Sampling Biases Associated with Driver Distraction Tasks in a Simulated Environment

Abstract

Participants recruited for driving simulator studies may need to be excluded if they fail to complete the study as designed. The data that is analyzed therefore includes a sampling bias that is not typically considered but important to recognize. This sampling bias was examined using a study involving five data collection sites in the US. The goal of the original study was to examine the effectiveness of a forward collision warning system. A distracting task is part of the study protocol to divert drivers’ attention away from the braking lights of the forward vehicle. In this experiment, two groups of participants became ineligible for further data analysis: (1) conservative drivers who had difficulty being artificially distracted and (2) risky drivers who did not maintain vehicle control while distracted. Two separate binary logit models were used to identify factors associated with these two driver types. As age increased, drivers were more likely to be in one of these driver types and are therefore, likely to be disqualified from further examination. Females had a higher likelihood of losing control during the distraction task when compared to males. Further, drivers with previous driving simulator experience were better able to maintain vehicle control while distracted. Drivers in two (Iowa City, IA and Seattle, WA) of the five sites were less likely to be distracted. The sampling biases observed among older drivers and female drivers may have implications with respect to the generalization of the findings. The site differences observed may be indicative of ordering effects in that IA and WA were the first two sites to begin data collection. Once issues with attrition became apparent, there was a greater emphasis on the experimental training in the latter data collection sites.