systems. This higher elasticity may reflect the easier access to alternate modes of transportation in Los Angeles County, since many local jurisdictions provide paratransit in addition to the ADA paratransit operated by Access Ser- vices. Also, the short-term elasticities determined from time series analysis may understate the impact of changes over the long term. The most likely use of time series analysis for this research will be in combination with cross- sectional analysis. In compiling data from the exemplary systems, it may be possible to obtain data for multiple years from some systems. If there have been significant differences from year to year, this data could enrich a cross-sectional analysis as described later. Stated Preference Analysis Stated preference analysis is a consumer survey-based method that has been developed to test consumer reactions to new choices in a more rigorous fashion than is possible with simple con- sumer surveys. At a time when there were no paratransit systems in compliance with ADA requirements, data about existing services mostly reflected the influence of service limitations. Stated preference would have offered a more sophisticated alternative to the consumer surveys described previously. The stated preference method relies on “contingent valuation” surveys. The contingent valuation survey is a measurement instrument in which statistically drawn respondents make trade-offs among experimentally designed choice situations designed to sim- ulate real-world conditions that might not presently exist. In principle, stated preference can also illuminate individual travel behavior in ways that system-level data cannot. For example, a stated

Excerpts from the First Interim Report (May 2005) 95 preference survey can show the value that disabled consumers place on various components of travel by paratransit and other modes and can show how these values differ depending on spe- cific types of disabilities. For purposes of this research, the stated preference method has significant drawbacks. Obtain- ing the necessary data can be quite expensive. The expertise needed to apply the stated prefer- ence method is not widespread, so it would be difficult for many transit and planning agencies to conduct their own analyses of new service alternatives as they become of interest. More fun- damentally, even though the method is far more sophisticated than simple survey analysis, it still relies on consumers’ statements about hypothetical responses to hypothetical situations. System-level Demand Modeling A system-level demand model would allow individual paratransit systems to obtain predictions of total ADA paratransit demand (and ideally, total people certified as ADA paratransit eligible) depending on future values of key factors such as population, rates of disability, income, ADA para- transit service policies and service reliability, availability of accessible fixed-route transit, and avail- ability of other specialized transportation services. To illustrate how this would work, Figure 10 provides a diagram of the influences that would ideally be included in such a model. The right-hand side of the diagram shows the many factors that influence the demand for paratransit, and the left hand shows the impacts of these influences, separated into stages. The top layer shows the factors that influence the size of population (i.e., number of individuals) in a service area that is theoretically eligible for ADA paratransit, regardless of whether these indi- viduals have actually applied and been certified as ADA paratransit eligible. The diagram then proceeds in stages showing the factors that influence: the percent of theoretically eligible people who actually apply for ADA paratransit and are certified as eligible; the percent of these people who actually use the service; the number of trips that these users reserve; and the number of reserved trips actually taken. The specific influencing factors listed in the diagram are provided as a starting point for discussion and analysis. There may be other factors that can be included, and some factors could be eliminated based on analysis results. The right-to-left arrows connecting the influencing factors with the demand outcomes repre- sent the strength and direction of each factor. In mathematical terms, these would be coefficients on equations that need to be estimated using statistical analysis of data from the exemplary sys- tems where possible. Where the data from the exemplary systems do not allow coefficients to be estimated (for example how community awareness affects the percentage of theoretically eligible people who apply for certification), expert opinion could supply default values or users could insert values based on their local knowledge and judgment. Similarly, expert opinion or local knowledge would be needed to supply input values for many of influencing factors such as the age distribution of the population at a future date of interest. Estimation Method: The principal statistical method to be used to estimate equations for the influences would be cross-sectional econometric analysis of data from exemplary systems. In a cross-sectional econometric analysis, data from numerous systems are gathered, usually for a single point in time (commonly the most recent fiscal year) and analyzed, usually with ordinary least-squares regression and/or analysis of variance.2 For the sake of illustrating the concept, Figure 11 shows an example of a very crude analysis of this type, just using actual paratransit demand and service area population. Points of the graph show population and paratransit 2 Analysis of variance is equivalent to regression with dummy variables and is relevant where variables are expressed in categories (yes/no) or discrete levels (low/medium/high) instead of continuous values.

96 Improving ADA Complementary Paratransit Demand Estimation Figure 10. Structure and logic diagram of a system-level demand model.

