This paper discusses considerations for the next series of personal travel surveys conducted by the Department of Transportation. After a brief discussion of the current National Household Travel Survey (NHTS) design, a broad range of methodological and design considerations are introducedoften in the context of other federal surveys and household travel survey experiences. This paper introduces topics such as whether the current design allows the NHTS to fulfill major objectives of the survey, the efficacy of simultaneous collection of daily and long-distance travel, considerations for improvement of data quality, the need to improve response rates, and the desire to maintain data on travel behavior trends.
KEYWORDS: Travel surveys, survey methodology, data quality, emerging technologies.
The primary objective of this paper is to introduce and discuss survey methodology considerations for the next series of personal travel surveys conducted by the U.S. Department of Transportation (DOT). After a brief discussion of the current National Household Travel Survey (NHTS) design and issues, a broad range of methodological and design considerations are introduced, often in the context of other federal surveys and household travel survey experiences. In addition, the following questions are posed that must be carefully considered in designing the next series of personal travel surveys.
The 2001 NHTS combined a daily travel survey, the Nationwide Personal Transportation Survey (NPTS), and a long-distance travel survey, the American Travel Survey (ATS). Both predecessor surveys were last conducted in 1995. The goal in combining the two surveys was to build a more comprehensive picture of household travel while reducing the cost and respondent burden.
The 2001 NHTS primarily employed the 1995 NPTS design with an expanded and more detailed long-distance travel section (i.e., trips of 50 miles or more) added at the end of the interview. The design consisted of a cross-sectional, random-digit dial (RDD) sample of approximately 26,000 households and 60,000 persons nationally, with additional samples in nine states and metropolitan areas.1 All interviews were conducted via telephone using a two-stage data-collection design. Interviews were conducted over a 14-month period, March 2001 to May 2002, to capture travel throughout the year.
Sampled households (with matched addresses) first received an advance letter with a $5 incentive, followed by a telephone screener interview to collect basic household information, and finally an extended telephone interview to collect trip detail from all household members on their assigned travel day and travel period. The travel day was pre-assigned for each household to ensure equal representation among days of the week and across the entire year. The travel period for long-distance travel was defined as the four-week period prior to and including the travel day. Attempts were made to collect travel information on all persons in the household. In order to be considered a completed or useable household interview, interviews had to be obtained from at least 50% of all household adults. Proxy interviews were required for all children under 14 years of age and were allowed, only in very limited situations, for adult household members.
One of the greatest challenges of any statistical survey is producing high-quality, useful data with a limited budget and resources. This will likely be an even greater challenge with the next series of personal travel surveys. In 2002, DOT commissioned the Transportation Research Board and the Committee on National Statistics to review and evaluate the NHTS. The group suggested several improvements that will need to be carefully addressed in the next survey (TRB 2003). In addition, federal statistical surveys are obligated to adhere to the policy and guidelines of external stakeholders, most notably Congress and the Office of Management and Budget (OMB).2
In June 2004, OMB in conjunction with the Federal Committee on Statistical Methodology, drafted a revised series of standards and guidelines for all federal surveys.3 Finally, the Confidential Information Protection and Statistical Efficiency Act of 2002 (CIPSEA),4 enacted in December 2002, mandates more stringent procedures pertaining to the collection, protection, and release of federal survey data. This legislation has significant implications for the accessibility of NHTS data and can impact what data are collected.
With the exception of modest changes,5 the NHTS design remained largely consistent for the collection of daily travel with that of the 1995 NPTS. Thus, the repeated design helped to preserve daily travel trend analysis over time. The long-distance travel component, however, underwent significant changes in definition, content, and methodology as compared with the 1995 ATS. Some key issues faced in conducting the 2001 NHTS and some important considerations for the next series of passenger travel survey(s) follow.
Sample size and methodology significantly changed for the long-distance component. The number of sampled households was reduced from almost 70,000 (1995 ATS) to 26,000 (2001 NHTS). The long-distance trip definition was also revised to include trips of 50 miles or more away from home (as compared with 100 miles in the previous survey). Long-distance trips were collected once for a four-week reference period, as compared with four waves of interviews over a one-year period. These changes resulted in a sample of far fewer long-distance trips, diminished ability to track long-distance travel trends, and difficulty in producing annual and seasonal long-distance travel estimates. In addition, the smaller sample size all but eliminated the ability to produce lower level geographic estimates and analyze travel flows.
Response rate to this multistage telephone survey was 41%.One of the biggest challenges for the 2001 NHTS was obtaining a high response rate, primarily meant to reduce the impact of nonresponse bias likely in surveys with lower response rates. In spite of many effortsuse of incentives, refusal aversion training for interviewers, refusal conversionthe 2001 survey achieved a household response rate of 41%. Given the complexities of the survey and the difficulty of achieving high response rates in an RDD design, the response rate was considerably lower than what is commonly expected in a federal statistical survey and the survey was only reluctantly approved by OMB.
Technology with positional information should be considered in future surveys. Research in the last 10 years suggests that using global positioning systems and, perhaps, cellular phone technology can be effective tools in capturing trips that are often missed using self-reported methods. Expenses incurred for incorporating these technologies have dropped dramatically as hardware costs decline, particularly in light of the Federal Communications Commission directives for positional accuracy for emergency calls (FCC E911). While many benefits may be gained, the decision must also consider the potential tradeoffs in cost, data quality, and statistical reliability.
