Improving Household Travel Survey Quality Through Time and

Improving Household Travel Survey Quality Through
Time and Distance Data Checking
Presentation Summary
Laurie Wargelin, Vice President
MORPACE, International 31700 Middlebelt Road
Farmington Hills, MI 48334
Phone: 248 737 3210
Fax: 248 737 5326
Email: [email protected]
Karen Faussett, Project Manager
Michigan Department of Transportation
P.O. Box 30500
Lansing, MI 48901
Phone: 517 335 2956
Fax: 517 373 9255
Email: [email protected]
Lidia Kostyniuk, Research Scientist
University of Michigan Transportation Research Institute
2901 Baxter Road
Ann Arbor, MI 48109
Phone: 734 763 2466
Fax: 734 936 1076
Email: [email protected]
This presentation addresses the problem of the accuracy of travel time
and travel distance collected in current household travel surveys, including the
2001 NHTS, and proposes a method that could help alleviate the problem in the
next NHTS survey. The problems with the accuracy of travel time and travel
distance data in household travel surveys stem from several sources including
respondent errors, geocoding errors, and interviewer errors. Respondents in
travel surveys frequently err when reporting both the distance between locations
traveled and the time that the travel took. Studies that have compared at selfreported travel distance against actual distance traveled, find that drivers in
urban areas are not very good at reporting accurate distances of their trips
(e.g., 1). Obtaining travel distance from geocoded trip’s origins and destinations
helps to decrease the reliance on self-reported trip distances. However, even
1
though the proportion of locations that can be geocoded has increased, and the
accuracy of geocoding has improved through Internet look-ups and interactive
geocoding software, considerable error still occurs. For, example, if the same
street name appears in two cities within a targeted area and the respondent
supplied a different city/village name, there is likelihood that the geocoded point
selected will be incorrect.
Interviewer error also contributes to the problem. Training and experience
in conducting household travel surveys help reduce error in recording time and
location information. Up-to-date Computer Assisted Telephone Interviewing
(CATI) or Internet systems can be programmed with checks to assure that time is
sequentially recorded. However, neither the CATI nor the Internet system can
prevent an interviewer from recording an incorrect arrival time at a destination.
The error will be identified by the CATI or Internet system checks only if the
respondent reports leaving a location at a time prior to the recorded arrival time.
Otherwise the error will remain in the data file.
Considerable time and effort are spent trying to correct or eliminate cases
with these time and distance problems, which significantly increases the time that
passes before the data can be used. To date, the most prevalent suggestion for
improving the quality of travel time and distance in household surveys has been
to use GPS in respondents’ vehicles as a supplement or replacement to
CATI/Internet household travel diary data collection. The expectation is that GPS
systems will accurately record both the time and distance between locations.
However, while GPS technologies are improving in their ability to meet the task,
the cost of deploying GPS in a household travel survey is currently above $500/
household. This high cost has prohibited the use of GPS for all households in
the travel survey sample and has limited GPS data collection to small
subsamples. Furthermore, at this time there are no standards or agreement on
how to expand the GPS data from the subsample to the main or stratified
2
sample, and use of correction factors derived from the subsample GPS data has
had only limited applications.
It is most likely inevitable that GPS systems will keep improving, become
less expensive, and at some point in the future be part of standard practice in
household surveys. Until then, however, there is a still a need for a cost effective
method to minimize the errors in travel time and distance variables in data from
household travel surveys.
A method was developed to improve the quality of travel time and distance
data in the Michigan Department of Transportation’s (MDOT) MI Travel Counts
household travel survey that is currently underway1.
The approach is based on
checking time and distance data using the capabilities of TransCAD software.
Time and distance outliers are identified while the survey is in progress so that
respondents can be quickly called back to clarify suspect information. As
questionable information is identified, the entire household’s travel records are
reviewed for logic, and are corrected or eliminated, (labeled incomplete) allowing
recruitment of replacement households on a continuing basis.