Excerpts from the First Interim Report (May 2005) 97 demand as of fiscal year 2000-01 in most cases. The line is a best-fit line determined using least- squares regression. The equation of the line is: Trips per year = 254,255 + 0.23 (Service Area Population) This equation, in effect, collapses most of the diagram in Figure 10, going straight from the top to the bottom of the diagram in one step with only one influencing factor. The challenge for this research would be to fill in as much additional detail as possible using the available data. At a minimum, the researchers are confident that an equation or equations can be estimated connecting many of the most important influencing factors with total paratransit demand. This would be a more refined version of the equation given before. For example, total paratransit trips may be a function of the population with disabilities in a service area, some measure of the rigor of the eligibility process, a measure of service reliability, a measure of the availability of non-ADA specialized transportation, and a measure of the availability of accessible fixed-route transit service. A single equation of this type, while more refined than the one given, would still collapse the diagram to a single layer, leaving out the intermediate stages concerning the percent of theoret- ically eligible people who become certified, the percent of those who use the service, etc. To fill in these intermediate stages would involve estimating multiple equations, one for each stage. Estimating these multiple equations is theoretically possible, but may prove impractical for three reasons: 1. Unknown variables: Estimating an equation to represent the first stage of the diagram would require knowing the size of the population theoretically eligible for ADA paratransit in the service areas of the exemplary systems. In practice this number is not known. Even the num- ber of certified individuals (needed for the second stage of the diagram) may be unknown in some cases. This uncertainty arises from the fact that some systems have poorly maintained lists of certified individuals, including many people who are no longer living or who may have moved out the service area. 2. Processes that are not independent: The processes represented by stages two, three, and four of the diagram (certification, becoming an active user, and reservation rate) are not independent. As shown in the diagram, the same factors influence becoming an active user and reservation rate, and these factors also probably play a role in influencing the certifi- cation rate. For example, service reliability certainly influences the rate at which users reserve trips, and most likely influences whether people who have become certified use the service at all. But it probably also influences the rate of applications for ADA paratransit eligibility. If the perception in the community is that service is very unreliable, many poten- tially eligible people will probably not even bother applying for certification. This would be more true for those people who have other transportation available (for example, in the 0 200,000 400,000 600,000 800,000 1,000,000 1,200,000 1,400,000 0 1,000,000 2,000,000 3,000,000 4,000,000 Service Area Population Tr ip s pe r Ye a r Figure 11. ADA paratransit demand and population.

98 Improving ADA Complementary Paratransit Demand Estimation form of rides from a family member) illustrating how this factor, too, influences multiple stages of the process. 3. Small sample size: At least the problem of lack of independence might be overcome with sufficiently large samples. However, it appears that on the order of 20 exemplary systems will be available for analysis. With a sample this size, it is unlikely that multiple equations can be estimated. The form of the equations to be estimated will be determined according to what provides the best fit to the data. In principle, a multiplicative form has desirable properties. This type of model would have an equation or equations of the following type: Demand = a × (Factor 1)b × (Factor 2)c × (Factor 3)d . . . and so forth. The superscripts represent exponents and are measures of the strength of each fac- tor that are equivalent to elasticities. These exponents and the constant “a” will be estimated using the data from the exemplary sys- tems. If the estimated exponent for a factor is close to zero, then, since (Factor)0 = 1 always, val- ues of that factor have no impact on demand. Large positive exponents mean that increases in a factor strongly influence demand to increase. Large negative exponents mean that increases in a factor strongly influence demand to decrease. “Demand” may be expressed in the form of trips per capita or total trips. Trips per capita has the desirable property that it “automatically” adjusts for the impact of population, providing a measure that is more comparable across service areas than total trips. This type of equation is typically transformed using logarithms to the equivalent form: log(Demand) = a + b log(Factor 1) + c log(Factor 2) + d log(Factor 3) . . . Here the exponents are seen as coefficients that can be estimated using linear regression. Other model forms will also be tested including simple additive and mixed forms. Uses of the Model: Regardless of the amount of detail that would prove practical to include in the model, the result could be used in two ways. First, by assuming future values of popula- tion, fares, transit accessibility, etc., a transit system could obtain an estimate of future paratransit demand. Second, systems that are not currently in full compliance with ADA could use the model to determine hypothetical