Three Approaches to Testing for Statistical Suppression

Area of Science:

• Psychometrics
• Statistical Modeling
• Quantitative Psychology

Background:

• Suppression effects, where an effect unexpectedly increases after adjusting for a third variable, are crucial in theoretical and applied research.
• Understanding and accurately testing for suppression effects is essential for robust statistical analysis.

Purpose of the Study:

• To investigate and compare three distinct statistical approaches for testing suppression effects.
• To evaluate the performance of these tests through simulation and real-world data analysis.

Main Methods:

• Compared three tests for statistical suppression: one based on zero-order and semi-partial correlations (1978), another on a necessary condition for suppression (1997), and a third extending the inconsistent mediation test.
• Derived standard errors for the Velicer, and Sharpe and Roberts tests.
• Conducted a statistical simulation study and applied tests to real data sets and published correlation matrices.

Main Results:

• The test based on inconsistent mediation demonstrated superior properties in the simulation study.
• When applied to example data, all three tests yielded consistent results.
• Analytical work identified conditions under which the tests produced conflicting outcomes.

Conclusions:

• The mediation test for suppression, specifically evaluating the sign of the product of mediated and direct effects, exhibited the best overall performance.
• Accurate identification of suppression effects is vital for advancing statistical understanding and application.