An important research topic in the field of Automated Driving (AD) is its interaction with human behavior. This requires algorithms for modeling human driver behavior, pedestrians, bikes, modeling of vehicle dynamics, and much more. One approach to this is to base the traffic simulations and traffic participant models on stochastic processes with stochastic distributions as input parameters. This leads to new traffic situations and edge cases that can be analyzed. Obtaining the input distributions and understanding how they influence the simulation result can be challenging. However, understanding the importance of input parameters can be crucial to further improve simulation models.

This thesis develops a framework for measuring the effect of input parameters for stochastic models on the results of traffic simulations. Global variance-based sensitivity analysis, specifically first- and total-order Sobol' indices, measure the sensitivity of the simulation models in this framework. The focus lies on the total-order effects because these models are supposed to have non-negligible parameter interactions and strong nonlinearities. Monte Carlo approximation of the Sobol' indices addresses the potentially high number of parameters, i.e., more than 28. The samples are generated quasi-randomly to improve the convergence of standard pseudo-random sampling. To further reduce computation time, the execution of the simulations is parallelized. The framework is applied to BMW's Stochastic Cognitive Model (SCM) in multiple traffic scenarios generated by the Open Platform for the Assessment of Safety Systems (openPASS). This application demonstrates the plausibility of the framework's analysis results and showcases the usage and interpretation of the sensitivity indices. The results show that the parameter's Sobol' index highly depends on various conditions, such as the traffic scenario, the analyzed times step of the simulation, and how the parameter was defined. The differences between the analysis results for first- and total-order indices indicate strong higher-order parameter interactions. This justifies the computational costs necessary for calculating total-order effects. With the total-order effects, researchers and developers can identify interactions between parameters in complex traffic scenarios and, therefore, obtain their influence with certainty. This supports the understanding of the way a model works. It can discover parameters with too much or too little influence and parameters that can be eliminated to simplify the model.