Suppressor

#1
Hi,

I am attempting to conduct a regression analysis in a study examining patient and caregiver variables and their impact on formal care satisfaction. I think I have a suppressor variable acting in my data - correlations without partialling out this variable are smaller and less significant than the same correlations repeated with this variable controlled for. Is there a test I can use to test if this is the case? If it is, how do I control for this variable in a multiple regression? All help greatly appreciated, I am so far beyond confused now!

R
 

Dragan

Super Moderator
#2
Hi,

I am attempting to conduct a regression analysis in a study examining patient and caregiver variables and their impact on formal care satisfaction. I think I have a suppressor variable acting in my data - correlations without partialling out this variable are smaller and less significant than the same correlations repeated with this variable controlled for. Is there a test I can use to test if this is the case? If it is, how do I control for this variable in a multiple regression? All help greatly appreciated, I am so far beyond confused now!

R
Well, actually, a suppressor variable is a good thing to have in a regression model. In short, if it is in fact a suppressor variable, leave it in.

The reason is because it may significantly reduce "noise."

Briefly, think of a model where you're predicting how well someone does on an essay test. Some participants may have an advantage because they can just simply read faster than the others and score better - yet have little knowledge of the subject matter.

A suppressor variable "reading speed" will penalize those who can read fast and help those who cannot.

You can test for its significance by using a traditional test of R^2.