linear regression question

HorseRider · February 28th, 2012, 2:13 pm

regress A on B, get very small R square, say 1% and B is not significant in t-testregress A on C, r-squared 57%, C is significant in t-testbut A regress on B + C, R square is 79%, and both B and C are significant in t-testQuestion: is A ~ B+C linear regression valid? What does it mean that A on B not significant, but when it combines with other factors, B becomes significant? Any intuition about this. Thanks a lot!

Traden4Alpha · February 28th, 2012, 2:30 pm

I'd look at the scatter of the three datasets. Perhaps C contains some outliers that are the inverse of some outliers in B.

Alan · February 28th, 2012, 3:28 pm

The R^2 increasing is no surprise. I will guess the t-test in unreliable, for example due to autocorrelated or heteroscedastic errors.Trying testing and correcting for those kinds of problems.