Consideration of alternative sample design and data-collection methodology.The next series of passenger travel surveys should consider different sampling strategies and data-collection methodologies to address concerns related to coverage, nonresponse, timeliness, and operational efficiency.
Survey designs can generally be classified into two broad categories on the basis of whether they obtain repeated measurements on the sample of units over time: panel surveys do and cross-sectional surveys do not. In the United States, most travel surveys rely on one-time cross-sectional designs to collect information on travel consumption and behavior (Tourangeau et al. 1997). The NPTS/NHTS series can be most accurately described as a repeated cross-sectional design, since essentially the same survey is repeated over time with different samples using very similar survey questions and procedures.6 Given the significant changes in long-distance travel data collection in the 2001 NHTS, the 1995 ATS is more accurately defined as a single, cross-sectional survey. (Although respondents were interviewed four different times, similar to a panel survey, this design was used to pool estimates over a one year period and not analyze the change within households between interviews.) Table 1 describes four common travel survey designs, along with a brief description of advantages and disadvantages, and provides examples of each.
Other countries have transformed many large cross-sectional and panel surveys into continuously repeated surveys for many of the same reasons currently given for transitioning the decennial census "long form" into the continuously collected American Community Survey. Primary reasons include: 1) to flatten the budget for all years, rather than having high peaks during periodic data collection; 2) to retain staff expertise, both in field implementation, data processing, and data analysis; and 3) to improve estimates over time (i.e., national-level data are more timely while retaining the ability to make small area estimates over time). Some examples of national household travel surveys that are conducted continuously include:
However, as smaller samples are collected each year, multiple years of data are required for reporting subpopulations, including geographic subregions or subgroups based on sociodemographic or economic characteristics, such as a specific age or income group.
For the purpose of making objective statistical inference, the sample must be selected using probability methods, that is, where everyone in the target population has a known, non-zero probability of selection (Kish 1965). Likewise, the sample frame utilized in a national probability sample must be complete, accurate, and up-to-date to ensure adequate representation of the larger target population. For telephone surveys, problems with completeness of frame include the growing number of persons who have only a cell phone and who are not included due to costs incurred by potential respondents for incoming as well as outgoing calls, and the small number of households without telephones. For address-based surveys, the completeness of the address list must be evaluated. For example, the Census Bureau has a "Master Address File" that is updated with a U.S. Postal Service Delivery Sequence File. Omissions of certain groups in the sample frame can introduce coverage bias because of the exclusion of these groups.
To the extent that the nontelephone households differ from telephone households in their travel behaviors, coverage bias makes the results less representative of the U.S. population. According to the 2000 Census,7 the number of households without telephone service was estimated to be 2.4%. While the number of households without telephone service has decreased in the last decade, the number with only a cell phone is rising. A recent analysis using the February 2004 Current Population Survey supplement estimated the number of cell-phone only households to be as high as 6% (Tuckel and O'Neill 2004). In addition, these households were found to be disproportionately single-person, central city, and renters.
Address frames are subject to errors of omission as well. In the pre-test of the 2001 NHTS, it was estimated that the address frame contained addresses for 90% to 95% of households (USDOT 2001).
As a result of continued problems with population coverage (and low response rates,) the future of RDD-only design remains in question. A number of other sample frame options exist, but each introduces other disadvantages related to survey costs, data-collection methodology, response rates, and estimation. Table 2 summarizes the sample frames currently being used by other national surveys, along with some associated benefits and limitations.
Other sample frame alternatives are currently being researched, yet they pose many coverage issues and operational difficulties for household travel surveys, even as a dual-frame or mixed-mode approach. Internet surveys, for example, offer limited household coverage, because less than half (41.5% from the 2000 Current Population Survey 8) of U.S. households have access to the internet. Offered as another alternative data-collection mode, they appear to have more merit, but results to date have been mixed with frequent reports of low response and other data quality issues. In addition, the use of cellular telephone sampling frames is also limited, because no comprehensive list of cellular telephone numbers exists.
The number of units to sample is another key decision in the survey design. In choosing the required sample size, these factors should be considered:
With these factors in mind, the 2001 NHTS was designed to achieve interviews from approximately 25,000 households and 60,000 individuals within these households. This resulted in the collection of approximately 250,000 daily trips and 45,000 long-distance trips. This sample size allows for reliable estimation of many national-level trip characteristics for both daily and long-distance travel, but affords very limited estimation for lower levels of geography.9
Long-distance trip analysis was greatly affected by the sample size. The 2001 NHTS captured less than 5% of comparable trips captured by the 1995 ATS (i.e., noncommuting trips of 100 miles one way), resulting in the inability to make state estimates or analyze flows between states and major metropolitan areas. In addition, only limited analysis can be performed for rarer transportation modes (e.g., trips by bus and train) and for certain subpopulations (e.g., elderly travelers). The sample size would need to be increased substantially in order to ensure reliability of estimates at lower levels and for rarer groups, as well as to measure flows; however, an increase would be very costly.
