118 – Issues, Challenges, and Solutions in Modernizing NHTSA's NASS
Estimating Population and Design Parameters for Nhtsa's New National Automotive Sampling System (Nass)
Rui Jiao
Westat
Yumiko Sugawara
Westat
Martha Rozsi
Westat
Sharon Lohr
Westat
James Green
Westat
William Cecere
Westat
The new National Automotive Sampling System (NASS) sample design uses a multivariate optimization method to solve for the sample sizes at the first, second, and third stages, with the considerations of the operation cost and the variance of the variables of interest. The calculation of variance called for building a Police Jurisdiction (PJ) level sampling frame that includes population crash count by Police Accident Report (PAR) stratum, population count of the key estimates, both for the new NASS General Estimates System (GES) and the Follow-on Passenger Vehicle (FOPV) modules by utilizing counts by PAR classification reported to the state by the PJ. Multiple Linear Regression models were developed for estimating other county level estimates from PAR classification counts using the current PSU level data of GES and FOPV. These models were then applied to the PAR counts provided by the population of PJs within sampled PSUs. Special error terms were added to the models in order to add variability / noise across counties. The PJ list contains all PJs in a sampled PSU who report PARs to the state, and the list was further geocoded to map to Census geographic files to associate urbanicity with each PJ.