Improving ADA Paratransit Demand Estimation: Regional Modeling (2012)

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.

10 C h a p t e r 2 This chapter explains the rationale for developing disaggregate models of ADA paratransit travel and how the typical travel demand model structure was adapted for the case of ADA para- transit. There is also a review of the steps carried out by the research team, including selecting sites for collecting data and the actual data collection process. Regional Travel Demand Models This project created a model of the type used by MPOs in the regional household travel mod- els with which MPOs project travel trends, including use of major highways and transit. The new model (actually a series of models) can be combined with existing regional travel models to add ADA paratransit to the mix of modes treated by those regional models. Regional travel models traditionally have been based on aggregate data for travel analy- sis zones (TAZs). More recently, disaggregate travel demand models have been developed that model choices by individuals at the behavioral level that they actually occur. A typical model of household travel demand treats travel behavior as a series of separate but interrelated decisions: • Frequency choice (usually called “trip generation”): the choice of how many trips to make for different purposes; • Destination choice (usually called “trip distribution”): the choice of where to travel to; • Mode choice: the choice of which travel mode (drive-alone, transit, carpool) to use; and • Route choice (usually called “trip assignment”): the choice of which route of travel to take. A disaggregate travel demand model attempts to explain these individual decisions in terms of individual or household characteristics (e.g., income, gender, and employment status), the available opportunities (e.g., work and shopping) at various possible destinations, and the cost or travel time associated with possible trips depending on the mode of travel. This research created a disaggregate model of the decisions of individual ADA paratransit users to make particular trips and of their decisions to make them by ADA paratransit or by some other mode. These choices are similar to those modeled in typical regional travel demand models, although ADA paratransit users have a different set of modes available to them. Also, choice of route of travel, which is part of a typical travel demand model, is not relevant to travel by ADA paratransit, because the system operator determines the route after the individual has chosen to make a trip. Description of the Research

Description of the research 11 The alternative to a disaggregate model is an aggregate model, which is one that treats only the combined results of travel decisions by thousands of people (e.g., total trips in an area). The model developed in the first phase of this research was an aggregate model that compared total ADA paratransit demand in 28 transit systems and attempted to explain the observed differences in terms of regional measures such as population and poverty level, as well as ADA paratransit system policies such as fares and on-time performance. There are two primary advantages of the disaggregate approach, relative to an aggregate model. First, the disaggregate approach may avoid the problem of spurious results. The more aggregated the data is, the more likely one is to find broad correlations between variables and the more difficult it is to attribute behavioral effects to any particular variable. For example, the first phase of this research found that high levels of poverty in a region correlate with lower ADA paratransit ridership. This effect was statistically very significant but it says nothing about how the incomes of ADA paratransit riders (as opposed to the community as a whole) affect their use of ADA paratransit. It could still be the case that lower income riders choose ADA paratransit instead of private automobile more than higher income riders. With disaggregate data, we can relate the ADA paratransit trip rates of individual persons or households to (1) their household incomes as well as the availability of an automobile within the household; (2) the accessibility to important destinations by automobile versus other modes (e.g., parking convenience, parking costs, and walking distance between stores); and (3) land use mixes (the proximity of different types of destinations). By using a large number of observed cases subject to different levels of these variables, we can overcome problems of correlation and sort out their relative effects on behavior. A second important advantage of the disaggregate approach is that it can overcome aggregation bias. This type of bias arises from the fact that most models that represent discrete choices at the individual level (e.g., logit models and gravity models) are nonlinear, and thus the probability share and model sensitivity at the aggregate average value is not necessarily equal to the average of the probabilities and sensitivity across all individual values. This is shown in Figure 2-1 and is true both for the predicted choice shares and the predicted elasticities. This means that if the data used to estimate and/or apply demand models are aggregated to too coarse a level, the predicted demand is subject to inaccuracies. As an example of aggregation bias, suppose that households with no automobiles have few alternatives to using transit, so their mode choice is not very sensitive to transit service levels. Also suppose that households with a car for every driver are very unlikely to use transit, so their mode choice is also insensitive to transit service levels. The intermediate households who own cars but do not have a car for every driver are the ones where transit and automobile are most competitive, and thus most sensitive to transit service changes. A model that uses aggregate average car ownership levels within a zone or a region would assign everyone an intermediate level of car ownership and thus would over-predict the sensitivity of mode choice to transit service levels. Similar logic could apply to predicting how ADA paratransit service quality affects the choice to use ADA paratransit instead of another mode. Aggregate regression models such as the one created in the first phase of this research (TCRP Report 119) can be subject to this same underlying behavioral inaccuracy. Such models are estimated using single average values for variables distributed across the population, and there is no guarantee that the predicted effects of changing those variables will be the same as what we would predict from more detailed models that segment the population into more homogenous categories.

