# Negative $\beta$ in mediation regression analysis

Now, $X$ is my IV and $Z$ is the DV. In my analysis, I also have a mediating Variable $Y$. The significant $\beta$ while doing MRA between $X$ and $Z$ is -.85. Other conditions being satisfied, when I had MRA between $X$ and $Z$ with mediating variable $Y$, the sig. $\beta$ is -.32. Here if we do not consider the negative sign , only the magnitude of $\beta$ decreases from .85 to .32 and thus $Y$ partially mediates. But if we consider the negative sign -.32 is greater than -.85, and no mediation occurs. Some people are saying that in psychology, negative sign is only seen to find the direction of the relation and not the magnitude. Others are saying, we have to consider the magnitude with negative sign, which in case of mediation is giving a completely opposite answer. How should I interpret these results?

I did all my analysis in SPSS.

-
Have you considered replacing $Z$ by $-Z$? Or $X$ by $-X$? –  whuber Sep 9 '11 at 17:40
Having gotten my PhD in psychometrics, I've been subject to what people in psychology say. Oy. In your example, partial mediation occurs. The sign of the regression coefficient is not important. You don't say what your variables are, and so it's hard to give specifics. –  Peter Flom Sep 9 '11 at 20:52