Office of Planning, Environment, & Realty (HEP)
Over the past few decades, several hundred travel surveys have been conducted within the United States, mostly by regional transit agencies and metropolitan planning organizations . The data from these surveys have been used for such diverse purposes as measuring the impact of changes in the transportation system on travel behavior, forecasting future travel patterns and demand, and developing marketing campaigns to promote transit use. Nearly all the surveys have relied on cross-sectional designs that measure variation in travel behavior among the members of a population.
The purpose of this report is to discuss a different kind of survey design that measures variation in travel behavior at the level of the individual household or person by taking repeated measurements on the same sample of units at different points in time. These designs, referred to as panel or longitudinal designs in the survey literature, provide direct information on how the travel behavior of individual households or persons changes over time in response to other factors.
Although panel designs have enjoyed widespread use in transportation studies in other countries, and in work in other fields, they have rarely been adopted in travel surveys in the United States. This report shows how they can be used to address a variety of transportation policy and planning issues, ranging from impact assessments of specific policy changes on travel behavior to the more general issues of predicting and planning for future trends in behavior.
The report begins by describing the differences between panel and cross-sectional approaches to the study of travel behavior. It then discusses the advantages and limitations of these approaches to data collection, identifies situations where panel data are desirable, and illustrates their benefits through examples drawn from the transportation literature. The final section of the report provides guidelines for the conduct of panel surveys, focusing on the special issues and difficulties that arise when the same sample of households or individuals is measured repeatedly over time.
All surveys can be classified into one of two broad categories on the basis of whether they obtain repeated measurements on the same sample of units over time. Panel surveys do and cross-sectional surveys do not.
Within these two approaches to data collection, surveys may be further distinguished according to whether they monitor changes in the population over time. Cross-sectional and panel surveys that incorporate this feature periodically draw new samples from the population and collect measurements on them using the same methods as in previous time periods.
The differences among these four approaches to the collection of survey data are summarized in Table 1. The table distinguishes between two types of cross-sectional designs- one-time cross-sectional designs, and repeated cross-sectional designs- and two types of panel designs- longitudinal panel designs, and rotating or revolving panel designs. It shows how the designs differ along four dimensions:
Differences Among the Features of Four Types of Survey Designs
One-time cross-sectional surveys. In the United States, most travel surveys rely on one-time cross-sectional designs to collect information on travel consumption and behavior. In these designs, a single sample of households or individuals, usually a cross section of the regional or national population, is asked to complete a survey at a single point in time. In other words, a single set of measurements is collected from each sample member. In practice, the time at which the measurements are actually taken varies somewhat across sample members. Nevertheless, the measurements are close enough in time to be regarded as contemporaneous, as occurring at the same point or period in time.
One-time cross-sectional designs capture the travel behavior of the population as it exists at the time of the survey. They provide a "snapshot" of travel behavior in a region by obtaining snapshots of the behavior of the individual sample members. Surveys of this type measure cross-sectional variation in travel behavior, that is, variation among the members of a population. They show how behavior differs from member to member, but they provide no direct information on how it changes over time.
A distinguishing feature of one-time cross-sectional surveys is that they make no attempt to replicate conditions of previous surveys. They may measure a similar set of variables, but the actual questions posed to the respondents may differ in wording or in meaning, and the sampling and field procedures may not be the same as in previous surveys. For this reason, one-time cross-sectional surveys conducted at different points in time are not well suited for assessing trends in population behavior since their results cannot be readily compared with one another.
Repeated cross-sectional surveys. Repeated cross-sectional designs, on the other hand, measure the travel behavior or attitudes of the population over time by repeating the same survey on two or more occasions. During each time period, a separate but comparable sample of units is drawn from the population and asked to complete the survey. Each sample member completes the survey once, unless they are selected by chance into more than one sample.
Because the field procedures, survey instruments, and samples are comparable from period to period, designs of this type allow for comparisons among and between measurement periods. They are ideally suited for assessing period trends in behavior at the population or other aggregate levels, and are often used to monitor changes in the population as a whole or in various subgroups within the population, such as those defined by demographic background characteristics. However, they provide no direct information on change at the level of the individual sample member since each measurement period relies on a distinct sample of households or individuals. Like one-time cross-sectional surveys, they measure cross-sectional variation in travel behavior, but at two or more periods in time instead of at one.