12 Improving aDa paratransit Demand estimation: regional Modeling Average probability is not equal to the probability at the average of explanatory variables. The average impact of a change (average of slopes at a and b) is not equal to the impact calculated at the average of the explanatory variables. Figure 2-1. Aggregation bias with nonlinear logit models. Adapting Travel Demand Models for ADA Paratransit For modeling ADA paratransit demand, the basic four-step approach has been expanded to account for issues unique to this mode of travel and to incorporate state-of-the-art methods in regional travel demand modeling. Aside from the details of implementation, the model still treats four key decisions that determine the number of trips by ADA paratransit that any given individual makes, but not the same ones used in a traditional four-step model and not in the usual sequence. The decisions are 1. The decision to apply/register for ADA service eligibility. This step is necessary, because a person with a disability has to apply for and be certified as meeting ADA eligibility criteria before ADA paratransit becomes a possible travel choice. 2. Tour generation: The decision to leave home to make one or more connected trips for some purposes.

Description of the research 13 3. Mode choice: The decision to make that series of trips by ADA paratransit or an alternative mode, or some combination of modes. 4. Destination choice: The decision to visit a specific destination. The ordering of the steps implies that each of these decisions is conditional on making the decision above it. However, we cannot treat the decisions as purely sequential, because each decision may also depend somewhat on the decisions below it as well. The decision to apply for ADA eligibility will depend on the number of trips a person makes and the propensity to make those trips by ADA paratransit. This is comparable to the decision to get a license to drive an automobile. For example, people in New York City are less likely to have driving licenses, not because they are less able to drive, but because driving is less attractive there, so they are less likely to make the effort. In the case of ADA paratransit, the decision to apply for ADA eligibil- ity (Step 1 above) depends to some extent on whether or not there are particular types of trips that a person wishes to make using the system (Steps 2 through 4 above). Another example is the fact that some people may have no feasible alternative to ADA paratransit for some trips, so the decision to make a trip at all may depend on the availability of ADA paratransit service. This inter-relationship is probably even stronger than it is for most other types of travelers who are able to use a wider variety of modes. In disaggregate travel demand modeling, the most effective way of modeling interrelated deci- sions is to use the expected utility, or “logsum” (the logarithm of the sum of the modeled utili- ties), across all available alternatives in the lower level model (i.e., a model of one of the lower decisions in the list above) as an explanatory variable in the upper model (i.e., a model of one of the upper decisions in the list above). This essentially leads to a system of simultaneous nested models which are internally consistent. This type of linkage is described further below, as part of a discussion of variables that should be considered in each of the four models above. The choice to model mode choice before destination choice departs from the usual practice in travel demand models in the United States, though it is not unusual in Europe. It is entirely possible that ADA paratransit users’ choice of destination depends more on the modes available than that mode depends on the destination. Also, in the case of ADA paratransit, the travel time by ADA paratransit to a specific destination is not generally known in advance. Therefore, it is not critical to model mode choice after destination choice. Moreover, NCTCOG’s model has very limited data on transit travel times. In the mode choice model, a logsum of the type just described is used to represent overall ability to reach destinations of interest by each mode. The next four sections describe modeling methods for each of the four decisions. Each com- ponent of the complete model system is described in detail in Chapter 3. The Decision to Apply/Register for ADA Paratransit Eligibility Initially, it had been hoped to estimate a disaggregate model of ADA paratransit registration that would predict the probability of any individual applying and registering for ADA para- transit. However, that would have required a very large survey of the general population, not just limited to ADA paratransit users, that obtained the variables related to ADA paratransit eligibil- ity as well as whether any household members were actually registered for ADA paratransit. Such survey data was not available and could not be obtained for a reasonable cost. In the future, it may be possible to obtain the data for such a model by adding a small number of questions to a general-purpose regional household travel survey. Respondents would need to be asked if any household members had any condition that limited their ability to travel, and then questions would need to be asked about the nature of the condition and whether the individual was registered for ADA paratransit. Ideally, the travel diary portion of the survey would include ADA paratransit among the modes that respondents could report on.

