Statistical monitoring of longitudinal categorical survey data

The longitudinal survey is conducted to collect responses from the target population repeatedly over long periods of time, aiming to analyze and monitor the response development as time goes on. However so far systematic detection methodology for survey response changes is still in its infancy. In this regard, this paper sheds light on this field by applying statistical process control (SPC) methodology into monitoring of longitudinal survey responses. Specifically, since generally the longitudinal survey responses are categorical variables which have temporal dependence on their previous values, i.e., autocorrelation, this research firstly proposes a hierarchical model to describe these categorical time series. The model can provide flexible autocorrelation structures for the categorical time series and therefore can be widely applied. Based on this model, this research secondly designs a SPC scheme using likelihood ratio test to monitor the survey responses. Estimation of the in-control (IC) response distribution is also discussed. Numerical studies demonstrate the satisfactory monitoring performance of the proposed scheme. Finally as an empirical evaluation, the scheme is applied to a real survey dataset to detect changes of consumer attitudes towards the economic conditions during a economic crisis.