Problems with step-wise regression in research on aging and recommended alternatives

Area of Science:

• Gerontology
• Biostatistics
• Statistical Modeling

Background:

• Step-wise regression techniques are frequently employed in aging research.
• Correlated predictors, common in aging studies, present interpretative challenges for standard regression methods.

Purpose of the Study:

• To evaluate the interpretative difficulties of step-wise regression in aging research with correlated predictors.
• To assess alternative regression techniques for determining predictor importance in the presence of multicollinearity.

Main Methods:

• Geometric and algebraic approaches were used to analyze step-wise regression with orthogonal and correlated predictors.
• Techniques for multicollinearity detection were discussed.
• Principal components regression, ridge regression, and hierarchical regression were evaluated.

Main Results:

• Step-wise procedures yield poor tests of predictor regression weights when predictors are correlated.
• Hierarchical regression offers a more stable and reliable method for assessing predictor importance.

Conclusions:

• Hierarchical regression is the most recommended technique for analyzing correlated predictors in aging research.
• This method is preferred because it is theory-driven, unlike empirical relations that can be sample-specific.