Logistic regression models for single–source data – a simulation study
Tue TjurCopenhagen Business School
1. INTRODUCTION
Birch (2002) has argued that logistic regression in longitudinal data, also called panel data, or single source datain the marketing context, can produce serious inference errors when heterogeneity between respondents is ignored. What can happen is that an explanatory variable with essentially no effect on the binary responses appears to be strongly significant. Roughly because the reuse of the same respondents again and again results in a phenomenon which – in the most extreme case where the behaviour...