14 Improving aDa paratransit Demand estimation: regional Modeling For this research, a model was estimated to predict the fraction of people in each census tract who apply for and obtain ADA eligibility. One set of variables of interest are those related to disability: • Age • Gender • Employment status (unemployment as an indicator of disability) • Income (provides access to healthier lifestyle, better health care) • Household size (people not living alone may tend to be healthier) These are not necessarily causes of disability, and certainly have nothing to do with whether an individual meets the ADA eligibility criteria; but they may have a statistical relationship to ADA eligibility that will be useful for modeling purposes. Other variables are related to the probability that eligible individuals will actually apply for eligibility and be accepted: • Household size (as an indicator of the availability of help from other family members) • Location within the ADA paratransit service area • Proximity to locations that can be reached by walking or wheelchair (reducing the need for ADA paratransit) In a disaggregate model, it would also be desirable to include the increase in mobility and accessibility that ADA paratransit would provide for the individual. This could be measured by the difference in the overall expected utility from the trip generation, distribution, and mode choice models with and without ADA paratransit as an alternative. Other variables could affect application rates among different regions, but could not be mea- sured within one study region. These include • The process used by the provider in determining eligibility (e.g., whether a simple paper application is used, or all applicants are subject to functional testing) • The level of awareness of the service (the degree of activity/sophistication of social service agencies/advocacy groups in the region may be an indicator) Tour Generation: The Decision to Make a Series of Trips The key variables are those that influence the propensity of a person to carry out various types of out-of-home activities. Depending on the type of activity (the trip purpose), variables may include • Age. Most activities generally decrease with age. • Income. People with higher incomes typically travel more for all purposes. • Employment status/student status. These determine the need for commute or school trips. • Gender. Females and males often have somewhat different participation rates for specific out- of-home activity purposes, particularly in older households. • Type of impairment/disability: People with certain disabilities may be more likely to travel for certain purposes (e.g., people with mental impairments to adult daycare). • Household size. In general, people who live alone may make more trips because they cannot delegate activities to others, but people with disabilities who live with others may make more trips because they have someone to assist them, accompany them, or provide rides for them. • Car availability. • Accessibility to activities. People living in areas where it is possible to reach more activities of interest in a reasonable time are likely to travel more. (“Accessibility” here is used in the broad sense generally used in travel behavior theory, namely access to destinations, taking account of travel times, availability of travel options, etc.)

Description of the research 15 • Regional effects. Climate and lifestyles may vary somewhat across regions in ways that affect travel. However, in a model estimated for one region, there is no way to measure these effects. Trip generation was modeled at the level of tours (i.e., tour generation) consisting of a sequence of chained trips from home to various destinations and back to home. Within each tour, additional steps modeled the number of intermediate stops that would be made. Mode Choice: The Decision to Travel by ADA Paratransit or an Alternative Mode Depending on the principal purpose for leaving home, the key factors in mode choice fall into two categories: (1) variables directly related to ADA paratransit service and (2) variables related to alternatives to using ADA paratransit. In principle, ADA paratransit service variables should be of major importance. These could include • Service reliability • Advance reservation requirements (how far ahead a trip can be reserved) • Availability of most convenient requested time • Conditional eligibility trip screening • Fare • Travel time (will typically be somewhat longer than travel time by private car, but somewhat shorter than the scheduled time on fixed-route transit) • Denial rates (not relevant in this study, because the scope was limited to ADA-compliant services) • Availability of a program to coach riders on how to use the system In practice, with results from only two very similar ADA paratransit services, and with no data about ADA paratransit travel times, it was not possible to estimate these effects from travel data. As a next-best alternative a Stated Preference analysis was used to provide limited evidence for the effect of some ADA paratransit service variables. The process of gathering Stated Preference data is described in the Data Collection section and the analysis is described in Chapter 3. Variables related to alternatives to using ADA paratransit—that can be modeled—include • Automobile ownership and availability. Ability to drive would be important, but few ADA paratransit users are able to drive. • Household size (may mean that a companion driver is available). • Cars per driver in the household (for households with a car). • Income (determines which options are affordable and may also influence availability of spe- cialized services). • Age (influences ability to drive and walk and may influence availability of other specialized services). • Disability type/need for mobility aids (also influences the ability to use alternative modes and services). • Trip purpose has a bearing on mode choice and is easily included in the model. For example, recreation and social trips may be less commonly made by ADA paratransit than other trip types, and some work and school trips may tend to be made using specialized services oper- ated by school districts or by agencies that provide supported work. Ideally, a mode choice model would incorporate the effect of differences in the cost (to the rider) and time required for a trip by ADA paratransit compared to the same trip by other modes. However, the cost for every ADA paratransit trip is essentially the same in the Dallas-Fort Worth region, and travel time for an ADA paratransit trip is not known before a reservation is

