The National Household Transportation Survey (NHTS), a survey of the U.S. Department of Transportation, is designed to assess the mobility of the American public (USDOT FHWA 2011). NHTS gathers data on daily personal travel, including information on household and demographic characteristics, employment status, vehicle ownership, trips taken, modal choice, and other related transportation data pertinent to US households. This survey is a continuation of the Nationwide Personal Transportation Survey (NPTS), which was conducted in 1969, 1977, 1983, 1990 and 1995; and the NHTS has been conducted in 2001 and in 2009. The 2009 NHTS collected travel data from a national sample of civilian, non-institutionalized population of the United States - 25,000 households in the national sample and separate samples from 20 add-on areas, which together provided data on 150,147 completed households.
While the NHTS is an excellent source of travel information for large geographic areas in the US, it has the difficulty of limited sample sizes for small area estimation. While statistically valid models relating such travel measures as person-trips, household-trips, person miles traveled and household miles traveled can and have been created for national and regional areas of the US (e.g., Scuderi and Clifton 2005, Pucher and Renne 2003), transferring the results of such models proves to be difficult based on the limited sample sizes of the small geographic areas. Such transferability studies often rely on other sources of more detailed information for the smaller geographic areas in order to estimate travel behavior.
In this study, the NHTS data are broken down into six geographical areas and urban/suburban/rural classifications to make estimates of several travel variables, based upon a set of household and demographic characteristics. These estimates are then transferred to individual census tracts using the household and demographic data for each of those census tracts. While these individual census tract estimates may have limited accuracy in some cases, they can be very beneficial to local governments, and other interested customers, who often do not have the budget and/or time for conducting their own surveys. Using these estimates can make economic sense for those agencies, even if the results are less accurate than if they conducted their own survey. Additionally NHTS has the advantage of using questions standardized across the geographic sample (with only small variations for the add-ons), which would not be possible when comparing local surveys with differing methodologies.
Henson and Goulias (2001) used the 2001 NHTS and local travel surveys by transferring a survey participant’s daily travel schedules to different geographic locations by, connecting travel behavior and land use, urban form and accessibility. In this study, the NHTS participants were set into cluster models representing persons, land use and travel. These clusters were then compared to similar groupings from two local travel surveys – but their research indicated that people with different geographic areas do travel differently, even if they share the same socio-demographic characteristics. Wilmot and Stopher (2001) found that updating transferred values with local values from small surveys results in a better transferred model from the larger survey. Mohammadian et al. (2010) propose a technique to simulate disaggregate and synthetic household travel survey data through spatial transferability of travel data. Bayesian modeling was used to create a ‘synthetic’ population for the State of New York (excluding Manhattan), which was then linked to travel estimates developed from the NHTS data. Stopher et al. (2005) used local survey data to supplement (through simulation) national data. In this study, distributions of travel characteristics were obtained from a nationwide sample, which were updated to a locality by using a small local sample and Bayesian updating. Long et al (2009) used small area estimation methods to produce reliable estimates of household travel characteristics at both the aggregated and disaggregated (household) level. Data were drawn from the 2001 NHTS and the CTPP.
The Federal Highway Administration (FHWA) contracted with researchers at Oak Ridge Laboratories to conduct work in this area of transferability to expand the usability of their NHTS data to small geographic areas – in particular, Census tracts. First utilizing the 1995 NPTS survey data, Reuscher et al. (2002) estimated local travel, which included vehicle trips (VT), vehicle miles of travel (VMT), person trips (PT), and person miles of travel (PMT). The three steps of research were used to obtain these estimates:
- Classify the Census Tracts into homogenous groups. The census tracts were first split by income (very low, very high, and the rest), then split by area type (urban, suburban and rural), and then split by cluster analysis, based on income, employment rate and number of vehicles.
- Use NPTS data to estimate driving characteristics for each of the clusters derived in the previous step.
- With the classification and estimations from the above two steps, make travel estimates for any census tract derived from the tract’s classification.
Census tracts without population or without vehicles were excluded from the analysis, as were the Manhattan census tracts.
In a subsequent study using 2001 NHTS data (Hu et al. 2007), the clustering approach fared poorly. Hu et al. tried to determine if the 2001 NHTS data had increased statistical noise – but that was not the case. They also thought that it may be due to a potential ‘survey/firm’ effect (different firms doing the surveys) – but that also was not the case. So – instead of utilizing the clustering by income and rural/urban/mega urban, the researchers reduced the breakouts to urban, suburban, rural, mega-urban and extreme poverty. Extreme poverty classified tracts with greater than 40% of the population being below poverty level. Mega-urban was defined by being densely populated with highly used transit; 19 cities were classified as mega-urban. Regressions were performed within each of these ‘geo-economic clusters’ - one regression for each parameter (PT, PMT, VT, and VMT) and geo-economic cluster. The tables (1 and 2) below show the variables for these regressions.
For measuring success, each NHTS add-on was split into two samples. The four variables (PT, PMT, VT, VMT) were calculated for one of the two samples to create baselines for each add-on. The researchers then performed five different classification clusters / modeling on the second set of samples (Census division and MSA-sized base, MSA size-based, Census division-based, Census region-based and the above regressions). To determine which technique worked the best, the average baseline was compared to the calculated values of the variables by taking the percent difference from the baseline (Table 3 and 4).
This study uses similar methodology, utilizing multiple regression analysis, to estimate travel variables as a function of significant demographic and household characteristics. The following section gives details on the methodology.