Base-rates, cut-points and interaction effects: the problem with dichotomized continuous variables

Area of Science:

• Social Psychology
• Statistical Modeling

Background:

• Recent discussions question the validity of linear models for analyzing interaction effects.
• Dichotomizing continuous variables can lead to unstable interaction effect estimates.

Purpose of the Study:

• To investigate the impact of dichotomizing continuous variables on interaction effects.
• To evaluate the appropriateness of linear difference and ratio models in this context.

Main Methods:

• Simulated a dataset with known interrelationships between stress, social support, and depression.
• Analyzed the simulated data using tabular methods.
• Examined interaction effects in ratio analysis models.

Main Results:

• Interaction effects are demonstrated to be unstable when continuous variables are dichotomized.
• The choice of cut-points for independent variables critically influences the size and presence of interaction effects.
• Ratio analysis models are particularly sensitive to the selection of cut-points.

Conclusions:

• Linear models analyzing dichotomized variables may produce misleading interaction effects.
• Researchers must carefully consider the implications of variable dichotomization and cut-point selection.
• The findings highlight the need for cautious interpretation of interaction effects in social science research.