Subgroups of Interest
As mentioned above, the NHTS sample size posed limitations for specific analyses due to the small size of the geographic and demographic subgroups. The 2001 NHTS national design included one stratum for metropolitan areas (MSAs) with rail service, but otherwise did not oversample specific, rarer groups. Historically, the NHTS sample design and size has limited the analysis of several important transportation modes and groups. A description of a few of these follows.
Transit users.Personal vehicle use dominates passenger travel and accounts for nearly 9 out of 10 trips in this country. Analysis of lesser used but important transportation modes, such as transit, is limited due to both the small sample and, in some cases, geographic sensitivity. Analysis of transit behavior has relied heavily on decennial census information on the share of transit for "usual mode to work" and the NHTS for transit share for all trips, regardless of purpose. These results are often cited by Congress in decisions related to transit investments. The add-on program in the NHTS that allows states and metropolitan planning organizations to purchase additional samples provides a disproportionate number of cases in certain geographical areas. For example, the New York state add-on sample in the 2001 NHTS also resulted in an unintended benefit. The New York metropolitan area has nearly 40% of the U.S. transit market based on the Federal Transit Administration National Transit Database. The addition of the New York add-on sample allowed for a unique analysis of transit behavior in an area representing the largest share of transit use. The added sample, reweighted for national estimation, also provided modest gains in precision for transit estimates at the national level.
Households without vehicles.Data from the 2001 NHTS showed that approximately 8% of households were without a vehicle for regular use. In the United States, households without vehicles are thought to include two main groups: 1) people who live in high-density urban neighborhoods (e.g., Manhattan, downtown Chicago, San Francisco, or Washington, DC) with good transit and taxi accessibility, and 2) recent immigrants who have not acquired a car (Murakami 2003; Pisarski 1996). Those in the second group may be less likely to participate in a national travel survey for several reasons: language barriers, potential distrust of government activities, and different concerns about privacy and confidentiality.
Race and ethnic groups.African Americans have been less likely to respond to travel surveys than white households. Little documentation is available on participation rates by Hispanic households in transportation surveys. However, larger households are more likely to fail to complete travel surveys as the burden of reporting is greater, and this finding may affect participation by Hispanic households (Murakami and Watterson 1992; DRCOG and Parsons 2000; Nustats 2003, 2004a, and 2004b).
Contrino and Liss (2004) conducted research on nonresponders in three regional surveys (Atlanta, Phoenix, and Ohio) and found that, in all three areas, minority and low income groups were more likely to be nonrespondents. As a result, Contrino and Liss recommend oversampling for low responding and special interest population groups and using targeted approaches for recruitment and retrieval, as well as the use of post-stratification weights to adjust for low participation. Oversampling strategies, if effectively employed, can be a valuable mechanism for producing more reliable estimates for specific subgroup analysis. However, assuming the overall sample size remains constant, it can also result in a loss of precision for national estimates. In addition, while oversampling may reduce the variance of these estimates, it does not necessarily reduce the potential nonresponse bias.
Many transportation analysts and modelers require data from all members of households for use in their analysis and models in order to capture joint decisions and household interactions. Consequently, attempts were made in the NHTS to interview all household members, and only households where at least half the household members were interviewed were considered complete or useable. Approximately 85% of NHTS useable households resulted in a complete enumeration of the household members. These households were provided with an additional set of weights to allow researchers the choice of using only those households with complete enumeration. Complete enumeration of a household is a challenging task and has negative impacts on the response rate. In the last several years, transportation researchers (Erhardt 2000) have begun to investigate whether changing the sampling unit to a person, rather than a household would improve response while maintaining the ability to simulate travel for a household.
Another key decision in travel surveys is the mode of data collection. Similar to sample design, each choice of data-collection mode has inherent advantages and disadvantages. Selection of an appropriate mode requires careful consideration of many factors, not the least of which is coverage of the target population. While the method of data collection might be largely dictated by the population coverage and sample frame, other common determinants include survey costs, response rates, and data quality issues. Mode selection can also be influenced by the complexity and length of the survey and timeliness needs.
Table 3 provides a summary of four popular data-collection modes: in-person, telephone, mail, and internet, along with their features. As this table illustrates, in-person data collection typically yields the most complete coverage, achieves the highest response rate, and produces the best quality data. Not surprisingly, in-person interviews are also the most expensive of the four modes. For this reason, telephone and mail modes are more commonly used despite well-recognized tradeoffs in data quality. Telephone interviews have been the most commonly used collection method in the United States over the past 20 years, as field costs for personal visits increased to prohibitive levels and other obstacles to personal interviews have arisen (e.g., personal security and gated communities).
The 2001 NHTS used a telephone data-collection methodology. Telephone interviews are often preferred over mail-back methods for travel/activity surveys as they allow for more probing for complete reporting of trips. In addition, because the NHTS captures travel by all household members, telephone retrieval allows for correction and validation of travel among household members.