demand under today’s conditions assuming full compliance. For example a system that currently denies trips but is otherwise in compliance would simply input values for current population and service variables. Since the model would be estimated using only exemplary systems, the output of the model would be demand under conditions of zero or near-zero denials. For use by practitioners, the model would be implemented as a spreadsheet or self-contained program that allows users to input values for local conditions and obtain estimates of demand. For inputs that users do not have available, values from the survey of exemplary systems would be provided that users can substitute based on their judgments about comparability. The model could also provide estimates of upper and lower probability bounds for estimated demand based on assumed probability bounds of the input variables as well as estimated statistical error in the values of the coefficients used in the model. Relationship to Selection Criteria: The system-level demand model meets most of the crite- ria listed earlier. The researchers are confident that it can be accomplished within the available resources with available data. It explicitly includes variables describing local conditions, so it would be transferable among cities. For the most part, systems should be able to obtain the data needed to use the model, although predictions of future population and other conditions are always uncertain. The model can be provided in a form that does not require great technical

Excerpts from the First Interim Report (May 2005) 99 sophistication. The model would be estimated using data from exemplary systems; as a result, it could be limited to ADA trips by ADA eligible individuals as long as the exemplary systems that also operate non-ADA service are able to provide data separating out those trips. The model would address many of the policy issues of interest to the extent that it proves possible to obtain good data. Compared with time-series models that project from current conditions, a model based on cross-sectional analysis should have good ability to produce estimates for periods five or more years in the future. This ability stems from the wide range of conditions represented by the exemplary systems used to estimate the model. A system-level model would provide little detail related to determining the cost of service pro- vision, for example, trip lengths, time-of-day peaking, or the portion of trips by people who use wheelchairs. A system-level model would also provide limited insight into fundamental issues of travel behavior by individuals with disabilities. However, the process of creating the model would at a minimum provide extensive input for an agenda to guide future research. The model results would have potential for implementation within a conventional regional travel demand model. Most regional models have zone-specific base year and forecast year val- ues for the non-paratransit variables in the model. By default, all zones served by a particular paratransit system would have the same values for variables describing the paratransit level of service. The model would be implemented as a demand equation (or set of equations) that can be scripted inside of the typical MPO software environment (TransCAD, TP+, Cube, etc.), pro- viding access to all zonal and network input variables and writing out zone-specific forecasts that can be easily viewed in GIS/network format to see where the ADA trips will most likely originate from. The user would need to be given the caveat that this model is likely to be less geographi- cally accurate than typical trip generation models, so any zone-specific output should be used only as an indication. The total regional demand forecast would be more accurate, as would the totals for major sub-regional areas such as counties or cities. Disaggregate Travel Demand Modeling The kind of modeling described in the previous section is “aggregate” in the sense that it groups together data for all of the individuals in a particular service area. Conventional travel demand models used for regional planning use aggregate methods. That is, they treat all trips origins and destinations within a given zone as if they were located at one point in space. They represent all households in a zone using average values or at most a small number of averages for ranges of income, household size, and car ownership. And they represent all trips within two or three periods of the day as subject to the same conditions of traffic congestion (Vovsha et al., 2004). In contrast, a newer generation of travel demand models use disaggregate methods, mean- ing they use data about individual people and individual trips. For purposes of this research, the differences can be described as follows: Aggregate: Data are counts or averages across geographic areas, such as cities, counties, zones, zip codes, etc. This includes both the dependent variables (e.g., ADA paratransit certification rates and trip rates) and the explanatory variables (e.g., population by age and gender, income distribution or average income, system-level paratransit service descriptors). Disaggregate: Data are collected from individual persons. The data items typically represent the same variables as in aggregate data, but are measured at the individual level. Again, this includes both the dependent variable (e.g., whether a person has certified as ADA paratransit eligible, or how many paratransit trips a person makes during a given day or has made in the most recent week), and the explanatory variables (the person’s age, gender, income, household size, distance from shopping, etc.).

