As is likely true in other textbooks there are important topics that should have been included, but do not seem to logically fit in the existing chapters. In Section 10.2 we provide an overview, with an example, of the use of propensity scores in logistic regression modeling, where the goal of the analysis is to estimate the effect of a treatment. Section 10.3 considers fitting a logistic regression model to sparse data where one cannot rely on large sample assumptions. This is often referred to as “exact” logistic regression. Throughout the text we have assumed any missing data would have no effect on the analysis. In Section 10.4 we consider methods for dealing with missing data. Section 10.5 considers choosing a sample size when the goal of the analysis is to estimate the effect of one covariate where the effect of others has been controlled using a logistic regression model. An introduction to the Bayesian approach to logistic regression modeling is presented in the Section 10.6. In Section 10.7 we consider regression modeling of a binary outcome using link functions other than the logit. The chapter concludes with sections on effect mediators and other modes of statistical interaction.

In a typical observational study, subjects who received a particular treatment have likely not been randomly assigned to treatment. Thus, estimation of a treatment effect ...

Start Free Trial

No credit card required