🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
suppression_effect
A suppression effect occurs when including a third variable in a regression model increases the magnitude of an existing association, or reveals a previously hidden association, by absorbing irrelevant variance in the predictor. This is the inverse of confounding: while a confounder inflates a relationship and needs to be controlled, a suppressor deflates a relationship and its inclusion strengthens or reveals the true effect.
Verbal test scores and academic performance are weakly correlated. When anxiety is added as a control variable, the correlation between verbal scores and performance increases substantially, because anxiety was capturing variance in verbal test scores that was irrelevant to academic ability.
A study finds almost no relationship between hours of physical exercise and work productivity. Once researchers control for chronic pain levels — which reduces both exercise ability and productivity — the true positive effect of exercise on productivity becomes clearly visible, having been masked by the confounding influence of pain.
In a political science study, income level and support for a social program appear nearly uncorrelated. When researchers add education level as a control variable, a strong positive relationship between income and program support emerges — because education was acting as a suppressor, pulling the apparent correlation toward zero by being negatively related to one variable and positively to the other.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Does the correlation between two variables change substantially when a third variable is added to the model?
Type: binaryDoes adding a control variable increase (rather than decrease) an estimated association?
Type: binaryIs the suppressor variable correlated with one predictor but not with the outcome?
Type: binaryCould the suppressor be absorbing irrelevant variance in the predictor, sharpening the predictor-outcome relationship?
Type: binaryA suppression effect occurs when including a third variable in a regression model increases the magnitude of an existing association, or reveals a previously hidden association, by absorbing irrelevant variance in the predictor. This is the inverse of confounding: while a confounder inflates a relationship and needs to be controlled, a suppressor deflates a relationship and its inclusion strengthens or reveals the true effect.
Predictors often contain measurement error or variance due to irrelevant constructs. A suppressor variable is correlated with the 'noise' component of the predictor but not with the outcome, so controlling for it removes the noise and clarifies the signal.
When adding controls increases an association, examine the correlations among all variables. Distinguish suppressor effects from collider bias, which can also increase associations when controlled.
Suppression effects appear in educational psychology, where controlling for test anxiety reveals stronger aptitude-performance links. They are often mistaken for statistical errors or Simpson's Paradox.
Use these tools to detect, analyze, or train this aspect.