Moderated Regression Non-Signficant but Subgroup Correlation Significant??

#1
Hi Guys,
I'm having a bit of trouble trying to figure out some of my statistics and wonder if any one has some ideas for me?
I've run a moderated regression analysis as I've read in many places that it is the best way to approach moderation, but it was not significant. I then divided my moderator variable into 3 groups (low, medium and high scorers, n = 35 for each group) and found that there was a significant correlation between my IV and DV in the high scorers of my moderator variable.
My concerns are:
1. Is my significant finding meaningful at all in light of the many criticisms against split-median type methods?
2. Why does this significant result not show up in my regression? Would it possibly be due to a step-function of my moderator?

Thanks very much! Any thoughts appreciated!
 

Karabiner

TS Contributor
#2
Please describe the correlation coefficients
and p-values from the 3 groups. Please
describe the p-value for the moderator test.

With kind regards

K.
 
#3
Hi Karabiner,
The p value for the moderated recession analysis is .944. In the high scorers of the moderator group: r=.404 P= 0.01; medium scorers: r = -.2 p= .125; low scorers r= .007 p= .484.

Is this problem perhaps due to normality violations? I have read that correlation is relatively robust to normality violations, but not sure about regression. Should I try a log formation?

thanks very much!
 
#4
I have the same issue (i.e. non significant overall result for moderation but significant results at certain levels of the moderator). Can I ask what you ended up doing when reporting your results?