100 Improving ADA Complementary Paratransit Demand Estimation In considering the possibility of applying disaggregate methods, it becomes even clearer than before that, in a fundamental sense, travel choices by people with disabilities are driven by the same factors that drive travel choice by non-disabled people. For example, the literature review at the beginning of this report identified several key principles of travel demand theory, including: • Travel time and discomfort are disutilities that consumers seek to minimize in choosing how, whether, and when to travel. • Travel demand has been shown to stem from demand for activities away from home, the con- sumption of which is strongly linked to disposable income. Gender is tied to trip frequency, with females engaging in more trips. • Car ownership, at least partly a function of income, is strongly linked to mode choice. Bus travel has been found to be an “inferior good” (demand declines with increasing income). In addition it is known that travel decisions are influenced not just by individual characteris- tics, but by household characteristics, including trips by other household members and avail- ability of rides with other household members. In considering people with disabilities and paratransit, we can hypothesize the following: • Disabling conditions make any trip more time consuming for people with disabilities than for people without disabilities; thus people with disabilities travel less than people without dis- abilities. This effect may be reduced but probably not eliminated by modifications to build- ings, sidewalks, and transportation vehicles. • The mode choices of people with disabilities are constrained by functional inability or greater difficulty in driving or using public transportation, even assuming availability of adapted vehi- cles and fully accessible public transportation. • People with disabilities of working age have higher unemployment rates and lower incomes than people without disabilities; thus people with disabilities travel less than people without disabil- ities. Eliminating employment discrimination will reduce but probably not eliminate this effect. • A high proportion of people with disabilities are older, thus they travel less than people with- out disabilities. • Because many people with disabilities are older, and possibly for other reasons, they are more likely to live alone than people with disabilities. Thus they have less ability to rely on others for rides or to perform activities that substitute for travel. The exact workings of these connections remain to be determined. However, given an under- standing of each of them, the travel behavior of people with disabilities could, for the most part, be understood by treating the identical explanatory factors used in modeling travel behavior of non- disabled people. An important caveat in this respect concerning ADA paratransit is that the eligi- bility process determines the availability of this mode for each person. Also, in the case of people with disabilities, modes that impact travel choices include some that are not considered in the analysis of travel by the general public, including specialized services operated by agencies that serve people with developmental disabilities, adult day health centers, senior centers, and Medicaid. Figure 12 provides a preliminary sketch of how demand for ADA paratransit might be viewed at the level of choices by individual people. The rectangular boxes at the left of the diagram represent stages similar to the stages represented at the aggregate level by Figure 10. Some of these stages are individual choices (whether to apply for eligibility, whether to travel by paratransit), while others are stages leading to individual choices (awareness, extent of functional limitation). The boxes with rounded corners to the right of these rectangular boxes represent the various fac- tors that influence these stages. As in the earlier figure, many factors influence multiple stages. As in the earlier figure, observa- tions representing many of the intermediate stages are not available. For these reasons, a practical

Figure 12. Influence diagram for individual paratransit travel.

102 Improving ADA Complementary Paratransit Demand Estimation disaggregate model would probably be much simpler than suggested by the diagram, representing the impact on paratransit demand of personal, household, and mode characteristics in one or two steps. Where the aggregate model uses regression analysis, with disaggregate data, there are more choices, including discrete choice models such as multinomial logit that can accommodate yes/no type choices or choices involving more than two alternatives, e.g., the mode used for a given trip. The immediate outputs of the model equations are typically probabilities that are translated into trip totals by various means. Considerations for this research in deciding whether to pursue a disaggregate approach include the following: Transferability: Models based on disaggregate data from one or a few regions are typically assumed to be more transferable to other regions because they measure behavior at a more fundamental level, rather than simply picking up aggregate correlations in the data that may not hold for other aggregate samples or at other points in time. As a corollary to this point, we typically learn more about behavior when analyzing disaggregate data compared with aggre- gate data. Detailed Data: Disaggregate data sets typically cover a much wider range of data items than can be collected from sources of aggregate data. Estimating a model requires a person-based sur- vey asking about trips made by modes other than ADA paratransit, including private modes, whereas aggregate data collected from the paratransit operators and other transit operators would not provide such information. Expense of Data Collection: Disaggregate data collection may be more expensive in general. This depends on the study context and how difficult it is to reach the target population. For this research, it may be possible, working with transit operators, to obtain data for people who have already been certified as ADA paratransit eligible. However, it would also be important to reach those who are disabled but do not use ADA services. A random telephone survey would be a very