16 Improving aDa paratransit Demand estimation: regional Modeling made. Instead, a summary measure of accessibility to all potential destinations by each mode was used; this is another example of the type of logsum described earlier. Mode choice is first modeled for tours. Then after intermediate stops are added to tours, a separate model treats whether subsidiary modes would be used (e.g., getting a ride home in a car after taking an ADA paratransit trip or else walking, going by wheelchair, or getting a ride to transit). Additional variables that may be important but could not be modeled include • Availability of specialized services provided by Medicaid, adult day healthcare, and programs for those who have developmental disabilities. This varies among regions, but is the same for all ADA paratransit users in the Dallas-Fort Worth region. However, some of this effect is captured by trip purpose as noted above. • Fixed-route transit convenience, fare, frequencies, transfers required, wheelchair lifts, etc. NCTCOG’s model included very limited transit data, and both DART and FWTA operate 100% wheelchair-accessible fleets. In particular, planners would find it useful to understand the potential impacts of • Establishing more convenient fixed-route transit service to major ADA paratransit trip gen- erators and attractors, • Improvements in accessible pathways to fixed-route stops and stations, • Offering fixed-route fare incentives for ADA-eligible riders or persons with disabilities in general, • Reducing walking distance for riders with disabilities by shifting to a flex-route design in certain corridors. Trip Distribution: The Decision to Visit a Specific Destination Destination choice models typically include two types of variables: (1) impedance variables and (2) attraction variables. Impedance variables (e.g., travel time and cost) measure the separa- tion between zones and typically have negative effects. Attraction variables (e.g., retail employ- ment) measure the attractive power of a zone as a destination and have positive values. Given that transit travel times were not available and ADA paratransit travel time is not known in advance, automobile travel time was used as an impedance measure. The 2007 zonal data provided by NCTCOG has six possible attraction variables: • The number of resident households • The number of retail jobs • The number of service jobs • The number of “basic” jobs (non-retail, non-service jobs, such as industry, production, and wholesale) • The number of jobs at shopping malls • The number of jobs at hospitals The number of jobs in a zone is considered a measure of the overall level of activity in the zone that should attract trips. For example, numerous jobs at hospitals in a zone indicate the presence of major medical facilities that would attract large numbers of medical trips. The first four variables are typical attraction variables used in travel models, although some regions have finer breakdowns of employment than just the three categories (basic, retail, service). The last two, however, are not typical, but fortunately NCTCOG includes hospitals and shopping malls as “special generators” in their models, so the model includes separate zonal data. Because these are common destinations for ADA paratransit, both of these variables were tested.

Description of the research 17 Site Selection Locations were needed to develop the models with good-quality ADA paratransit and an adequate travel model system. For each location, the process required conducting a travel diary survey of ADA paratransit users and using an existing regional travel demand modeling system. After consideration of more than 40 candidate locations, DART and the Fort Worth Transpor- tation Authority (FWTA) were chosen. Each candidate location was evaluated based on the following criteria: • The ADA paratransit system in operation there was believed to be operating without capac- ity constraints and generally employing best practices in operations and eligibility screening. • The ADA paratransit operator maintained a list of registered users and would allow the list to be used to survey those riders. • The local MPO had an existing regional model system that was typical (or better than average) in terms of the specification of input and output data. For this study, it was important that the system include an automobile ownership model that predicts car ownership distribution by combination of income group, household size, and number of workers in the household. • The local MPO was willing to provide access to their model data and a modest level of guid- ance in using it. • One or more members of the study team had worked with the MPO and with their data and were familiar with their models. Other locations declined to participate, were unable to provide rider information due to con- fidentiality concerns, had inadequate travel models, were believed to have possible ADA service quality issues, or were unfamiliar to the study team so that information was lacking to make an assessment. DART and FWTA met the criteria and agreed to participate in the study. Having two systems in the same metropolitan area raised some concerns, but also had significant advantages, because both are within the jurisdiction of the same MPO with the same travel model. The MPO, the NCTCOG, agreed to participate. The agency’s travel model was already known to the study team. Also, despite being in the same metro area, Dallas and Fort Worth have the positive feature of being significantly different with respect to demographics and ADA paratransit system char- acteristics. Compared to Dallas, Fort Worth has about one-fourth the population, less density of fixed-route transit service, a higher poverty rate, and a higher percentage of older women. Look- ing at ADA paratransit service, Fort Worth makes less use of conditional eligibility and uses a 30-minute on-time window (compared to 20 minutes in Dallas); both charge $2.75 per trip. As it turned out, it was not possible to make detailed use of these differences in service characteristics. DART’s ADA paratransit system is known just as DART ADA paratransit, while FWTA’s is known as Mobility-Impaired Transportation Service (MITS). Data Collection The principal source of data for model estimation was a travel diary survey of 800 ADA para- transit users, including 400 DART users and 400 MITS users. The travel diary survey was similar in concept to surveys conducted by MPOs for estimating urban travel demand models. These surveys typically ask for details of all trips made by all household members during one or two specific days. A typical sample size for regional travel surveys is 3,000 to 6,000 households. The number of ADA paratransit trips reported in such a survey tends to be quite small and not ade- quate to estimate separate models for those trips. In fact, ADA paratransit is rarely given its own mode category—it is often relegated to the “other” category to be written in by the occasional

