Two ways to measure how far vehicles are driven are: to ask knowledgeable respondents to estimate the miles traveled over a period of time; and to ask the respondents to report at least two odometer readings. The findings described here compare odometer readings with self-reported estimates of Vehicle Miles Traveled (VMT) to investigate to what extent self-reported VMT is a reliable surrogate for odometer-based VMT. Results from two national surveys, the Energy Information Administration's 1994 Residential Transportation Energy Consumption Survey (RTECS) and the Federal Highway Administration's (FHWA) 1995 Nationwide Personal Transportation Survey (NPTS), show that self-reported VMT values were at least 4 percent greater than odometer-based VMT (see Figure 1).
Sources: Energy Information Administration, 1994 Residential Transportation Energy Consumption Survey and the U.S. Department of Transportation, Federal Highway Administration, 1995 Nationwide Personal Transportation Survey.
Asking the respondent to estimate the miles traveled over a reference period is less complicated and less expensive than computing it from odometer readings. Collecting odometer-based VMT requires contacting the respondent at least twice to obtain beginning and ending odometer readings. Also, the respondent must have the vehicle at the time of the contact, or have recorded the odometer reading at an earlier time, which presupposes additional prior contact. These findings should help researchers determine which method to use for collecting VMT.
Between 1979 and 1994 the RTECS was conducted seven times. Due to budget reduction, the RTECS was discontinued after the 1994 cycle. Each RTECS covered the population of household vehicles based on a nationally representative sample of about 3,000 households and over 5,000 vehicles.
The RTECS was an adjunct survey to the Residential Energy Consumption Survey (RECS), a household energy survey conducted normally in the last part of a calendar year via personal interview. In addition to household information, energy-related information, basic demographic information and characteristics of the household's vehicles are collected. A subset of the RECS households was selected for RTECS and odometer readings were collected for all vehicles in these households near the beginning of the calendar year during the RECS interview. Approximately 6 months later, information about any vehicles that had been acquired or disposed was collected during a telephone interview. At the end of the year a postcard was mailed to each respondent to record final odometer readings for all vehicles disposed during the second half of the year or still in the possession of the household at the end of the year.
In addition, RECS respondents reported the number of miles each vehicle in their possession at the time of the RECS interview had been driven the previous 12 months. This respondent-reported VMT is referred to as self-reported VMT.
RTECS estimates of self-reported VMT are based on subjective data. Respondents were asked, "How many miles had each vehicle been driven in the past 12 months?" This question was asked during the RECS interviews conducted in late 1990 and 1993. Thus, the respondent estimates covered mostly 1990 and 1993, plus parts of 1989 and 1992, for vehicles in the household at the time of the RECS interview.
Odometer-Based VMT 1
When possible, VMT were determined for a sample vehicle by taking the difference between two odometer readings, which spanned a period of time. This method was used in 1994 to determine VMT for about half of the RTECS 5,553 sampled vehicles. Attempts were made to obtain odometer readings during the RECS interviews; the RTECS Mid-Year(M-Y), and the End-of-Year (E-O-Y) interviews. A "span" of odometer readings was the difference between two odometer readings. In most cases, this span was a Beginning Of Year (B-O-Y) to E-O-Y span; although, due to an occasional nonresponse, sometimes only shorter spans were available, such as RECS to RTECS M-Y. Odometer spans of less than a full year were also obtained for vehicles that were either acquired or disposed of during the survey year.
The VMT that were assigned to each RTECS vehicle corresponded to the period of time during 1994 that the vehicle was in possession by the sample household. In most cases, however, this period of possession did not correspond exactly with the beginning and ending dates for the odometer span. This was true even for vehicles with a complete B-O-Y to E-O-Y odometer span; because odometer cards were mailed to respondents in several distinct waves at the beginning and end of the survey; and because the exact dates of odometer readings were left to the convenience of the respondents. Therefore, all VMT obtained from odometer spans were adjusted to correspond to the period of time that the vehicle was in possession by the sample household. A 2-step adjustment procedure was used. Step 1 adjusted the odometer-span VMT to a standard annualized mileage covering 365 days; Step 2 readjusted the annualized VMT to correspond to the exact period of time that the vehicle was in possession by the household. These adjustments took into account a typical distribution of VMT fractions among the different months of the year. Step 2 was performed only for vehicles that were not in the possession of the household for the entire calendar year.
