Correlated parameters in driving behavior models: Car-following example and implications for traffic microsimulation

Behavioral parameters in car following and other models of driving behavior are expected to be correlated. An investigation is conducted into the effect of ignoring correlations in three parameters of car-following models on the resulting movement and properties of a simulated heterogeneous vehicle traffic stream. For each model specification, parameters are calibrated for the entire sample of individual drivers with Next Generation Simulation trajectory data. Factor analysis is performed to understand the pattern of relationships between parameters on the basis of calibrated data. Correlation coefficients have been used to show statistically significant correlation between the parameters. Simulation experiments are performed with vehicle parameter sets generated with and without considering such correlation. First, parameter values are sampled from the empirical mass functions, and simulated results show significant difference in output measures when parameter correlation is captured (versus ignored). Next, parameters are sampled under the assumption that they follow the multivariate normal distribution. Results suggest that the use of parametric distribution with known correlation structure may not sufficiently reduce the error due to ignoring correlation if the underlying assumption does not hold for both marginal and joint distributions.