Roman Pavelka, Statistical Office of the Slovak Republic, Slovak Republic

Type of article: informative article
Pages: 50 – 66

Abstract

The aim of the paper is to present the methods of interpretation of statistical interaction by analysing the conditional effects that constitute the interaction, using the analytical capabilities of selected procedures of the SAS statistical system. The presented paper will mainly deal with the estimation of simple effects or slopes of general linear models and testing their significance. The conditional effect is estimated by regression directly on the reference level of the moderating variable. Further steps are required to estimate effects at other levels. The differences between simple effects or slopes will also be tested as tests of differences in effects at the reference level of the moderating variable and effects at other levels of this variable. The visual representation of the interaction will be illustrated using graphs of simple effects or slopes, which usually provide the simplest and fastest interpretation. Examples of modelling of statistical effects and their visualization will be demonstrated using a PLM procedure. However, the functionality of this procedure gives data analysts a wide range of possibilities in analysing the estimated statistical model, testing statistical hypotheses about effects and their visualization, reparametrizing the model for estimating contrasts, and modelling prediction values. The GLM analytical procedure will be used for the estimation of statistical models.