Incomplete Odometer-Based VMT Data. In 1994, no odometer span was available for 35 percent of the sampled vehicles, although an estimate of annual VMT had been obtained from the respondent during the RECS interview. VMT for these vehicles was imputed from a regression of actual annual miles (from those households in the 1994 RTECS that had both mileage figures) on the estimated VMT obtained from the RECS. For another 17 percent of sampled vehicles, no odometer span was available and a VMT estimate was not obtained during the RECS interview. VMT for these vehicles was imputed using a multiple linear regression model, where the independent variables were number of drivers, household income, age of household head, type of vehicle, and use of vehicle on the job. This regression was also used for imputing VMT for vehicles that were being acquired or disposed. Both of the regression models described above yielded estimates of annualized VMT. The Step 2 adjustment described previously was then used to adjust this VMT to correspond with the time the vehicle was in the possession of the household.
To compare the two methods (odometer-based and respondent self-reporting) for estimating VMT, odometer-based and respondent self-reporting VMT were calculated for each vehicle in the 1991 and 1994 RTECS. A ratio of odometer-based VMT to self-reported VMT was used to determine the magnitude of the difference between the two methods. Because the 1991 RTECS results were very similar, only the 1994 RTECS results will be presented.
There were two sets of comparisons made. First, comparisons were made using the full 1994 dataset. Second, a reduced 1994 dataset was created by eliminating vehicles that did not have two odometer readings. In addition, the impact of eliminating part-year vehicles was reviewed but not included because it did not impact the final results. The reduced dataset only included vehicles with two odometer readings so not to bias the results by having a portion of the odometer-based VMT values computed with two odometer readings and a portion computed with one odometer reading and an imputed mileage value. Table 1a presents the reduced dataset and Table 1b presents the full dataset. However, only the results from the reduced dataset will be discussed here. In general, comparing the two methods for computing VMT revealed the following:
NPTS Odometer-Based VMT -- PDF Format. In 1995, each NPTS respondent was sent a form to record two odometer readings for each vehicle, along with the day, month, and year of each reading. The two odometer readings were then used to produce an annualized VMT taking into account seasonal patterns of driving.
NPTS Self-Reported. This method of estimating VMT by NPTS was similar to the RTECS self-reported VMT. Respondents were asked, "How much was that car driven during the last 12 months?"
1995 NPTS Results for The Two VMT Methods
The 1995 NPTS web site's analysis tools were used to compute average VMT per vehicle for odometer-based VMT and self-reported VMT. These results are included in Table 2. The 1995 NPTS average odometer-based VMT was 11,801 and the average self-reported VMT was 12,226 which gives a ratio between these two variables of .96. Therefore, NPTS self-reported VMT was 4 percent higher than the odometer-based VMT.
Both surveys show that self-reported VMT were higher than odometer-based VMT. Since odometer-based VMT is known to be more accurate than self-reported VMT, if possible, odometer-based VMT should be obtained. However, when only self-reported VMT values are obtained, adjustments should be made to reflect the differences noted here between self-reported and odometer-based VMT. Also, the RTECS and NPTS values for odometer- based VMT were within 3 percent of each other with NPTS being slightly larger. These results are amazingly consistent, given that VMT is understood to be slowly increasing over time.4End Notes
1. Energy, Information Administration, Household Vehicles Energy Consumption 1994, Appendix B Estimation Methodologies, August 1997, DOE/EIA-0464(94).
Top of Page
Release Date: August 3, 2000