Pilot Implementation of a Vehicle Occupancy Data Collection Program

About the report:

This study developed an approach for partially automating the extraction of vehicle occupancies from crash data.  The approach can be implemented within 2 to 3 days based on a Tableau file available from the Virginia Department of Transportation’s Traffic Operations Division, the execution of a Python script developed for this work, and the resultant creation of online maps showing occupancies by corridor, block group, jurisdiction (city or county), and district. 

With additional effort, it is possible to reduce potential crash bias.  One way of removing bias is heuristic: synthesize what are believed to be missing vehicles from the crash data.  This Type 1 bias correction method was highly productive in that it took only about 2 hours and did not require field data collection; this approach is suitable for towns and small cities or counties.  Although it had little impact for populous jurisdictions (e.g., it shifted the occupancy for Amelia County from 1.38 to 1.39), such correction had a substantial impact on small jurisdictions (changing the occupancy for Burkeville from 1.20 to 1.30). 

Another way of removing crash bias is statistical and is more labor intensive.  With Type 2 bias correction, there is a ground truth value: one measures occupancies in the field and then uses a regression model from these field estimates to adjust the crash occupancies for a specific corridor.  This method requires considerable effort (e.g., 22 hours in the Richmond District) and is not always productive.  Accordingly, guidance for when Type 1 and Type 2 bias correction should be performed is given herein; for instance, the research shows that Type 2 bias correction should be attempted only with sites where the occupancy for injury crashes is higher than the occupancy for property damage only crashes.

Historically, detailed occupancy data such as those developed in this study have not been available, and thus in consultation with the project’s technical review panel, uses of these data to support planning decisions were explored.  One application pertains to land development: how do various land use factors influence occupancy throughout Virginia?  Geographically weighted regression showed that income, mean travel time, population density, and degree of land use mix explains about 40% of the variation in occupancy.  For instance, for the urban state capital of Richmond, a 10% increase in households earning less than $15,000 annually was associated with an occupancy increase of 0.06.  For rural southwest Virginia, occupancy increased by about 0.19 for areas having long travel times to work (compared to those having shorter commuting times).  As these results suggest modest potential for regional variation to inform land development decisions, such findings can support more nuanced land development reviews for interested localities.

Supplemental files can be found at: https://library.vdot.virginia.gov/vtrc/supplements

Authors