Abstract
This research explores the impact of inflated variances on statistical significance and identification of variables correlated with median crossover accident frequencies. In roadway and accident data that is comprised of cross-sectional panels, serial-correlation effects across time arise due to shared unobserved effects through random error terms. We use an empirical technique to adjust for downward bias in standard errors in count models of traffic accidents, so that proper identification of variables is ensured. The identified model structures reflect the impact of inflated significance values, and provide specifications that include truly significant geometric, traffic and environmental effects contributing to median crossover accidents. It should be noted that the results from this study are localized and hence limited to inferences from the median crossover context in Washington State in the United States. Further study is required to ensure the transferability of these findings to other contexts.
