|
|
|
|
Measuring
Day-to-Day Variability in
Travel Behavior Using GPS Data
IMPORTANCE
OF MULTIDAY TRAVEL SURVEYS
Introduction
Over the past several decades, increasing attention has been paid to the collection and use of multiday travel survey data for measuring day-to-day variability in travel behavior. However, a review of the literature and planning practice indicates that the predominant procedure for the conduct of household travel surveys has been the traditional "one-day travel diary" approach.
In the 1940s and 1950s, transportation planning focused on the identification and relief of congestion on highway networks and usually used travel demand models to estimate the additional number of lanes required to improve operational level of service on the roadway. The accuracy and precision levels required to answer to such capacity-expansion planning questions were modest, as were the data requirements of the zonal-based models in use at the time. In that context, data from one-day surveys was generally considered sufficient.
Since then, the emphasis of planning has shifted from mere capacity expansion to the development of strategies and implementation of systems that would more effectively manage travel demand. It was recognized that greater accuracy and precision levels would be required of travel demand models to address such planning questions. Early examples of multiday surveys include the 1971 household travel survey in Uppsala, Sweden that collected travel information for a 35-day period [Hanson and Huff, 1982] and the 1973 seven day activity survey conducted in Reading, England [Pas, 1986]. Since then, many multiday travel surveys have been conducted around the world with a view to capture the variability in travel across different days.
The next few sections describe the primary considerations that warrant the use of multiday travel survey data in transportation planning.
Policy Considerations
Why is the measurement and modeling of day-to-day variability of travel behavior important from a policy perspective? An excellent discussion in response to this question is provided by Jones and Clarke [1988]. They note that as the emphasis of transportation planning shifted from capacity expansion to travel demand management, some of the issues facing transportation planners could not be addressed by one-day data (regardless of the sample size) because "by their nature, they are questions about variations in behavior over time". They provide several key examples to illustrate their point.
One example pertains to an electronic road pricing feasibility study conducted in Hong Kong in the early 1980s. A policy question facing planners was "how many car owners would be affected by a road pricing scheme, if charges were imposed only on Monday through Friday?". A one-day travel survey was conducted in 1981 using a trip recall survey methodology. Analysis of the data revealed that around 40 percent of the household vehicles were not used on an average weekday. But the data did not provide information that could answer the policy question facing planners, namely, the percent of cars that would not incur road charges over a typical week. While the one-day data provided a lower limit (0 percent) and an upper limit (40 percent), it could not provide more precise information.
On the other hand, Jones and Clarke [1988] refer to a study of car use conducted in Oxford that collected multiday data and was able to address such issues. In that sample, it was found that 18 percent of the cars were left unused at home on a typical weekday, with 65 percent of cars being used every weekday and only 5 percent not being used at all over a week. If the Hong Kong road pricing scenario were applied in this context, it would be apparent that only 5 percent of the vehicles would be unaffected by the road charge in a typical week.
In other words, multiday data provides information about the distribution of the frequency of participation, in addition to the mean participation rate. Jones and Clarke [1988] indicate that such information on the frequency of participation allows the planner to gauge the exposure of different demographic and travel segments to various policy scenarios. In essence, the measurement of day-to-day variability in travel behavior has important policy implications in the current planning context.
Efficiency Considerations
Even when policy considerations do not necessarily require the use of multiday travel survey data, efficiency considerations may call for the collection and use of such data. Pas [1986] shows that substantial economies can be achieved with respect to survey cost and parameter efficiency if a multiday sample, as opposed to a one-day sample, is employed in travel demand analysis.
Koppelman and Pas [1984] report on an examination of linear regression models of trip generation estimated using multiday data. They find that the estimators of multiday models are more efficient than those of the single day model. Using five-day data from the 1973 Reading, England survey, they find that the standard errors of parameter estimates are reduced by 31 percent through the use of a five day sample rather than a single day sample. The corresponding reductions are 18, 25, and 29 percent respectively for two, three, and four day samples. They note that the diminishing benefit of additional observational days suggests that an optimal number of survey days may exist depending on the extent of intrapersonal variability in the phenomenon under study.
In subsequent research, Pas [1986] further extends this work to actually compute the gains in efficiency for different reporting periods under a wide variety of cost assumptions. In particular, he shows that it is possible to obtain more precise estimates of parameters in linear regression models of trip generation by taking advantage of the intrapersonal variability captured in multiday data (within a fixed data collection budget). Alternatively, he shows how precision levels can be maintained at a desired level while decreasing the overall data collection costs when multiday samples are used for model estimation.
In the Reading, England data, assuming that the variable cost of data collection is 25 percent of the fixed cost, he found that three-day data from a 75 person sample can yield the same level of precision as a single day sample of 136 individuals at a cost saving of approximately 23 percent. On the other hand, a three-day sample of 98 individuals would cost approximately the same as a 136 person single day sample, but would yield parameter estimates that are approximately 23 percent more efficient than the single-day sample. Alternatively, a two-day sample of 114 individuals can provide a 20 percent gain in parameter efficiency at the same cost, or a two-day sample of 91 individuals can provide the same parameter efficiency (as a 136 person single day sample) at a 20 percent cost savings. Thus, he finds that a two-day sample is almost as beneficial as a three-day sample and is possibly the preferred choice as the level of respondent fatigue is likely to be lower in a two-day sample than in a three-day sample.
It should be noted that the figures noted above were obtained in the case of the Reading data set. Pas [1986] notes that, as the proportion of intrapersonal variance in a data set increases (i.e., as more of the total variance can be attributed to within-person variability), the efficiency gains at a fixed budget or the cost savings at a fixed precision level increase. In other words, if the intrapersonal variability is zero, then single day data provide the same amount of information as multiday data (additional survey days provide no new information).
The discussion in this section clearly shows that the collection of multiday survey data is important and offers substantial benefits from several perspectives. It is in this context that this research project focuses on measuring day-to-day variability in travel behavior.
