Hello all,

I am attempting to test the significance of a dichotomous, categorical moderator (role type: "service employees" = 0; "sales employees" =1). Is it the case that to test for the significance of this dichotomous moderator, I simply do the following:

1.) Obtain the Cochran's Q result of both groups pooled together,

2.) Obtain the Q from each group individually,

3.) Sum the Q results of both of these groups,

4.) Subtract the above sum from the 'pooled' Q results (i.e. what i did in step #1),

5.) Test the right-tailed probability of the chi-squared distribution on the remaining Q value with 1 degree of freedom? (i.e. use the Excel command "=CHIDIST(Q_between, 1)")

6.) If the resulting alpha level is significant (i.e., p < .05), this suggests that group membership (i.e. sales vs. service roles) is a significant predictor of variance ?

Is it truly that easy?.. I ask because I have tested this a variety of different ways and have obtained wildly different results. (Also, most of the effect size relationships I am exploring are "significant" when I use this approach and I feel that I must be missing something..)

Any help you all could give me would be much appreciated!