Effective, Fair, and Over-the-Air: Data-driven policy recommendations for passenger vehicle inspection programs in the United States

This dissertation presents three examples of how jurisdictions can employ large data sets and statistical methods to assess and design efficient and equitable passenger vehicle inspection and maintenance programs (I/M programs). Chapter 2 quantities the effects of I/M programs on road fatalities by applying panel specifications to a data set covering all 50 U.S. states over a 44-year period. Fixed effects (FE) estimates suggest that states with I/M programs had 5.5% fewer roadway fatalities per 100,000 registered passenger vehicles each year. A supporting two-stage least-squares (2SLS) specification also implies causality. Chapter 3 proposes transparent models to identify over-emitting vehicles, using only data which may be collected over-the-air through telematics systems. These modelswere up to 24% more accurate than the current test methodology, in a stratified data sample. These statistical models are a proof-of-concept for I/M programs where jurisdictions may remotely identify over-emitting vehicles, without requiring a physical inspection. Chapter 4 breaks down the annualized user costs of emissions inspection programs. An assessment of data from Pennsylvania finds that in 2016, the average user cost was $56 per vehicle—which includes $44 in test fees, $3 in repair costs, and $9 in indirect costs. Further, the results show that costs vary substantially: that on average, older vehicles have lower test fees but higher repair costs, and that vehicle owners in neighborhoods with lower median incomes or more minority residents may pay more in test fees and repair costs. The proliferation of telematics, driver assistance technologies, and electric vehicles is ushering in an era of ‘intelligent’ transportation systems. As modern vehicles rely on increasingly large amounts of data, this dissertation presents a case for why the regulations and policies governing their inspection and maintenance, may be improved by leveraging data-driven assessments and models.