inefficient way to find such a limited population. If suitable disaggregate surveys have already been done in the past and the data already exists, then this option becomes more attractive. If future regional household travel surveys include the necessary questions, they could be used for this purpose. Appearance and Use of the Model: In terms of what the final models “look like” and how difficult they are to apply, both types of data and both types of modeling methods end up pro- ducing models that look very similar in terms of the variables they include and the type of infor- mation needed to apply them. As mentioned above, disaggregate data can be used to include more variables in the models, but the “final” model need only include the variables for which information is available to apply it. If the model is applied to zonal or regional aggregate data, it will be applied in essentially the same manner regardless if it was estimated on aggregate or dis- aggregate data. However, aggregation bias will also occur to some extent if a model is estimated from disaggregate data but then applied to aggregate zonal or regional data. This is one of the main reasons for the growing popularity of micro-simulation methods in regional travel demand modeling to simulate individual decisions and then aggregate those simulated choices. There are ways to use aggregate population statistics to generate representative “synthetic” populations of individuals, though it may be more challenging to synthesize a representative population of dis- abled individuals. Level of Expertise: Discrete choice modeling methods require more expertise than more com- mon regression methods, although most packages such as SPSS and SAS now include routines for binary and multinomial logit model estimation. Also, the fact that there is usually a wider

Excerpts from the First Interim Report (May 2005) 103 range of explanatory variables available in disaggregate data is a good feature, but it can require more time and judgment on the part of the analyst to decide on the best model specification. Looking forward to eventually incorporating paratransit into regional travel demand models, the disaggregate approach offers important advantages. With aggregate data, if the segment of the population being studied is a small percent of the general population, then aggregate statis- tics for any given geographic area will not be very accurate for that particular segment. With dis- aggregate data collection focused on that specific segment of the population, this problem does not occur (although it does mean they may be more expensive to contact). Referring again to the criteria for choosing tools to develop in this research, a disaggregate model would have good ability to address policy issues of interest. More than a system-level model, it would contribute to increased understanding of travel behavior and needs of people with disabilities and has greater potential for incorporation in the next generation of regional travel demand models. Like the system-level model, a disaggregate model could be limited to ADA paratransit eligible trips and individuals. A major drawback of disaggregate modeling is that it requires data beyond what can be obtained in this project. A survey of people with disabilities that would obtain the necessary data would be a major undertaking even in one region. To produce a model with a reasonable degree of transferability, it would be necessary to conduct similar surveys in several regions with exem- plary systems. Even then, the credibility of the results would be less than a model based on data from the larger sample of exemplary systems planned for the system-level model. Applying the model in another region would require similarly detailed data about people with disabilities in that region and a degree of expertise not generally available within a transit agency. Appendix C Bibliography Batchelder, J.H.; Forstall, K.W.; Wensley, J.A., Estimating Patronage For Community Transit Services, Multisys- tems for FHWA and UMTA, 1884. Bearse, Peter, Shiferaw Gurmu, Carol Rapaport, Steven Stern, “Estimating Disabled People’s Demand for Spe- cialized Transportation,” Transportation Research, Part B, Volume 38, Issue 9, November 2004, pp. 809-831. Benjamin, J.M.; Sakano, R., “Equilibrium Model Based on a Microfoundation for Forecasting Dial-A-Ride Rid- ership,” Transportation Research Record, 2001. Bureau of Transportation Statistics, Freedom to Travel, BTS03-08, Washington DC, 2003. Crain & Associates, Inc and HLB, ADA and Options Paratransit Demand Estimation Study: Final Report (for King County Metro), 1995 (basis of the TRB paper by Koffman and Lewis). Crain & Associates, San Francisco Bay Area Regional Paratransit Plan, Working Paper No. 6, Service Needs Analy- sis, for Metropolitan Transportation Commission, Oakland CA, January 1990. Domencich, Tom and Daniel L. McFadden, Urban Travel Demand: A Behavioral Analysis, North-Holland Pub- lishing Co., 1975. Reprinted 1996. Available online at http://emlab.berkeley.edu/users/mcfadden/ travel.html. Fitzgerald, James, Donna Shaunesey, Steven Stern, “The Effect of Education Programs on Paratransit Demand of People with Disabilities.” Transportation Research Part A, Vol. 34, 2000, pp. 261-285. Franklin, Joel T. and Debbie A. Niemeier, “Discrete Choice Elasticities for Elderly and Disabled Travelers Between Fixed-Route Transit and Paratransit,” Transportation Research Record 1623, Transportation Research Board, National Research Council, Washington DC, 1998. Glaister, Stephen, Fundamentals of Transport Economics, Basil Blackwell, London, 1981. Grey Advertising, Incorporated, Technical Report of the National Survey of Transportation Handicapped People, Urban Mass Transportation Administration, Washington DC. Report No. UMTA-NY-060054-78-1, 1978. Hickling Corporation, “Demand for Paratransit: A Cross Sectional Analysis” (Chapter II of Regulatory Impact Analysis, Americans with Disabilities Act, U.S. Department of Transportation). 1991. HLB Decision Economics, Demand Forecasting Model for Dayton Metropolitan Area ADA Paratransit, Dayton Metropolitan Area Transit Authority, Washington DC, 2002. HLB Decision Economics, Demand Forecasting Model for LA Access ADA Paratransit, Los Angeles Metropolitan Area Transit Authority, Washington DC, 2003, 2004.