Mixed-mode approaches are commonly used to strike a balance between survey costs and data-quality issues (most often response and coverage). A commonly used approach, taken in the American Community Survey (ACS), is to employ the least costly mode for initial contact, followed by a more costly mode for nonresponse followup, such as using a mail survey with telephone nonresponse or a telephone survey with in-person nonresponse followup. In dual-frame designs aimed at improving coverage, different modes are often required to capture the sampled units from each frame. While a mixed-mode approach can offer an effective mechanism for improving response and coverage, it also potentially introduces bias resulting from mode effects (i.e., a difference in responses due entirely to the method of data collection). Therefore, it is important to first evaluate the tradeoff of improved coverage and response with potential response error (bias) before deciding on a mixed-mode methodology. Recognizing the greater likelihood that future surveys, including the NHTS, will need to allow for multiple modes of data retrieval, appropriate research on modal influences on travel behavior data collections will be needed.
Electronic or computerized data-collection options are also commonly used for the entirety of the interviews or as an alternative methodology. The NHTS used a computer-assisted telephone interview (CATI) with an additional option for reporting specific information via the internet. Use of CATI was especially beneficial for the NHTS, allowing for trip rostering and capture of trips made by multiple household members. Therefore, trips already capture during preceding interviews could be verified by a subsequent household member instead of captured anew. Computer-assisted designs require more upfront planning and increased time to implement compared with paper-and-pencil surveys, but result in faster access to data and higher quality control as the need for data entry from paper forms is eliminated. In the future, the use of CATI, whether by telephone or in-person, presents opportunities for better integration of geographic information and other location-based data.
Nonresponse is not just an issue for the NHTS. In June 2004, Robert Groves presented an overview of the status of current household nonresponse to the Committee on National Statistics (Groves 2004). He found declining response rates in attributes that match those of the NHTS (i.e., surveys conducted in the developed world, one-time surveys compared with longitudinal surveys, and telephone surveys compared with in-person surveys). He listed techniques to reduce nonresponse, including prepaid incentives, cash incentives, more followup calls, and longer data-collection periods.
Many efforts were implemented in the 2001 NHTS design to achieve as high a response rate as possible. For example, respondents were sent incentives prior to contact. Interviewers were provided with special refusal aversion training, and refusal conversion efforts were attempted with survey nonresponders. The following are some factors thought to contribute to nonresponse in the NHTS:
Most cross-sectional household travel surveys utilize a multistage approach for interviewing households about their travel. As with the NHTS, advance letters introducing the survey were sent first, followed by a telephone contact to conduct a basic household-level screener interview. Respondents were mailed a travel diary, with information retrieved by another telephone interview. In any multistage approach, nonresponse can occur at each stage and compounds the overall nonresponse rate. In the 2001 NHTS, the recruitment rate was 58.2%, and the subsequent completion of the extended survey was 70.8%. The composite response rate was 41.2% (USDOT 2004).
The 2001 NHTS did not include a nonresponse followup survey. Traditionally, low response rates have been suspected of resulting in more biased results. However, more recent research (Groves 2004) also cites examples of surveys with high nonresponse rates and low bias, and interestingly, some attempts to reduce nonresponse that resulted in greater bias.
Nonresponse followup (NRFU) surveys have not yet been incorporated as standard practice in activity and travel behavior surveys. Some exceptions include work done by Richardson (2003) in Australia and a small test funded by the Federal Highway Administration (FHWA) in Denver in the late 1990s. The Victorian Activity and Travel Survey in Melbourne, Australia, conducted in-home interviews with a sample of nonrespondents to the main mail-back survey. This study indicated that nonrespondents to mail-back surveys were more like early respondents than late respondents in daily trip rates.
In the Denver Regional Council of Governments project, a brief mail-out/mail-back survey was conducted for nonrespondents to an RDD telephone survey (for those where an address could be found). Small cash incentives were found to double the response rate to the NRFU survey. They did not find statistically significant differences in household trip rates between the households who completed the full survey compared with those in the "quick refusal" and "noncontact" households. Therefore, the hypothesis that these nonrespondents to the telephone survey led to underreporting of trips was not supported.
For the ACS, the U.S. Census Bureau found that response rates varied widely, with particularly low mail-back responses in neighborhoods that were predominantly Native American (17%), Hispanic (34%), and African American (35%) (USDOC 2002) . The original plan for the ACS nonresponse followup, which was tested in their pilot, was a one in three field followup. The response rate to the field followup has been uniformly very high (between 92% and 95%). The Census Bureau now plans to implement differential nonresponse followup, with higher followup rates in areas with low mail-back returns (USDOC 2004).
Current NHTS methodology requires that an interview be completed from a respondent within six days of the assigned travel day. For respondents who neglect to complete a diary, recall errors are felt to be much higher after six days, especially for daily travel. The limited six-day window also eliminates the confusion of referencing the particular travel day that was assigned (e.g., Tuesday this week as opposed Tuesday last week). Although the six-day data-collection period appears to help in reducing response problems, it could also potentially contribute to nonresponse, especially if attempts are made to interview all household members within this relatively small window. In addition, potential bias may also be introduced in the capture of long-distance travel. Respondents who travel often and are away from home for longer periods of time are more likely not to respond.
