Survival analysis: A fresh approach for analyzing response times in driving-related hazard perception paradigms

• Hazard perception (HP) is the defined as drivers ability to anticipate hazardous events.
• Response time (RT) is a typical measurement to examine drivers’ HP skills.
• Common practices were less likely than survival analysis to accurately estimate the actual RT distributions.
• Survival analysis is arguably a favorable method for analyzing driving related RT data.

Hazard perception (HP) is a driving skill and is described as drivers’ ability to ’read’ the road and anticipate hazardous events. Typically, exploring differences in HP abilities between experienced and young-inexperienced drivers involves a paradigm in which drivers observe short traffic-scene movies and press a response button each time they identify a hazard. The time measured from when the hazard is detectable up to when the driver pressed the response button is labeled response time. While some studies revealed that experienced drivers detect hazards faster than less experienced drivers, other studies did not report any differences. The current study suggests a possible explanation for this apparent contradiction by arguing that the way response time data are analyzed differs across studies and may lead to different conclusions. Specifically, it is argued that common practices fail to manage cases where a participant did not respond to a hazard, usually by replacing it with the mean response time or any other central tendency parameter. The current study proposes survival analysis as a more suitable methodology for analyzing response time in the HP realm. Survival analysis was compared to other commonly used methods on the basis of their ability to estimate the actual distribution from where response time data were generated. Both simulation of pre-defined distributions and real-world data revealed that commonly used methods underestimate the real distribution whereas survival analysis technique accurately estimates its values. The implications of these findings are discussed.