Correlation analysis to investigate unconscious mental processes: A critical appraisal and mini-tutorial

Area of Science:

• Cognitive Psychology
• Neuroscience
• Psychometrics

Background:

• Investigating unconscious mental processes often involves correlating task performance with stimulus awareness.
• A common but flawed approach is to infer unconscious influence from a non-significant correlation between these measures.
• This 'null-correlation' approach is problematic, especially with unreliable psychological measures.

Purpose of the Study:

• To highlight the pitfalls of using null-correlation approaches to infer unconscious processes.
• To provide guidance on appropriate statistical methods for analyzing task performance and stimulus awareness.
• To re-evaluate a study on memory suppression using a more robust statistical framework.

Main Methods:

• Critique of Null Hypothesis Significance Testing (NHST) for interpreting non-significant correlations.
• Introduction to Bayesian statistical methods for comparing evidence for null vs. alternative hypotheses.
• Application of Bayesian analysis, accounting for measurement unreliability, to existing memory suppression data (Salvador et al., 2018).

Main Results:

• NHST inappropriately interprets failure to reject the null hypothesis (correlation = 0) as evidence for the null.
• Measurement unreliability can attenuate observed correlations, masking true relationships.
• Bayesian analysis of Salvador et al.'s data showed no to moderate support for unconscious memory suppression, contingent on the alternative hypothesis model.

Conclusions:

• The null-correlation approach is statistically unsound for inferring unconscious processes.
• Bayesian methods offer a superior framework for analyzing the relationship between performance and awareness, especially when accounting for measure reliability.
• More reliable data and advanced analytical techniques are needed to definitively understand unconscious influences on cognition.