Daily travel. As previously mentioned, the reference period for the travel day in the NHTS is a pre-assigned one-day period (from 4:00 a.m. on the travel day through 4:00 a.m. the following day). Other national travel surveys, such as the U.K. National Travel Survey and the German Mobility Panel use a seven-day diary. These surveys allow for examination of travel variability over a longer period. For example, a respondent may not go grocery shopping each day but only once a week. Similarly, a respondent may ride transit only two days a week. These longer reference periods, however, are more burdensome, typically achieve very low participation rates, and may result in fewer trips reported each day as the survey period continues. The Dutch National Mobility Panel found significant "trip reporting fatigue" in a seven-day diary (Golob and Meurs 1986).
Cost efficiency might suggest that a smaller sample with a larger reference period should be considered in order to continue generating similar numbers of trips overall. However, moving from a larger sample size with a one-day reference period to a smaller sample with a longer reference period would create additional estimation issues for lower geographic levels and subgroups. Although we might have the same number of trips, the effective sample size would be lower given the increased correlation between trip reports. Given current criticisms, reducing the household sample sizethus requiring more aggregation on characteristics and even more limited analytic potentialwould not likely be perceived as an improvement.
Long-distance travel.The NHTS reference period for long-distance travel was the four-week period before and including the travel day. Therefore, if a respondent's travel day was July 30, their assigned travel period would be July 3 to July 30. This brings into question the respondent's ability to accurately recall trips for this period, and telescoping effects are potentially introduced (i.e., they might be reporting trips taken outside the travel period, e.g. on July 1 to 2). Due to the rotating nature of the travel period, it further introduces difficulties in producing seasonal and annual estimates of long-distance travel. Introducing longer, more salient reference periods, however, can also be problematic. For example, given that the respondent is interviewed only once, it is unlikely that he or she would be able to accurately recall all trips for one year, not to mention the burden of this request.
In the 1995 ATS design, respondents were interviewed quarterly over a one-year period. This methodology allowed for bounding and dependent recall of previous trips, thus reducing telescoping effects. As is common in panel designs, however, time-in-panel or conditioning effects were also evidenced by the declining trip rates in later waves of interviewing.
Only limited proxy reporting for adult household members was allowed in the NHTS, resulting in approximately 80% of the interviews being conducted with the respondent. Self-reports are preferred in travel surveys due to diminished accuracy and completeness in trip reporting often experienced when proxy reporting is allowed. In one travel survey conducted in Toronto, researchers found that home-based discretionary and nonhome-based trips were underreported by proxy, with gender a related factor (Badoe and Steuart 2002). Bose and Giesbrecht (2004) found in the 2001 NHTS that average trip rates for persons interviewed by proxy were much lower than those who reported for themselves. The average daily trip rate was 4.5 for self-reports as compared with 3.7 for proxy reports. Proxy reports were more likely in the NHTS if the respondent was male, a nondriver, had less education, was away from home on the travel day, or had a disability that affects travel. Proxy reports also tended to have fewer daily, long-distance, walk, and bike trips, and transit usage.
Broeg and Ampt's (1983) continuing work assigns "caseloads" to individual interviewers or "motivators." Their methods rely primarily on mail-out/mail-back techniques, with telephone calls used for reminders and for queries when responses are missing or other problems exist. Because the 2001 NHTS used telephone retrieval, it was nearly impossible to assign cases to an individual as call-backs were scheduled over many different hours and different days of the week. A small test was completed in 2002 in the Washington, DC, region and a small team was assigned a caseload. The survey period was very compressed, and thus results were inconclusive but seemed positive (Freedman and Machado 2003). In debriefing the interviewers, the interviewers felt more confident and comfortable when making subsequent phone calls. Some respondents said they wanted the first caller to call them back, not someone else on the team. One of the drawbacks was that the current scheduling software was not optimized for team assignments.
Improvements in questionnaire design should be made to assist the respondent in providing complete and accurate information that the analyst is attempting to collect. The main issue for travel surveys is to ensure that all trips are reported, otherwise, they become a serious problem of item nonresponse. Techniques that have been used to improve reporting include:
Daily trips.A daily trip in the NHTS was defined as each time the respondent went from one address to another. One of the greatest difficulties in travel surveys is to capture short stops, because people may neglect to report them for several reasons:
Long-distance trips.In contrast, people in the United States do not have a good estimate of distance, so questions about trips of over 50 miles in length are often overreported (in spite of interviewer aids such as maps). Respondents will often report trips that are closer to home than the long-distance definition. In the processing of both the 1995 ATS and the 2001 NHTS, 20% to 25% of long-distance trips were later excluded after the calculated route distance illustrated that these trips were under the specified mileage required (i.e., 100 miles for the 1995 ATS and 50 miles for the 2001 NHTS). Current issues facing the next collection of long-distance data include trip length criteria, the amount and type of detail most important, and how to define and collect trips or journeys with multiple stops and/or side trips.
Diary formats have evolved primarily through focus group testing. Some of the questions on the visual appearance and format of travel/activity diaries include:
In addition, NHTS respondents were also mailed a map delineating a 50-mile radius from their home location. Although the map was somewhat misleading, it was thought to have served as an effective memory jogger.
A couple of pilot tests have been conducted (Bachu et al. 2001; Stopher et al. 2004) using passive global positioning systems (GPS) and then supplying a map to respondents to use as a recall device. They found that people were able to recount their trips, even two weeks later by looking at the printed maps showing their GPS-recorded travel. This approach relied on the map-reading ability of respondents. In the Australian pilot (Stopher et al. 2004), respondents were also given the option of looking at a tabular description of each stop showing street names, arrival times, and travel times.
