Linear Regression: Disagreement in confounding with stratification

Hi everyone,

I am performing a simple linear regression testing whether an the results from an aptitude test for our sales force is associated with their profitability (binary variable profitable or not). The score variable is a percentile.

I performed a simple logistic regression (profitable = score) which was significant. Since we have multiple sales centers I then decided to stratify on center location and found that only one of our centers was significant. Therefore, I believed this finding indicated that location was a confounding variable. I then ran another logistic regression, this time adding location (profitabe = location score) and location was not significant but score was. I then tried an interaction variable (profitabe = location score location*score) and score was significant again while location and interaction term were not. At this point I wondered if the fact that score was not normally distributed was an issue so I transformed the score variable into z-scores and got the same results.

A similar phenomenon takes place when I run a linear regression with profitability ($ amount) and score variables.

At this point I am unsure how to interpret the results. Is location confounding? Are the two non significant locations really not significant? Is the overall just significant because they are included in the overall, and if so then why would location not be significant in the model? Any help with interpretation would be greatly appreciated!



TS Contributor
"Singificant / not significant" unfortunately is not very informative (0.049 versus 0.50? 0.00001 versus 0.89?) You could tell us something about sample size, coefficients and p-values, perhaps.

With kind regards