The Role of Interaction Effects in Eliminating Biases of Regression Coefficients Caused by the Presence of Random and Non-Random Missing in the Data

The problem of incomplete data is particularly relevant for a sociologist, especially if the data contains non-random missing (MNAR) that is almost impossible to distinguish from random (MAR) in practice. Such well-known scientific methods of dealing with missing data as Complete Case Analysis and Missing Indicator Method do not provide accurate estimates of regression coefficients. This paper will focus on the role of interaction effects in eliminating bias of regression coefficients caused by the presence of random and non-random missing in data in case of Complete-case-analysis and Missing-indicator-method. The basis of this study was a statistical experiment on simulated data (N=2000), the immediate goal of which was to understand whether the interaction effects in linear regression contributes to unbiased estimates, and therefore the reliability of the results when using two different methods of handling with missing data. During the experiment, it was found that the Missing Indicator Method as a method, that avoids reducing the sample size during the analysis, can be used for both random and non-random missing without losing the quality, even if there are only direct effects in the regression model. Complete Case Analysis does not give an accurate result in the case of non-random missing data, but the inclusion of interaction effects in the model, along with direct effects, eliminates the bias of regression coefficients.