During the last couple of decades, travel behavior researchers have focused increasingly on activity-based travel survey approaches. (One of the first known uses of an activity diary for travel research was in Belgium in 19861987.) This is due in part to the desire to understand travel in the context of daily activities and allows analysts to bring this context into travel analysis and modeling. Activity approaches allow transportation researchers to examine the activities and relationships that generate the need for travel (Harvey 2003). Traditional trip-based travel surveys, such as the NHTS, enumerate all trips taken by persons during a specified time period, followed by the collection of trip detail that typically includes items such as origin and destination, time, purpose, and mode. Activity-based travel surveys, on the other hand, collect all activities undertaken by the respondent in the given time period. Trips are captured as just another activity. Much of the trip detail is not asked directly but is inherent to the activity diary structure and can be derived. Harvey's review of approximately 10 activity surveys showed that travel accounts for approximately 19% of reported activities.
Time-use surveys recently conducted in Europe found a lower proportion of persons who were "immobile," that is, not traveling on a given day, compared with a travel survey. For the French, 8% were immobile in the time-use survey compared with 17% in a travel survey (Armoogum et al. 2004). One hypothesis is that the reporting of "no trips" in a travel survey is that the answer is given as a "soft refusal." Now that the American Time Use Survey data are available, it is important for transportation analysts to do a similar comparison.
While some preliminary comparisons between trip-based and activity-based surveys have been performed, additional research is still needed. Important measurestrip rates and trip frequency distributionsshould be analyzed across survey types while controlling for survey conditions. While some research has shown that activity surveys produce better data quality, it is still unclear what tradeoffs there may be between quality and cost (Pendyala 2003).
Over the last several years, several passenger travel surveys have introduced multiple approaches for integrating GPS into travel surveys. Most commonly these have included vehicle-based passive surveys, person-based passive surveys, and vehicle-based interactive surveys. Original benefits were expected to include reduction of missing (unreported) trips, improved accuracy of travel distance and time, routing and speed data previously unobtainable, and ability to capture longer periods of travel.
Between 2001 and 2004, several regional household travel/activity surveys incorporated a GPS component as a subsample to their household dairy (Wolf 2004). The ability to capture unreported trips (item nonresponse) has ranged widely from 20% to 80%. Zmud and Wolf (2003) found that unreported trips were most likely to be trips of less than 10 minutes. Household characteristics leading to less complete reporting include having three or more vehicles, three or more workers, an annual income below $50,000, and persons younger than 25 years.
One of the most exciting passive GPS studies is the Commute Atlanta project (an FHWA value pricing pilot) with 365 days of 1-second GPS data for over 450 household vehicles. This long period of data collection allows for examination of the variability of travel and a better understanding of long-distance trips made by private vehicles. A "sunset clause," in which the data must be destroyed within six months of the end of the project, is one major handicap of this project.
How a GPS component could be incorporated into a national survey raises many questions, because completeness of responses in self-reported diaries, compared with GPS-recorded information, may be linked to demographic characteristics (e.g., English-language capability and education) and metropolitan characteristics (e.g., population size, density, and transportation network complexity).
Several tests have been conducted to trace personal movements using cellular/mobile phones. Some advantages of mobile phones relative to a GPS system are that they function underground and inside buildings more often, and the density of cellular base stations/towers is higher in the densest urban areas. Also, the market penetration of cellular phones is very high, so the cost of equipment is low.
In Germany, Wermuth et al. (2003) tested tracking of cellular phones for a long-distance survey. Recently, Kracht (2004), also in Germany, began testing cellular phone use for tracking daily personal travel, especially as many phones already have the capability of recording and storing the position (cell) over time. The positional accuracy afforded by cellular phones is not as good as using GPS, so while gross measurements of distance and travel time are achievable, specific routes or travel modes are less likely to be determined without greater respondent interface.
Using the internet as a response method is becoming more robust; however, it has more often been applied to shorter origin/destination travel surveys, for example, after license plate capture with a mail-out postcard and responses allowed either by mail-back or the internet. One test using only internet responses for household travel diaries, completed by the Resource Systems Group (2002), showed higher interest among older men with higher incomes and young men. In a regional test, the ability to incorporate a pre-geocoded electronic yellow pages was an advantage for selecting destinations. However, in the 2002 National Transportation Availability and Use Survey conducted by the Bureau of Transportation Statistics, only 3 out of over 5,000 respondents chose to complete the survey using the web option.
Obviously, any changes made to the design of the future NHTS surveys has an obvious impact on the ability to monitor travel behavior trends. Any change will need to be carefully weighed and tradeoffs made between needed improvements and the continued ability to track trends over time. Although minimal changes were made to the capture of daily travel, certain changes were introduced that obscured the ability to detect what was an actual change in travel behavior versus a change due solely to methodological or definitional differences. For example, additional probes were added to the 2001 NHTS to better capture more incidental types of trips thought to be underreported, such as walk and bike trips. As a result, the number of walking trips reported increased significantly from 1995 to 2001. Due to this change, it is not possible to discern what was a real change in walking behavior as compared with the improved capture of walking trips. As previously described, the substantial changes in the collection of long-distance travel data severely limited the ability for longer distance trend analysis.
