Exploration of Data-Pooling Techniques: Modeling Activity Participation and Household Technology Holdings

As data collection costs escalate and travel behaviour models become more data hungry, it becomes increasingly important to exploit existing sources of data to the greatest extent possible. Data fusion as a means of combining disparate sources of data, collected from entirely unconnected surveys, is therefore an avenue worth exploring. In this paper, the authors explore the possibility of pooling data from the UK National Travel Survey (NTS) and the UK Time Use Survey (TUS) to model the impacts of household technology holdings on leisure activity participation. The authors test three different data pooling techniques: ad-hoc cluster sampling, Rubin’s multiple imputation, and a Bayesian conditional probability model. The Bayesian conditional probability model uses the TUS data to develop a posterior distribution of the technology holdings and then integrates the leisure type model estimated on the NTS data over this posterior distribution. The results reiterate the fact that this is the most behavioural of the three data pooling techniques, and also support our hypothesis that household technology holdings are correlated with out-of-home (OH) leisure activity patterns.