Two-Part Predictors in Regression Models.

Area of Science:

• Statistics
• Social Sciences
• Psychology

Background:

• Regression models are common for analyzing relationships between variables.
• Predictor variables may be undefined for certain participants or time points in longitudinal studies.
• This unique type of missing data requires specialized handling beyond standard imputation methods.

Purpose of the Study:

• To present a straightforward statistical method for regression analysis with undefined predictor variables.
• To address the challenge of predictors that are not applicable to all observations.
• To provide a practical solution for researchers encountering this specific data structure.

Main Methods:

• Development of a novel statistical technique to incorporate undefined predictors.
• Illustrative example using a romantic partner substance use predictor.
• Algebraic justification and simulation studies to validate the method.
• Practical guidance on implementing the technique in statistical software.

Main Results:

• The proposed method effectively accommodates predictor variables that are undefined for some observations.
• Simulation results demonstrate the method's validity and performance.
• The technique provides a robust way to analyze data with this distinctive missingness pattern.

Conclusions:

• The presented method offers a valuable tool for researchers dealing with undefined predictor variables in regression.
• This approach enhances the accuracy and applicability of statistical models in complex observational studies.
• It provides a statistically sound way to handle a common yet challenging data scenario.