Despite methodological hurdles, data from surveys on personal travel in the United States are a valuable commodity. Over time, the community of data users has grown and the application of these data in numerous studies has increased exponentially, especially as data accessibility has increased (ORNL 2004). Understanding who is traveling, how much they are traveling, and why they are traveling assists decisions on transportation investments and the potential implications of these investments on different communities.
Due to the disparate objectives and uses of the daily and long-distance travel components, current plans call for the next series of travel surveys to be separated again into two different data-collection efforts. Continued outreach and participation with the data user community to better understand data needs and anticipate emerging needs will be necessary for their successful design. Future surveys could greatly benefit from shared methodological improvements, including the following areas identified as priorities:
The next long-distance survey, in particular, will need careful consideration of its design. The disparate designs employed for the 1995 and 2001 surveys resulted in inconsistent definitions and trip characteristics, limited data utility, and diminished ability to monitor longer travel behavior. The next long-distance travel survey needs to be developed with a sustainable design, so that data users can come to expect a more useful and consistent product that allows for monitoring of long-distance travel behavior trends. In addition, the design needs to focus on two critical areas:
The NHTS/NPTS series, as a whole, has provided the daily travel data user community with a fairly consistent, useful product for nearly 30 years. However, further improvements are still needed to expand on its utility and better respond to changes in transportation, the population in general, and technology. Below are a few areas considered priorities in the next daily travel survey:
Armoogum, J., K. Axhausen, J.-P. Hubert, and J.L. Madre. 2004. Immobility and Mobility Seen Through Trip-Based vs. Time-Use Surveys, presented at the 7th International Conference on Travel Survey Methods, Playa Herradurra, Costa Rica, August 2004.
Bachu, P.K., T. Dudala, and S.M. Kothuri. 2001. Prompted Recall in Global Positioning System Survey, Proof of Concept Study. Transportation Research Record 1768: 106113.
Badoe, D.A. and G.N. Steuart. 2002. Impact of Interviewing by Proxy in Travel Survey Conducted by Telephone. Journal of Advanced Transportation6(1):4362, Winter.
Bose, J. and L. Giesbrecht. 2004. Impact of Proxy Reporting in the 2001 National Household Travel Survey, presented at the Joint Statistical Meetings, Toronto, Canada, August 2004.
Broeg, W. and E. Ampt. 1983. State of the Art in the Collection of Travel Behavior Data: Workshop on Data Needs and Collection. Transportation Research Board Special Report 201. Washington DC: Transportation Research Board. 4862
Contrino, H. and S. Liss. 2004. Sources and Impacts of Nonresponse in Household Travel Surveys: Three Case Studies, presented at the 7th International Conference on Travel Survey Methods, Playa Herradurra, Costa Rica, August.
Denver Regional Council of Governments (DRCOG) and Parsons Transportation Group. 2000. Travel Behavior Inventory: Decribing and Reaching Nonresponding Populations, report for the Federal Highway Administration, U.S. Department of Transportation. June.
Erhardt, G.D. 2000. Validation of Trip Production Rates for Synthesized Households, draft paper, Cornell University. January.
Freedman, M. and J. Machado. 2003. Pilot Study of Modified CATI Interview Technique, draft proposed for NCHRP Project 08-37, Standardized Procedures for Household Travel Surveys. July.
Golob, T.F. and H. Meurs. 1986. Biases in Response Over Time in a Seven-Day Travel Diary. Transportation13:163181.
Groves, R.M. 2004. Household Survey Nonresponse: Research Relevant to Statistical Policies, presented at the 94th Meeting of the Committee on National Statistics: Household Survey Nonresponse: What Do We Know? What Can We Do? Washington, DC, June 18, 2004.
Harvey, A.S. 2003. Time-Space Diaries: Merging Traditions. Transport Survey Quality and Innovation. Edited by P. Stopher and P. Jones. Amsterdam, The Netherlands: Pergamon Press.
Kish, L.F. 1965. Survey Sampling. New York, NY: John Wiley and Sons, Ltd.
Kracht, M. 2004. Tracking and Interviewing Individuals with GPS and GSM Technology on Mobile Electronic Devices, presented at the 7th International Conference on Travel Survey Methods, Playa Herradurra, Costa Rica, August 2004.
Murakami, E. 2003. Households without Vehicles, 2000. CTPP Status Report, January. Available at http://www.fhwa.dot. gov/ctpp/status.htm, as of October 2004.
Murakami, E. and W.T. Watterson. 1992. The Puget Sound Transportation Panel after Two Waves. Transportation 19:141158.
NuStats. 2003. BCD-COG Household Travel Behavior Survey: Final Report for the Berkeley-Charels-Dorchester Council of Governments. September.
______. 2004a. Goldsboro Urbanized Area Travel Survey: Final Report for the City of Goldsboro. January.
______. 2004b. Kansas City Regional Household Travel Survey: Final Report for the Mid-American Regional Council. June.
Oak Ridge National Laboratory (ORNL). 2004. Citations and Uses of NPTS/NHTS Data. Available at http://nhts.ornl.gov/2001/pub/NHTS_Citations.pdf, as of Sept. 27, 2005.
