Логистическая регрессия с категориальными предикторами и эффектами взаимодействия и CHAID: сравнительный анализ на эмпирическом примере

Социология: методология, методы, математическое моделирование. 2016. № 43. С. 63–99.

Rotmistrov A., Popova P.

The focus of this article is the methodological aspect of political activism determinants identifying; specifically variants of handling with categorical predictors which hypothetically explain the level of activism. When using regression for explaining the issue, one may transform such predictors into dummy variables. Such a popular solution makes the model bulky and causes troubles with assessing this model’s quality. Moreover, if a researcher wants to consider interaction effects of the mentioned predictors, the supernumerary combinations of the mentioned predictors values are pended because regression modeling does not take into account the degree of similarity of the mentioned predictors values’ effects. The article authors proposed CHAID as the alternative to the mentioned solution. The research’s aim was i) a comparison of the two mentioned methods leaning on their a priori known properties; ii) arguing CHAID’s some theoretical advantages comparing to logistic regression, iii) parallel implementing the two methods, iv) a comparison of gained empirical results and v) arguing that it is useful to examine multiple interaction effects when developing a predictive model. The raw data were extracted from ESS 2012. The dependent variable was Political activism; the hypothetical predictors belonged to the socio-economic bloc of the Panel.

Keywords: логистическая регрессия, political activism, политический активизм, logistic regression, деревья классификации, детерминанты, determinant, interaction effects, categorical predictors, категориальные предикторы, dummy variable, фиктивная переменная, classification trees, эффект взаимодействия