|
|
|
|
Measuring
Day-to-Day Variability in
Travel Behavior Using GPS Data
CONCLUSIONS AND FUTURE RESEARCH
Summary
and Implications
In this study, day-to-day variability in travel behavior is measured using GPS
data collected from a sample of 100 households in the Lexington, Kentucky area.
One vehicle in each of the 100 households was fitted with a GPS device and a
personal digital assistant (PDA) device in order to track and record movements
of the vehicle over a period of several days. Most households provided at least
three days of complete and usable information, thus providing a valuable data
set for measuring day-to-day variability and for examining the potential of
adopting GPS technologies in larger scale travel data collection efforts. The
analysis in this study focused on trips reported by primary drivers and did
not include secondary and tertiary drivers who may have also used the GPS-fitted
vehicle for trip making purposes.
The first
analysis performed for this study focused on comparing aggregate sample-wide
averages across the days for which travel information was reported. The comparison
was done across seven days without regard for the day of the week, across five
weekdays, and across the two weekend days. In general, it was found that sample-wide
averages show considerable stability across multiple days. However, when one
controls for day of week effects, some differences in overall trip frequency
are observed towards the end of the week (Thursday and Friday). This finding
is consistent with that found by Zhou and Golledge [2000]. As expected, some
differences in trip making patterns were found between weekdays and weekend
days with regard to trip frequencies and first time of departure from home.
Interestingly,
travel distances and total travel time were quite similar between weekdays and
weekend days. In general, it was found that the GPS data set yielded trip frequencies
slightly higher than the typical average frequencies reported in NPTS and other
traditional travel surveys. This is presumably because of the selective nature
of the sample which includes people who drive at least three days per week and
are 18 years or older. In addition, the GPS travel data collection method is
better able to capture short and infrequent trips that occur within trip chains
[Yalamanchili, et. al., 1999].
As tracking
aggregate sample-wide averages from one day to the next may mask disaggregate
within-person variability that is present in a data set, further analysis was
performed. First, an analysis was done to assess the degree of repetition of
behavior in the sample. The sample was divided into groups based on the total
number of days for which they provided complete usable travel information. Then,
the percents of each sample group that repeated behavior on all reported days,
all but one reported days, and all but two reported days were determined. In
general, it was found that the percentage of individuals who repeated their
behavior on all days is extremely small regardless of the travel characteristic
being examined. If one were to control for day-of-week, then the degree of repetition
of behavior is found to increase, but only modestly. Consistent with research
reported in the literature, this analysis showed that there is considerable
variability in behavior across multiple days.
In an effort
to quantify day-to-day variability in travel behavior and measure the extent
of intra-personal (within-person) variability, the methodology adopted by Pas
[1987] and Pas and Sundar [1995] was applied to the Lexington GPS data set.
In this approach, the total variability in the data set consists of two components
- inter-personal variability and intra-personal variability. By representing
total variability by the total sum of squares, inter-personal variability by
the between-person sum of squares, and intra-personal variability by within-person
sum of squares, the percent of total variability that may be attributed to intra-personal
variability may be determined. This analysis was done for several different
travel characteristics including trip frequencies, travel times, travel distances,
and selected departure/arrival times.
The analysis
showed that the extent of intra-personal variability measured in a data set
is sensitive to two dimensions. First, it is sensitive to the day of the week
effect. If one controls for the day of the week when analyzing variability across
multiple days, then the degree of intra-personal variability measured is lower.
Second, it is sensitive to the period or duration for which travel is observed.
The longer the period of observation, the greater the degree of intra-personal
variability. Both of these findings are consistent with the literature, thus
showing that GPS-based travel data sets are capable of offering plausible and
reasonable inferences on travel characteristics.
In comparing
the estimates of intra-personal day-to-day variability in travel offered by
the GPS-based data set with those reported in the literature by Pas [1987] and
Pas and Sundar [1995], it is found that the estimates obtained from the GPS
data set in this study are consistently greater than those reported in the literature
by nearly 10 percentage points. Further research is needed to clearly identify
the sources of this difference; however, several hypotheses may be put forth
here. First, there may be socio-economic differences between the samples under
consideration that may be partially contributing to this difference. The GPS-based
data set used in this study is a small sample data set, includes only those
trips undertaken by the primary driver using the GPS-fitted vehicle, and includes
a self-selected sample of households that has larger household sizes, vehicle
ownership, income, and education levels. The extent to which socio-economic
differences and sample composition contribute to differences in measures of
intra-personal variability merit further investigation.
A second major
hypothesis, and more noteworthy in the context of the growing interest in GPS-based
data collection methods, is that GPS technologies are able to better capture
variability in travel behavior across multiple days. There are two potential
inter-related reasons for this. First, as demonstrated by Yalamanchili, et.
al. [1999], the GPS-based data set captures (to a greater extent) the short
and infrequent trips that may not be obtained in a traditional travel diary
survey. As these short and infrequent trips are quite different from one day
to the next, they contribute to a greater degree of intra-personal variability
than if they were omitted from the analysis. Second, in a multiday travel diary
survey (especially those exceeding two days), there is likely to be diary fatigue
[Pendyala and Pas, 2000]. Even if a respondent provides very detailed information
(including that on short and infrequent trips) on one or two days, it is likely
that the quality of the information will deteriorate on subsequent days. Usually
diary fatigue manifests itself in the form of unreported trips (usually the
short and infrequent trips go unreported) and missing data items. Then, it is
likely that a multiday GPS travel data set that includes detailed information
on all trips across all days will provide a greater (and potentially more accurate)
measure of intra-personal variability than a multiday travel diary data set
where some infrequent trips go unreported in latter days of the survey.
Overall, it
appears that GPS based travel data sets are able to:
More importantly, GPS-based
data collection methods accomplish the above without placing undue or additional
burden on the respondent. Battelle [1997] reports that it took 75 percent of
the respondents less than one minute to enter the trip information into the
PDA device. This is in stark contrast to the 10 minute reporting durations for
paper diary surveys and 20-25 minute durations over the phone. Also, the presence
of the GPS receiver greatly reduces the number of questions or attributes that
are required to be entered manually by the respondent. For example, in the GPS
based survey, start time, end time, origin and destination locations, distance,
travel time, and route information are automatically logged. Given these potential
advantages, it is no surprise that GPS-based data collection methods are gaining
increasing popularity for use in travel surveys. Several major ongoing travel
surveys include GPS-based data collection components.
Within the scope of this
study, the full potential of GPS based travel data sets was not exploited. Detailed
route choice, spatial location, and travel itinerary data present in the data
set was not used in the analysis of this study. Further research should examine
day-to-day variability in route choice, spatial location, and action space;
such an analysis would be unique as it is very difficult to perform such analyses
using traditional travel diary survey data sets.
United States Department of Transportation - Federal Highway Administration