18 Improving aDa paratransit Demand estimation: regional Modeling respondent. Because of these limitations, a special-purpose ADA paratransit travel survey was needed for this research. This survey was used for estimating all of the model components, except the model of ADA paratransit registration. Because data was needed only from people who had already applied for and obtained eligibility to use ADA paratransit, the survey could be conducted using a well- defined sampling universe (all ADA paratransit registrants of the participating transit operators) whose contact information was already in the databases held by the transit operators. Compared to the process for a typical regional household travel survey, this process had several advantages: • It was only necessary to collect data from the ADA-eligible person(s) in a household, rather than from every household member. • Because people with disabilities tend to make fewer trips than the average person, the travel diary period could be extended beyond a single day without adding significant respondent burden. • It was possible to use the ADA paratransit operators’ registration data to contact respondents, greatly increasing response rate compared to a survey using a typical random sample. • By combining the registration lists with the operators’ databases of actual trips, it was possible to stratify the sample according to trip frequency and age group and to calibrate the results of the travel diary reporting using actual total trip-making. The ADA paratransit travel survey was conducted from September 2009 through June 2010 by a professional research firm experienced in this type of data collection. DART and FWTA pro- vided the contact information of ADA paratransit customers who could be asked to participate in the survey. The study was divided into two survey efforts. First, a pilot survey was conducted from September 2009 to December 2009 that yielded complete demographic and travel behav- ior characteristics for 17 ADA paratransit users. The purpose of the pilot survey was to test the survey methodology, evaluate respondent materials and comprehension, and gauge participa- tion rates. The full survey was fielded subsequent to the pilot survey, from January 2010 to June 2010. As an incentive, respondents were told that, for participating in the study, they would be entered into a random drawing for a cash prize of $200. The $200 honorarium was awarded to five respondents selected at random at the conclusion of data collection after all data had been processed. The procedures for the survey were similar to standard procedures for conducting a travel survey and included the following 10 stages, which are described in more detail below: 1. Sample Selection 2. Opt-Out Mailing 3. Recruitment Telephone Interview 4. Respondent Packet Mailing 5. Reminder Call 6. Data Retrieval Telephone Interview 7. Reminder Postcard 8. Processing 9. Real-Time Geocoding 10. Data Edit Checks and Cleaning Sample Selection The sample consisted of a stratified selection of DART and FWTA customers. The sample was limited to adult (age 16 and older) ADA-eligible riders (excluding personal care attendants or companions) who had either ridden the ADA paratransit service in the preceding 12 months or (in Fort Worth only) become eligible in that same time period. DART and FWTA provided lists

