| Abstract:
Evaluators often wish to use multiple predictors to predict or model a dichotomous outcome (e.g., success/failure, persist/dropout, admit/deny, self selection into a treatment vs. control as in propensity analysis). Ordinary regression does not provide an appropriate model for this type of analysis, but logistic regression is a readily available alternative that is accessible in SPSS and other statistical packages. Logistic regression is not difficult to use and understand although new terminology and unfamiliar statistics can be challenging for first-time users. In this demonstration we will examine the logic and application of logistic regression for dichotomous dependent variables, show why ordinary regression is not appropriate, and demonstrate applications with dichotomous predictors, continuous predictors, and categorical predictors. Participants will be given a packet with SPSS syntax and annotated output for a range of applications. Familiarity with multiple regression analysis will be helpful, but not required.
|