Application of logistic regression in experimental research

In experimental practice we often face the situation where the measured dependent variable takes one of two values only: 0 - lack of the measured characteristic or 1 - observation of the measured characteristic (behavior, consent to something, displaying an attitude or an opinion etc.). Both the general linear model as well as the linear regression analysis cannot be applied to dichotomous, nominal dependent variables. In such cases we are forced to use the non-linear analysis. Logistic regression is the model used for this type of dependent variables. This article presents application of the binomial logistic regression in experimental research. It explains specification and interpretation of typical logistic regression coefficients such as odds ratio, Wald coefficients, likelihood ratios. It presents the estimation procedure of the model parameters with maximum likelihood procedure and the Hosmer-Lemeshow goodness of fit test. Introduced were simple sample analyses (with nominal and quantitative predictors), a two-factor analysis as well as a two-factor analysis with interaction effect. The number of formulas and algebraic transformations were cut to the necessary minimum and the shown sample analysis and their interpretation were conducted step by step with the SPSS Statistics Pack version 17.0 PL. peerReviewed publishedVersion