Description of the research 19 of eligible customers, including date of eligibility certification, date of birth (for calculating age), and the number of trips taken in the preceding 12, 6, and 3 months. The sample was stratified by age and trip frequency using four trip frequency categories and four age categories. The purpose of this stratification was to ensure adequate coverage of all ages, including both frequent and infrequent users. An initial stratification used in the pretest produced numerous respondents who did not travel at all during the 2 days assigned for recording their travel. In response, the stratification was modified as described here and summarized in Table 2-1. The first table in Table 2-1 shows the actual distribution of the client bases used for drawing the sample (7,220 DART records and 4,584 FWTA MITS records) across the age and frequency categories. This excludes people who had not ridden or registered in the past 12 months, records with missing age data or age under 16, missing telephone numbers, or people who were already included in the pilot survey. The second table shows the target distribution for the survey, which was used for drawing the sample to be contacted. The row and column totals in bold were set to achieve the desired totals in each age group overall and each frequency group overall. Compared to the pretest sample, these targets increased the sample of frequent users in the sample (to avoid too many respondents who would have no travel to report) and also increased the number of Actual Distribution of ADA Paratransit Registrants Age Trip Frequency (trips in last 3 and 12 months) Total 6+ trips per month in the last 3 months 1-5 trips per month in the last 3 months No trips in the last 3 months, but trips in the last 12 months No trips in the last 12 months, but registered in the last 12 months Dallas 16-44 10.3% 5.8% 5.4% 21.5% 45-64 12.6% 15.4% 11.5% 39.5% 65-79 5.3% 12.2% 9.2% 26.7% 80+ 2.1% 5.5% 4.6% 12.2% Total 30.4% 38.9% 30.8% 100.0% Fort Worth 16-44 6.7% 3.9% 3.1% 2.0% 15.7% 45-64 10.6% 13.1% 10.1% 7.4% 41.2% 65-79 5.4% 9.9% 7.1% 5.9% 28.3% 80+ 1.9% 5.7% 4.2% 3.2% 14.9% Total 24.5% 32.6% 24.4% 18.5% 100.0% Target Survey Sample Distribution Age Trip Frequency (trips in last 3 and 12 months) Total 6+ trips per month in the last 3 months 1-5 trips per month in the last 3 months No trips in the last 3 months, but trips in the last 12 months No trips in the last 12 months, but registered in the last 12 months Dallas 16-44 21.39% 5.45% 3.15% 30% 45-64 22.08% 12.18% 5.74% 40% 65-79 7.94% 8.18% 3.88% 20% 80+ 3.59% 4.19% 2.22% 10% Total 55.0% 30.0% 15.0% Fort Worth 16-44 21.49% 5.74% 1.02% 1.74% 30% 45-64 21.51% 12.29% 2.08% 4.12% 40% 65-79 8.77% 7.42% 1.17% 2.64% 20% 80+ 3.24% 4.54% 0.73% 1.49% 10% Total 55% 30% 5% 10% Table 2-1. Sampling targets.

20 Improving aDa paratransit Demand estimation: regional Modeling younger persons in the sample—both because their behavior may be different from the older age groups and because younger people tend to be more difficult to contact during the survey process. In summary, the sample stratification and weighting was done in order to obtain a useful sample for modeling as efficiently as possible. It is desirable to include some respondents who make no trips or very few trips in the sample in order to model travel frequency, but for model- ing mode choice and destination choice, it is more efficient to survey people who will make a fair number of trips, including trips by ADA paratransit and by other modes, during the 2-day survey period. When modeling the data, the observations were re-weighted to be representative of the full population of possible (certified) ADA paratransit users. The fractions in the cells in the second table were calculated using iterative proportional fitting (IPF), starting with the actual cell percentages in the first table, and making the small- est possible adjustments which would match both the row targets and column targets in the second table. By dividing the cell percentages in the second table above by those in the first table, weights were calculated to use in drawing a probability sample for the survey that would match the cell targets. Based on the results of the pretest, the sample provided to the survey company included 5,304 records to allow for those who would opt out, not be contacted, or decline to participate at any stage. Opt-Out Mailing In order to address privacy concerns, DART and MITS customers were contacted via mail to inform them of the upcoming study and to invite them to opt out of the study if they so chose. Respondents were able to opt out by calling, emailing, or mailing their respective transit agen- cies. Those users choosing to opt out of the study were removed from the sample list used for data collection. Only 176 people chose to opt out within the 2 weeks allowed for this. An addi- tional handful requested to opt out after data collection had begun and they were also removed from the sample list. Recruitment The selected sample of ADA paratransit users were contacted by telephone. The purpose of the survey was explained, and respondents were assured that none of the information they would provide would be given to DART or FWTA. They were told participants would be entered in a drawing for a $200 honorarium. They were then asked if they would participate. For those respondents who agreed to participate in the study, a demographic interview was conducted to obtain data about the household and their members, including household size, number of vehicles, household income, dwelling type, age, gender, education level, driver’s license status, employment status, student status, and address. At the end of the recruitment interview, the respondent was assigned 2 travel days and arrangements were confirmed to call back and retrieve information about travel on these days. In total, 1,455 respondents were recruited and agreed to complete a 48-hour travel diary. Respondent Packet Mailing Travel diaries and Stated Preference materials were mailed to each respondent who had agreed to participate. The travel diary was an 18-page booklet with space for respondents to record information about locations visited, time of travel, mode used, traveling companions, and activ-

Description of the research 21 ity performed (purpose of the trip). Pages 2 and 3 from the diary, including instructions and a sample page for recording trip data, are shown in Figure 2-2. The Stated Preference materials consisted of a 4-page booklet, “ADA Paratransit Choices,” which presented five scenarios in which respondents were asked to make tradeoffs among service variables, including fare, travel time, reliability, telephone hold time, and advance reservations period. There were eight versions of the “ADA Paratransit Choices” booklet, each with a differ- ent set of scenarios. Figure 2-3 shows a sample. Table 2-2 shows the choices in each of the eight versions. The service variables measured were • Fare: the fare for a one-way ADA paratransit trip; • Travel time: multiple of automobile travel time that an ADA paratransit trip can take; • Late: number of trips out of 20 in which the pick-up or drop-off might be late (half of respondents received choices based on pick-up time and half received choices based on drop-off time); • Reservations: how many days in advance reservations can be made (by law, reservations must be accepted up to close of business 1 day in advance); and • Hold time: how many minutes customer may wait on hold to reserve a trip. Figure 2-2. Travel diary pages.

