Satchi (and others),

Suppression is when the strength of an association between two variables changes (statistically, i.e. coefficient change) when an additional variable is added to the model to predict the DV. Most commonly, the coefficient actually increases in the presence of the "suppressor" (the third variable), because the third variable accounts for a portion of the "irrelevant variability" (variability not associated with the original IV) of the DV. In its simplest conceptualization, a suppressor effect may be present when a regression coefficient between an IV and DV is stronger in the presence of a third variable, than the original correlation indicated (when the the third variable was not included).

As an example, lets say that we are examining a potential link between stress and depression (purely hypothetical), but our trusty correlation matrix surprisingly tells us that the two are not related. Surprised and determined to find an explanation, we decide to test for differences in gender on our outcome (depression), since the literature suggests that we might expect to see them. Sure enough, males and females differ on depression in our sample and when we include gender as a control (suppressing the variability of depression that was related to gender, but not stress), we are now able to see that stress and depression are significantly and positively related. In this example variability of depression based on gender was obscuring between person differences based on stress, so the effects of gender needed to be suppressed to reveal the effect of stress. I hope this is helpful! If you have additional questions, please feel free to visit my site www.varyyourstats.com or email me at [email protected]. Best of luck!

hi there! thanks for your reply. Makes sense and able to understand.
Ok I will look at your website. Thanks a lot for your advice and also wish you very good stats business (up)
satchi

Thanks Satchi, I'm glad it was helpful!