BINARY REGRESSION MODEL DIAGNOSTICS: AN APPLICATION TO CHILDHOOD DIARRHOEA DATA

Abstract

Binary logistic regression model is applied in many public health studies, that involve binary response variable such as presence or absence of diarrhoea in a child. In such applications, most studies in literature have focused on inferences and implications on relevant policies. There has been little effort to exhaustively understand the fit of the binary regression model to the data the at hand, before making conclusions and recommendations on policies. This study focused on utilization of the post-estimation diagnostic statistics that are available for fitting binary regression models to data, which are usually ignored in most applications of the model. This was done by applying diagnostic statistics that analyze the presence of outliers, influential observations, high leverage subjects and multicollinearity among independent variables, upon fitting binary logistic regression model to child dirrhoea data from 2015-16 Malawi demographic and health survey. The results showed that there were outliers and high leverage points in the model. Region and toilet sharing variables were mostly affected by outliers. But using Cook’s distance, the individual children did not have influence on all estimated regression parameter values. The study recommends that analysts should throughly examine the fit of the logistic regression model.