22 Improving aDa paratransit Demand estimation: regional Modeling Figure 2-3. Stated preference materials.

Description of the research 23 Reminder Call The night prior to the assigned travel day, reminder calls were made to the respondent. This reminder call served three key purposes: • To confirm that the respondent received the packet and to answer any questions respondents might have about using the diary to track their travel; • To schedule an appointment for the retrieval interview; and • To increase the likelihood that the household would follow through with recording their travel by reiterating the importance of the study and the household’s commitment to participate. For those instances where an answering machine was reached, the interviewers left brief mes- sages that referenced a toll-free number for respondents to call if they had questions. Data Retrieval Telephone Interview The day after the assigned travel period or at the appointed time, telephone calls were made to retrieve the travel data recorded by each respondent in his/her travel diary. The interviews were guided using Computer-Aided Telephone Interviewing (CATI) programs of the retrieval interview form. The average retrieval interview length was 20.31 minutes. The retrieval rate was 55 percent. This was calculated by dividing the completed retrieval calls (807) by the number of recruited households (1,455). Processing Data processing took place throughout the survey, beginning with the release of the sample for recruitment, processing recruitment data for the respondent mailing, appending the retrieval Choice 1 Set 1 A B Choice 1 A B Choice 1 A B Choice 1 A B Travel time 2 3 Travel time 2 3 Late 2 in 10 3 in 10 Reservations 14 1 Fare 200 100 Hold time 3 1 Fare 300 100 Hold time 5 3 Choice 2 A B Choice 2 A B Choice 2 A B Choice 2 A B Fare 300 200 Hold time 1 3 Fare 200 300 Hold time 1 5 Late 1 in 10 3 in 10 Reservations 1 4 Travel time 2.5 2 Travel time 2.5 2 Choice 3 A B Choice 3 A B Choice 3 A B Choice 3 A B Late 2 in 10 1 in 10 Reservations 14 4 Travel time 3 2 Travel time 2 3 Reservations 14 4 Fare 300 100 Hold time 1 5 Late 3 in 10 1 in 10 Choice 4 A B Choice 4 A B Choice 4 A B Choice 4 A B Reservations 1 14 Late 2 in 10 1 in 10 Hold time 5 1 Fare 100 300 Hold time 1 5 Fare 100 200 Reservations 14 4 Late 3 in 10 1 in 10 Choice 5 A B Choice 5 A B Choice 5 A B Choice 5 A B Hold time 5 3 Late 2 in 10 1 in 10 Reservations 1 14 Fare 200 300 Travel time 2.5 3 Travel time 2 2.5 Late 2 in 10 3 in 10 Reservations 1 4 Choice 1 A B Choice 1 A B Choice 1 A B Choice 1 A B Travel time 3 2 Travel time 3 2 Hold time 1 5 Late 3 in 10 1 in 10 Late 2 in 10 3 in 10 Reservations 14 1 Late 2 in 10 1 in 10 Reservations 14 1 Choice 2 A B Choice 2 A B Choice 2 A B Choice 2 A B Late 3 in 10 1 in 10 Reservations 1 4 Late 3 in 10 1 in 10 Reservations 1 14 Hold time 3 5 Late 1 in 10 3 in 10 Travel time 2.5 3 Travel time 2.5 3 Choice 3 A B Choice 3 A B Choice 3 A B Choice 3 A B Hold time 1 3 Late 1 in 10 3 in 10 Travel time 2 3 Travel time 3 2 Fare 300 100 Hold time 5 1 Reservations 4 14 Fare 100 300 Choice 4 A B Choice 4 A B Choice 4 A B Choice 4 A B Fare 100 300 Hold time 5 3 Reservations 14 1 Fare 300 200 Reservations 1 14 Fare 100 200 Fare 300 200 Hold time 1 5 Choice 5 A B Choice 5 A B Choice 5 A B Choice 5 A B Reservations 4 1 Fare 300 100 Fare 100 300 Hold time 3 1 Travel time 2.5 2 Travel time 2.5 3 Hold time 5 1 Late 2 in 10 3 in 10 Set 2 Set 3 Set 4 Set 5 Set 6 Set 7 Set 8 Table 2-2. Stated preference choice sets.

