The potential of joint application of logistic regression, variance analysis, and log-linear modeling for predicting personal health assessment
During the RSG open seminar, there was a pre-defense of the master thesis of Zaruhi Simonyan, the student of the HSE master program 'Applied Methods of Social Analysis of Markets', whose scientific supervisor is Alexey Rotmistrov.
Categorical variables are very often encountered in sociological practice, however not all analytical methods are adapted to work with categorical features. Logistic regression, variance analysis, and log-linear modeling can be identified as the methods that allow us to work with categorical variables. In the research, which formed the basis of the master's thesis, Zaruhi assessed the effectiveness of the joint application of these methods to predict the assessment of personal health.
Zaruhi stressed that the use of logistic regression is quite common in sociological studies of health, but the obtained models have acceptable quality indicators (for example, high values of the R-square analogs) quite rare. This fact led the student to think about the joint application of several methods to find the effects of interaction, which, if included in the logistic regression model, could improve its precision. As part of her master thesis, Zaruhi applied the variance analysis to estimate the maximum possible R-squared, then switched to the use of log-linear analysis to identify significant interaction effects and further built a logistic regression model with the inclusion of revealed effects in it. The result of her work is that the used algorithm did not significantly improve the quality of the regression model: the Nagelkerke R-squared increased by only 0.002. However, the use of interaction effects in the analysis of variance made it possible to increase the ANOVA's R-squared by 0.163, which gave the student a right to doubt the efficiency of modeling of social processes using linear relationships.
At the end of her speech, Zaruhi received useful recommendations from the RSG members, which she will be able to take into account in the forthcoming defense of the master thesis. RSG members, in turn, were introduced to another example of the successful application of interaction effects.