104 Improving ADA Complementary Paratransit Demand Estimation HLB Decision Economics, Demand Forecasting Model for New York City and Five Borough ADA Paratransit, Met- ropolitan Transit Authority of New York, Washington DC, 1998, 1999, 2000, 2001, 2002. HLB Decision Economics, Demand Forecasting Model for OC Transpo Paratransit, Ottawa-Carleton Transit Authority, 2003. HLB Decision Economics, Demand Forecasting Model for SEPTA ADA Paratransit, Southeast Pennsylvania Tran- sit Authority, Washington DC, 2002, 2003, 2004. HLB Decision Economics, Demand Forecasting Model for WMATA ADA Paratransit, Washington Metropolitan Transit Authority, Washington DC, 2001, 2002, 2003, 2004. Koffman, David and David Lewis, “Forecasting Demand for Paratransit Required by the Americans with Dis- abilities Act,” Transportation Research Record 1571, Transportation Research Board, National Research Coun- cil, Washington, DC, 1997. Levine, J.C., ADA And The Demand For Paratransit, Transportation Quarterly, Volume 51, Issue: 1, 1997. Lewis, David, “Making Paratransit Decisions when Budgets are Constrained,” Journal of Specialized Transporta- tion Planning and Practice, Gordon and Breach, April 1992. Lewis, David, Economics of Serving the Travel Needs of People with Disabilities, London School of Economics (unpublished Ph.D. dissertation), 1983. Lewis, David, Todd Evans and David Koffman, Impact of Reliability on Paratransit Demand and Operating Costs (in Seattle and New York City), Transportation Planning and Technology, Gordon and Breach, Vol. 21, 1998. Lewis David, Transportation for Disabled Persons: Issues and Options, Congressional Budget Office, 1979. McFadden Daniel, The Theory of Modal Choice (as presented in) Stephen Glaister, Transport Economics, Cam- bridge University Press, 1985. Menninger-Mayeda, Heather, Peggy M. Berger, Dale E. Berger, Ph.D., Beth McCormick, Daniel K. Boyle, “Demand Forecasting and the Americans with Disabilities Act: Orange County, California, Transportation Authority’s Access Program,” Transportation Research Record 1884, Transportation Research Board, National Academies, Washington, DC, 2004. Multisystems and Crain & Associates, Transit Cooperative Research Program Web Document No. 2: Evaluating Transit Operations for People with Disabilities, Transportation Research Board, National Research Council, Washington, DC, 1997. Noland, Robert, et al., Estimating Trip Generation of Elderly and Disabled People: An Analysis of London Data, Imperial College, Centre for Transport Studies, July 2004. Quandt, Richard, The Theory of Abstract Modes, Review of Economics and Statistics, 1963. Schmoecker, Jan-Dirk, Fumitaka Kurauchi, Michael GH Bell, John Polak, “A d2d Travel Budget Scheme for Lon- don.” Center for Transport Studies, Imperial College London, 11 August 2002 (Draft). SG Assocs. et al., TCRP Report 3: Workbook for Estimating Demand for Rural Passenger Transportation, Trans- portation Research Board, National Research Council, Washington, DC, 1995. Starks, JK, “Overview of the Transportation Demand of Mentally Retarded Persons,” Transportation Research Record 1098, Transportation Research Board, National Research Council, Washington, DC, 1986. Stern, S, “A Disaggregate Discrete Choice Model of Transportation Demand by Elderly and Disabled People in Rural Virginia,” Transportation Research. Part A, Volume 27, Issue 4, 1993. White, C; Wilks, S; Levine, J; Purves, W; Torng, G-W, Paratransit Demand Management Evaluation Handbook, Ann Arbor Transportation Authority, 1995.