The method was tested during the pilot of MI Travel Counts, conducted in
January and February 2004 and with interim samples of 2,000 households in the
spring and summer of 2004. First, all non-vehicular trips were removed from the
file as well as all trips that were non-geocoded. TransCAD automatically
converted the corresponding geocoded coordinates for each record into a
standard geographic file with points located onto the network file. Average
private vehicle travel speed was then calculated for each trip using TransCAD
shortest route distance and respondent travel time. Checks were conducted on
calculated distances, calculated travel time, and respondent travel time: Average
1
MI Travel Counts is collecting 48-hour travel (using diary) from a statewide sample of 14,280
households, with retrospective long distance (over 100 mile) travel collection for the prior 3 months. Data
collection began in March 2004 and will continue through March 2005. See MI Travel Counts website at
www.michigan.gov/mitravelcounts
3
private vehicle travel speed was calculated for each trip using TransCAD shortest
route distance and respondent travel time. A private vehicle trip was flagged if
the average travel speed was less than 5 miles per hour (mph). If a private
vehicle trip was greater than 30 miles in length, the trip was flagged if the
average travel speed was greater than 80 mph; otherwise it was flagged if the
average speed was greater than 65 mph. Private vehicle trips that were flagged
then have 10 minutes added and subtracted from the trip time and a new speeds
were computed. If at least one of the new speeds still was not within the speed
parameters, the record was flagged and reviewed.
The example provided here is for the July interim MI Travel Counts
Report.2 There were 1,979 new households in this file with 33,230 new trip
records (without including change in mode locations). The following records
were disregarded/removed due to the following reasons (in this order):
•
•
•
•
•
373 records did not make a trip
864 records were removed due to type of trip (walk)
204 records were removed due to type of trip (bicycle/moped)
3,600 records had either the origin or destination not geocoded
217 records had the origin and destination geocoded as the
same point (turn-around trips).
Thus, of the original 33,230 records, 5,258 were removed, leaving 27,972
trips, which is approximately 84%. A total of 1,502 trip records (4.5% of total)
were flagged for the following reasons:
•
•
•
same city/township with trip over 60/90 minutes: 63 records
travel time difference over 60 minutes: 976 records
average travel speed above or below threshold: 1,385 records
The 1,502 identified trip records were further examined. It was found that
909 trip records (2.7% of total) were geocoded incorrectly. For example, an
origin/destination would be listed as one city but geocoded to a different city.
The remaining 593 records (1.8% of total) needed further investigation for
4
discrepancies in travel time. Of these (593 trip records), it was found through
further review and callbacks to respondents that the information on 124 trips was
correct based on reasons given for extended travel time, such as traffic
congestion or weather conditions; 305 needed to have their location or
departure/arrive time checked; and 164 were still questionable. Of the 164 that
were still questionable, 140 had a trip time that was too long for the distance of
their trip and 24 had a trip time that was too short for the distance of their trip.
Thus overall, 469 trips or 1.4% of trips in this file need either further investigation
or reconsideration of their inclusion in the final dataset. While this proportion of
trips ultimately found to be questionable is small, the 1,502 records originally
identified, investigated, and mostly corrected through this process, represent 453
households or 22.9% of the new households in this interim data file.
This method is straightforward and simple. By monitoring the quality of
the data as it is coming in, there are more opportunities to identify problem data,
and either correct them or eliminate them and still be able to recruit replacement.
Most states have TransCAD of similar network models and can easily apply this
audit methodology to monitor the quality of their household travel data. The full
methodology can be found in the MI Travel Counts quality control (2) and
geocoding manuals (3). Monitoring the quality of the travel time and travel
distance data during the implementation of the survey is efficient. More
resources would are required to find the outliers at the end of the survey, and
opportunities to recruit replacement households would be lost.
2
7/24/04: As subconsultant to MORPACE International, Inc., PB Consult provided the TransCAD audit
review.
5
References:
1. Kostyniuk, L. P, Eby, DW, Kristoff, C., Hopp, M.L., Stress, F. M., 1997,
The FAST-TRAC Natural Use Leased-Car Study: An Evaluation of User
Perceptions and Behaviors of Al-Scout by Age and Gender, Report
UMTRI-97-09, University of Michigan Transportation Institute, Ann Arbor,
Michigan.
2. MDOT, MORPACE International, Inc., MI Travel Counts Quality Control
Manual, 2004.
3. MDOT, MORPACE International, Inc., MI Travel Counts Geocoding
Manual, 2004.
6