The impact of social networks and coordinated destination choice on the spread of epidemics using Episim

Abstract

The emergence of GPS-enabled smartphones and crowdsourcing tools are unique opportunities for understanding transport behaviour. However, the datasets they generate are often unbalanced, as individuals may use the service collecting data at different frequencies and periods. This raises important questions: are typical discrete choice models robust to this unbalance? Are model estimates biased towards over-represented individuals?

This paper tackles the issue of handling unbalanced panel datasets for route choice modelling. It first develops a simulation experiment to study to which degree Mixed Logit Models with panel effects reproduce the population preferences using unbalanced data. It then investigates bias reduction strategies, using subsampling and likelihood weighting. These strategies are compared to give guidelines that fit the model purpose. We show that weighting and subsampling techniques can reduce the bias when interpreting the model output for tastes. Combining these techniques helps to find an optimal trade-off between bias and variance of the estimates.