24 Improving aDa paratransit Demand estimation: regional Modeling data to the master data tables, and performing quality control on the data. A master control file tracked the progress of each respondent through the various survey stages with codes to allow for immediate identification of problem cases that were not progressing according to schedule, as well as for confirmation that cleared cases moved along as appropriate. Real-Time Geocoding All trip-ends and habitual addresses were geocoded during the retrieval telephone inter- view. The geocoding software was designed to provide interviewers with study-area details (e.g., road names and landmark references). Interviewers used this additional detail to con- firm respondent-reported locations in real time. Once the interview was completed, full address information, with matching x-y coordinates, for 100% of the locations was imme- diately available. Data Edit Checks and Cleaning Routine and customized data quality checks and data cleaning were performed on master data files. Routine checks included such items as • Data range checks (was there data outside the expected range?); • Checks for intra-household travel inconsistencies; • Checks for missing data (this was done by a combination of queries and direct data viewing of the internal delivery files, and minimized processing problems); • Checks for proper data skips; • Checks to ensure that deliverable files included the data items on the matrix and that variables were properly named; and • Checks for high frequency of item non-responses (checked throughout data collection). Data cleaning and preliminary data analysis reduced the number of usable, completed travel surveys to exactly 800. Characteristics of the Survey Sample Because of stratification, the survey is not representative without weighting (which was done for the model estimation described in the next chapter). Chapter 4 presents weighted tabula- tions. The highlights below represent characteristics of the sample used for the research. Survey Response Characteristics 1. The sample was split almost exactly 50/50 between Dallas and Fort Worth: 406 from Dallas DART and 394 from Ft. Worth MITS. 2. Only about 6% refused to answer each of the Stated Preference questions. 3. 88% said they were willing to participate in future surveys. Demographics of the Sample 4. There were more African-American (49%) than White (41%) respondents; 12% identified as Hispanic or Latino. 5. Very few people live in nursing homes or assisted living. (This was deliberate as part of survey administration procedures. Initially, some attempts were made to interview nursing home residents, but response rates were very low, and nursing home staff were generally not cooperative.) 6. Regarding household characteristics: 41% lived alone and 30% lived with one other person; 52% lived in a single-family unit and 44% lived in a duplex; 61% had an annual household income under $15,000; 52% lived in a household with no vehicle.

Description of the research 25 7. 15% were employed and 1% were students; 54% described themselves as “disabled/on dis- ability status” and 22% described themselves as retired. 8. 67% of the respondents were female, 68% were age 50 or older, and 29% were age 65 or older. This apparently low representation of older people matches the target set in the sam- ple stratification and was corrected in the weighting procedure. 9. 58% of respondents had a physical/motor impairment, 13% had a visual/sensory impair- ment, 11% had a mental/cognitive impairment, and most of the others had some combina- tion of impairments. It is likely that the sample is somewhat biased toward physical/motor because they have the easiest time completing the survey. 74% schedule their own trips, which may also be a bit biased compared to all users. Also, only 8% gave their answers through a proxy. Travel Characteristics 10. About 37% of respondents stayed home all day on Day 1, and about 43% stayed home on Day 2. Only 22% stayed home on both days. That is lower than had been feared. There is some non-response/survey fatigue bias on Day 2 compared to Day 1, which is typical. This was adjusted for in the model estimation process. 11. The most common reasons for not traveling on the diary day were “homebound elderly or disabled” or “no plans to travel that day.” Most people who used the first reason did so on both days, whereas most people who used the second (no plans to travel that day), made trips on the other day. 12. 34% had a valid driver’s license. 13. Only 6% had another specialized transit service available. 14. There were 2,681 person-trips, or about 1.7 per person-day. Of those, 27% were DART ADA paratransit trips, 22% MITS ADA paratransit, 25% automobile passenger, 8% scheduled transit, 7% walk, 5% wheelchair/scooter, 4% car driver, and only 2% other modes. 22% of trips used wheelchair or scooter in combination with other modes. 15. For almost half of the trips by auto/truck/van passenger, there was no other household member in the vehicle, meaning that the persons often get a ride with somebody else from outside the HH (household). 16. About 38% of reported ADA paratransit trips on both systems were subscription and 59% were scheduled for that trip. Most scheduled trips were scheduled 1 or 2 days in advance, although some were scheduled to 7 days (DART) and 14 days (MITS) in advance. 17. All trip-ends were successfully geocoded.