ABSTRACT：Driving simulator has been used for geometric design safety evaluation during design stage. Proper sample size needs to be determined based on the expense of experiment setup for data collection. Conventional ways of determining sample size include loss function minimization and power analysis. This study puts forward an empirical method to decide the acceptable sample size that can be properly used in driving simulator experiments for geometric design safety evaluation. Acceptable sample size is measured based on the parameter coefficient convergence of the mean squared error (MSE) curve of significant variables chosen from mixed effects model. Tongji University eight-degree-of-freedom driving simulator was utilized to collect continuous driving behavior data from a mountainous freeway during design stage. Vehicle speed and lane departure events were selected as operational indices to reflect the influence of geometric design features on driving behaviors as these two variables are well related to operational efficiency and safety. Mixed linear model and mixed logistic regression model were used respectively to model the relationships between speed and lane departure with geometric design features. Random sampling method was applied to choose samples from 5 to 50 drivers ten times with every step of 5 drivers in the total of 55 drivers to investigate the model coefficient convergence. Acceptable sample size is determined based on the converging trends of significant variable coefficients. The clear converging trends of MSE curves indicate that 30 should be the acceptable sample size.
Xuesong Wang*, Shuang Liu, Bowen Cai, Qiming Guo, Xiaomeng Wang. Application of Driving Simulator for Freeway Design Safety Evaluation: A Sample Size Study. Transportation Research Board 98th Annual Meeting, Washington D.C., USA, 2019. 1.13-17.