I have 3 variables. My DV $Z$ of interest are reaction times, IV $A$ and IV $B$ are performance scores in psychometric tests. Looking at bivariate correlations, $A$ and $B$ are negatively correlated (-.40), $A$ and $Z$ are marginally negatively correlated (-.10) and $B$ and $Z$ are highly negatively correlated (-.50). Putting $A$ and $B$ in a regression model with criterion $Z$ leads to two negative standardized coefficients about -.40. In other words, If I am good in $A$, I have a fast reaction. If I am good in $B$, I have a fast reaction. But if I am good in $A$, I am not good in $B$?! Is this logical?

Coding of the variables should be alright.


1 Answer 1


Maybe the answer lies in the fact that some of the correlations are simply not significant in the sense that a large correlation factor might arise from a chance alignment of a small number of points. Looking at the different correlation coefficients, I would advance that maybe the correlation between Z and A is not that significant in this sense. A way of testing for this this is to do a permutation test (as a simple non-parametric test) which would associate a p-value to the correlation coefficient.

So in a nutshell, given the information you provided, the results you obtained are not fully incompatible.

  • $\begingroup$ the bivariate correlation is not significant, nor the predictor. still, a standardized beta of -.36 is hard to ignore. i will do what you told me $\endgroup$ Commented Jun 10, 2014 at 16:09
  • 1
    $\begingroup$ @pedrofigueira: Could you explain in what way 'significance' is relevant here? To the OP: could you please provide some graphical description of the situation? This is much easier than to speculate using correlations. $\endgroup$
    – Michael M
    Commented Jun 10, 2014 at 17:57
  • $\begingroup$ @Michael Mayer I just did. Hope it helps. $\endgroup$ Commented Jun 10, 2014 at 21:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.