Repeated cross-sectional designs are often referred to as longitudinal designs in the survey literature because they measure variation in the population over time. Following the convention adopted in the Travel Survey Manual, we reserve the term longitudinal to refer to designs that collect measurements on the same sample of units at different times. Such designs are discussed below. 1
Longitudinal panel surveys. Longitudinal panel designs differ from cross-sectional surveys in that they collect information on the same set of variables from the same sample members at two or more points in time. For a household travel survey, this means that the same sample of households is asked to answer questions about their travel behavior and other variables on two or more occasions. Each distinct occasion when data are collected from the sample members is referred to as a wave or round of data collection. In a two-wave panel survey, sample members are asked to provide data twice, once during each wave. In a three-wave panel survey, panel members are asked to provide data three times, and so on. Within each wave the measurements are close enough in time to be considered contemporaneous. Typically, each wave collects some of the same items of information and some new items as well.
Although there is no upper limit on the number of waves a panel survey may contain, in practice most panel surveys consist of between 2 and 10 waves. This feature is often used to distinguish them from time series, which collect a series of measurements over a relatively large number of time points. Time series differ from panel surveys in two other important respects: 1) they collect data on a single entity, such as a person or a nation, while panel surveys obtain measurements on a collection of units, usually individuals or households, and 2) the time point rather than the individual sample member is the unit of analysis.
Longitudinal panel surveys are similar to cross-sectional surveys in that they measure cross-sectional variation in travel behavior by collecting information on a sample of units. What sets them apart from cross-sectional surveys is that they also measure longitudinal variation in travel behavior- that is, variation over time at the level of the individual sample member- by repeating the survey on the same sample of units at two or more points in time. In other words, they provide information on how the travel behavior of individual sample members changes over time in response to changes in the travel environment, household background characteristics, or other factors.
Longitudinal panel surveys are similar to repeated cross-sectional designs in that they permit comparisons across time by asking the same questions under comparable conditions. But, unlike repeated cross-sectional designs, they ask the questions of the same sample members and thus provide for direct measurement of individual change.
During the first wave of data collection, longitudinal panel surveys provide the same information as one-time cross-sectional designs. They assess current population levels and measure cross-sectional variation in travel behavior. During the second and subsequent waves, longitudinal panel surveys also measure cross-sectional variation, but they may not measure current population levels since the composition of the current population may no longer be the same as it was in the first wave when the sample was drawn. However, if the time span of the survey is relatively short and the panel sample is periodically refreshed, chances are high that data obtained in each wave will reflect current population levels.
Revolving or rotating panel designs. Rotating or revolving panel surveys are a combination of repeated cross-sectional and panel designs. They collect panel data on the same sample of units for some specified number of measurement periods. Portions of the sample are then gradually dropped from the panel and replaced with new but comparable samples drawn from the current population. The process of retiring portions of the existing sample and adding new members to the sample continues until the original panel is completely replaced. The new sample members are retained in the survey for some specified number of measurement periods and then gradually replaced with a comparable but more current sample and so on. The survey may continue indefinitely or be limited to a certain number of replacement samples. Each sample of units selected at the same time and adhering to the same schedule of data collection is called a rotation group.
The strength of rotating panel designs lies in their ability to allow for short-term analysis of individual or household change and long-term analysis of population and subgroup change. As in panel surveys consisting of a single sample of the population, rotating panel designs provide direct information on change at the level of the individual household or person over the period in which the sample member is retained in the survey. As in repeated cross-sectional designs, they provide information on how travel behavior changes over time at the population or other aggregate levels by periodically drawing comparable samples from the current population and obtaining similar measurements on them.
Other variations. Although most surveys fall into one of these four categories, there are many variations within each category not discussed here. For example, it is possible to have a rotating panel design in which portions of a sample are retired from the survey for some specified number of time periods and then returned to the survey for additional measurement periods.