Pendyala, R. 2003. Quality and Innovation in Time Use and Activity Surveys. Transport Survey Quality and Innovation. Edited by P.R. Stopher and P. Jones. Amsterdam, The Netherlands: Pergamon Press.
Pisarski, A. 1996. Commuting in America II. Lansdowne, VA: Eno Transportation Foundation.
Resource Systems Group. 2002. SBIR Phase II Final Report: Computer-Based Intelligent Travel Survey System, Report for the Federal Highway Administration. October. Available at http://www.fhwa.dot.gov/ohim/trb/sbir/sbir.htm, as of Oct. 18, 2004.
Richardson, A.J. 2003. Behavioral Mechanisms of Nonresponse in Mail-Back Travel Surveys. Transportation Research Record 1855:191199.
Stopher, P., A. Collins, and P. Bullock. 2004. GPS Surveys and the Internet, presented at the 27th Australasian Transport Research Forum, Adelaide, Australia, Sept. 29Oct. 1, 2004.
Transportation Research Board (TRB). 2003. Measuring Personal Travel and Goods Movement: A Review of the Bureau of Transportation Statistics' Surveys, Special Report 277. Washington, DC: National Research Council.
Tourangeau, R., M. Zimowski, and R. Ghadialy. 1997. Introduction to Panel Surveys in Transportation Studies, Travel Model Improvement Program, DOT-T-98-3. Washington, DC: U.S. Department of Transportation, Federal Highway Administration.
Tuckel, P. and H. O'Neill. 2004. Ownership and Usage Patterns of Cell Phones: 20002004, presented at the Annual Meeting of American Association for Public Opinion Research, Phoenix, Arizona, May 1316, 2003.
U.S. Department of Commerce (USDOC), U.S. Census Bureau. 2002. Meeting 21st Century Demographic Data Needs: Implementing the American Community SurveyReport 2: Demonstrating Survey Quality. Washington, DC. May. Available at http://www.census.gov/acs/www/AdvMeth/Reports. htm, as of August 2004.
______. 2004. Meeting 21st Century Demographic Data Needs: Implementing the American Community SurveyReport 6: The 20012002 Operational Feasibility Report of the American Community Survey. Washington, DC. 2004. Available at http://www.census.gov/acs/www/AdvMeth/Reports.htm, as of August 2004.
U.S. Department of Transportation (USDOT), Federal Highway Administration. 2001. Issues in the 2001 Address vs. List-Assisted Design, unpublished document.
______. 2004. National Household Travel Survey User's Guide, January 2004 (Version 3) (National Sample with Add-Ons), p. 4-3. Available at http://nhts.ornl.gov/2001/usersguide/index. shtml, as of December 2005.
Wermuth, M., C. Sommer, and M. Kreitz. 2003. Impact of New Technologies in Travel Surveys. Transport Survey Quality and Innovation. Edited by P. Stopher and P. Jones. Amsterdam, The Netherlands: Pergamon Press.
Wolf, J. 2004. Application of New Technologies in Travel Surveys, presented at the 7th International Conference on Travel Survey Methods, Playa Herradurra, Costa Rica, August 2004.
Zmud, J. and J. Wolf. 2003. Identifying the Correlates of Trip Misreporting: Results from the California Statewide Household Travel Survey GPS Study, presented at the International Conference on Travel Behaviour Research, Lucerne, Switzerland, 2003.
The following is a list of improvements and changes in the 2001 NHTS as compared with the 1995 NPTS:
In addition, a number of questions were added to (or data elements later derived from) the 2001 NHTS to cover emerging trends pertinent to personal travel behavior.
At the household level:
At the person level:
At the individual daily trip level:
At the individual long-distance trip level:
1. Five states (Hawaii, Kentucky, New York, Texas, and Wisconsin) and four metropolitan areas (Baltimore, MD; Des Moines, IA; Lancaster, PA; and Oahu, HI) purchased an additional sample for their areas through the NHTS "add-on" program.
2. The Paperwork Reduction Act of 1995 requires federal agency requests submitted to OMB, ". . . to use effective and efficient statistical methodology appropriate to the purpose for which information is to be collected and directs OMB to develop and oversee the implementation of government-wide policies, principles, standards, and guidelines concerning statistical collection procedures and methods."
6. Modest changes were implemented between data collections to improve overall data quality. These revisions made analyzing trends across years more difficult. See appendix A for a more detailed description of changes.
9. Reliable aggregated estimates can be made for the nine census regions (and divisions based on metropolitan area size and the presence of rail). An exception to limited geographical analysis is the selected add-on areas; nine states and metropolitan areas purchased additional samples for their areas to obtain a more reliable estimation.
Corresponding author: J. Sharp, Bureau of Transportation Statistics, Research and Innovative Technology Administration, U.S. Department of Transportation, 400 Seventh St, SW, Room 4432, Washington, DC 20590. E-mail: email@example.com
E. Murakami, Federal Highway Administration, U.S. Department of Transportation, c/o FTA, 915 Second Avenue, Room 3142, Seattle, WA 98174. E-mail: firstname.lastname@example.org