Bayesian highest density intervals of take-over times for highly automated driving in different traffic densities

Current human factors research on automated driving aims to ensure its safe introduction into road traffic. Although informative results are crucial for this purpose, most studies rely on point estimates and dichotomous reject-nonreject decisions that have been declared obsolete by more recent statistical approaches like new statistics (Cumming, 2014) or Bayesian parameter estimation (Kruschke, 2015). In this work, we show the objective advantages of Bayesian parameter estimation and demonstrate its abundance of information on parameters. In Study 1, we estimate take-over times with a relatively uninformed prior distribution. In Study 2, the resulting posterior distributions of Study 1 were then used as informed prior distributions for interval estimations of mean, standard deviation and distribution shape of take-over time in different traffic densities. We obtained 95 % credible interval widths between 490 ms and 600 ms for mean take-over times, depending on the condition. These intervals include the 95 % most probable values of the mean take-over time and represent a quantification of uncertainty in the estimation. Given the data and the experimental conditions, a take-over requires most likely 2.51 seconds [2.27, 2.76] when there is no traffic, 3.40 seconds [3.11, 3.71] in medium traffic and 3.50 seconds [3.21, 3.78] in high traffic. Bayesian model comparison with Bayes Factor is discussed as an alternative approach in conclusion.