Abstract
Previous studies of rectangular rapid-flashing beacons have found a wide range of driver yielding rates (19% to 98%). As part of the efforts to study yielding at different types of crosswalk locations better, this research compared alternative statistical methods to analyze staged crossing data. A database was compiled of 12,040 individual crossings at 73 sites representing 128 crossing periods, which included a set of 27 potentially influential variables. Analysis of this data set began with considering the hypothesis that negative binomial regression could be the most appropriate method of analysis, given the characteristics of the data collected at each staged crossing. However, alternative methods—in particular, logistic regression and analysis of variance—also appear appropriate. Even when these alternative methods may not explicitly account for some theoretical implications of staged-crossing data, the methods may prove to model yielding rates acceptably. A potential benefit of using the alternative methods is that the results may be easier to interpret in the context of decision making in regard to pedestrian crosswalk treatments. This research investigated the methodological aspect of analysis to advance the understanding of the explanatory factors, the results, and the phenomena under study. Results from various analyses indicated that logistic regression tended to yield results that were more accurate. Additionally, this research found that the statistical power of two of the techniques evaluated increased when explicitly accounting for the total number of cars observed at each crossing.
Get full access to this article
View all access options for this article.