Abstract

The logistic regression model is commonly used in biomedical research to analyze binary outcome data. However, its assumptions have not been widely studied, which limits the post-estimation analysis of the model. This chapter explores the properties of the error term in logistic regression using mathematical and numerical techniques. The results from a simulation study showed that the error term of the model has a trimodal standard logistic probability distribution with a mean of zero and a variance between 0 and 1. The distribution is continuous and bounded below at \(-1\) and above at \(+1\). By utilizing these properties in the analysis of child mortality data in Malawi, seven outlier children were identified in the model. These children experienced death despite being in low-risk groups, such as having well-educated mothers, and a birth interval of over 36 months. The observed model properties provide a clear basis for conducting post-estimation analyses, such as detecting outlier observations in the model.

Original language	en
Pages (from-to)	247-281
Publication status	Published - 2025