Identifying and Correcting Errors with Odometer Readings from Inspection and Maintenance Data: Rollover Problem for Estimation of Emissions and Technical Change

The analysis of data from state inspection and maintenance (I/M) programs and remote-sensing studies has revealed a number of fundamental insights about vehicle emissions. These include a deterioration effect (emissions increase as vehicles age) and the vintage effect (because of technological improvements, newer model year vehicles are cleaner at any given age than older model year vehicles). As a result, vehicle age and vehicle mileage (from odometer readings) are both important proxies for estimating deterioration and identifying high-emitting vehicles. In this paper, Colorado data are used to demonstrate systematic problems with odometer readings from state I/M data. The paper shows that odometer readings are notoriously poor indicators of actual vehicle miles because of the rollover of five-digit odometers (i.e., from 99,999 to 0) and recording errors. Visual inspection of a distribution of I/M records plotted against odometer readings shows obvious issues with the use of raw odometer readings and implausibly high emissions levels for low-mileage vehicles. A three-part iterative process for identifying and correcting these readings is proposed. After the method is applied, it is found that the rollover problem affects roughly 1 in 20 of the observations in the data set. The problem is particularly acute for older model year vehicles, and negative correlations are observed between miles driven and emissions for certain model years. The results suggest that some results of empirical studies of I/M data need to be reexamined and that emissions improvements resulting from technical change may be overstated.