Handling Intra-Household Correlations in Modeling Travel: A Comparison of Hierarchical (Random Effect) Models and Marginal (GEE) Models

In this paper, the necessity for treating intra-household correlation is investigated by analyzing two travel behavior indices, i.e. travel time and travel distance, for three important travel motives (commuting, shopping, and leisure). Data stemming from the 2010 Belgian National Household Travel Survey are used in the analysis. Two model approaches that accommodate for intra-household correlation are compared, namely the generalized linear mixed model (GLMM) and GEE model approach. Both model approaches show that high levels of intra-household correlation are present, and therefore the use of models that take into account intra-household correlation, is strongly recommend. Results indicate that this requirement is the most urgent for non-commuting trips. Moreover, the results show that GLMM and GEE yield comparable estimates in the case of normally distributed data. Furthermore, evidence was provided that the more the estimates of the intra-household correlation provided by the two approaches differ, the less the homogeneity of the parameters is assured. In this regard, if one has to choose between the GLMM and GEE methodology, especially the negative consequences of choosing an inappropriate covariance model in the case of a GLMM model favor the selection of the GEE methodology. Further research is needed to compare the two approaches in the context of non-normally distributed travel behavior data.