Study design and data source
We conducted a cross-sectional analysis of the 2001 NHTS. The NHTS was developed by the US Department of Transportation, with input from the Bureau of Transportation Statistics, the Federal Highway Administration, and the National Highway Traffic Safety Administration. The 2001 NHTS obtained information from a nationally-representative sample of households. Eligible participants were civilian, non-institutionalized persons who considered themselves primary residents of households sampled. Group housing settings were excluded. Types of data collected include information on the purpose, mode, transit time, trip length, and other related aspects of daily trips taken within a 24-hour period.
The survey was designed as a "list-assisted random digit dialing survey" , (p.3-3). To draw the sample, all phone numbers in the US were grouped in "100-banks" (lists of numbers for which only the last two digits differ). Numbers were first sorted by Census division and by metropolitan/non-metropolitan area location. Metropolitan areas were then sorted by size. Non-metropolitan areas were sorted by state and within state, by county. A serpentine ordering, north to south and east to west, was used to proceed through counties and states.
The survey took place in multiple stages: an introductory letter mailed to selected households, a screening/recruitment phone call, a travel diary package mailed to participants, and a final phone call to record results of the travel diary. Households were not restricted to those with private vehicles, as use of public transportation, walking and other modes was also of interest. Small cash incentives ($5 US) were provided to enhance response. The travel diary contained instructions for recording travel of each household member over a specified 24-hour period, their "travel day" . Data collection took place from March 2001 through May 2002 to provide a representative year of travel.
The total NHTS data set includes responses from 69,817 households, comprised of a national sample of 26,038 households, plus two add-on surveys conducted for selected states, of 28,899 and 14,880 households, respectivel . The overall response rate for the NHTS was 41% . The low response rate is believed to result from consumer resistance to unsolicited phone calls, language barriers, the multi-stage interview design (respondents could decline or drop out at several points), and the high level of participant burden . Response rates differ significantly by home value, race/ethnicity, the number of adults and presence of married persons in the household, and the type and size of the dwelling. Low socio-economic status was correlated with high response rates for the screener/household-level interview, and with low response rates for the extended/personal-level interview . Survey responses are weighted to account for under-response among specific populations. However, adjustments may not fully compensate for under-represented groups.
We subset the 2001 NHTS to households in which one member made at least one trip for "medical/dental services." [, p. M-40] No finer distinctions, such as between medical and dental, or within categories of medical care, were made by the data collection instrument. We used the term "medical/dental" throughout the paper to indicate the non-specific nature of the trip. The NHTS data described 3,914 trips made by 2,432 households, which were then weighted to provide national estimates. The unit of analysis was the trip.
Definition of variables
The dependent variables were road distance traveled for medical/dental care and time spent in travel. Distance was recorded as miles along roads from a starting location (which may be home, work or other) to a final destination for that trip. Thus, the NHTS measures one-way distance to care. (For purposes of this manuscript, distances are converted into kilometers in text, while the original units are retained in tables.) Time is recorded as the minutes spent on that trip. To provide context for readers not familiar with transportation patterns, we also provide descriptive statistics for travel to work among the households reporting travel for medical/dental care. Consistent with recent research, we selected 30 miles (48.3 kilometers) [35, 36] and 30 minutes per trip  as measures suggesting a "high" travel burden. Each of these measures is examined separately in multivariate analysis.
Independent variables of key interest
Our study uses the urban/rural variable developed by Claritas, Inc, included with the NHTS to characterize each household. The Claritas variable categorizes geographic units based on population density and proximity to an urban center. First, a 4-square mile grid network was overlaid on the US and, using population estimates from the underlying census block groups, population counts for each square were calculated. Then, to calculate "contextual density," Claritas added the population counts for each grid cell to those of the eight surrounding grid cells, and then divided that by the total area (36 square miles). These "contextual population density" values were then ranked from 0 to 99 to create density centiles. Grid squares with values 19 or less were defined as "rural" by Claritas and by the present study, and all grids with a value of 20 or more were grouped as "urban." Subdivisions within the urban category were not distinguished.
The use of population density and the proximity to an urban/metro area to define level of rurality is common to many classification schemes, including the Rural-Urban Continuum Codes (RUCCs)  and the Urban Influence Codes (UICs) . Both RUCCs and UICs are county-based, while the Claritas measure is not derived from political boundaries. The Claritas classification scheme, which uses much smaller geographic units of analysis (4-square mile grid squares), has finer resolution than do county-based codes.
Race and ethnicity is reported as white, African American, Hispanic, and other. The NHTS obtains information on multiple race/ethnicity groups, such as Asian Americans and American Indian/Alaskan Natives. However, persons in these racial and ethnic groups were not sufficiently represented in the travel-for-care population to support independent analysis, and were included in an "other" race and ethnicity category.
Factors in addition to residence and race/ethnicity are known to influence travel for care. Control variables, held constant in multivariate analysis, were conceptualized at three levels: characteristics of the traveler, the trip, and the community. Traveler characteristics included age, sex, educational attainment, occupation, income, family size, and the presence of "a medical condition that makes it difficult to travel outside of the home"  (p. M-9). The NHTS did not obtain information on specific medical conditions. Trip and travel characteristics included day of the week, the time of day, and the mode of travel. Ecological factors were perceived traffic conditions, region, and job density. The perceived traffic conditions variable is based on three NHTS questions about travel problems. Persons who agreed that the price of gasoline, rough pavement, or highway congestion were "very much" or "severe" problems for them were categorized as having traffic problems; others were coded as not having such problems. (The NHTS inquired about a total of 11 potential problems, but only the three items listed were asked of all respondents. Remaining items were randomly asked of 50% of respondents.) Region was based on four major Census regions. Job density refers to the number of jobs per square mile in the Census tract in which the household is located and was a Claritas-developed variable included with the NHTS data set.
Descriptive and multivariate analyses were conducted in SAS-callable SUDAAN. All estimates presented are weighted appropriately to reflect the complex NHTS sampling design and yield nationally representative estimates. The need for weighting to account for under-represented groups has already been described. In addition, specific states could purchase larger sample sizes, allowing them to make sub-analyses of interest. Survey weights correct for the over-representation of such states when developing national projections.