Cochran's Q
Message posted by Ben Ditkowsky (via 172.170.48.163) on November 24, 2001 at 1:42 PM (ET)

I have a data set composed of within subjects data in which each paricipant is given a score (0=not identified, 1=identified)on 6 different measures. From my searches it appears that 2x2 tables would best be analyzed with McNemars test for dependant samples. As a first step I think I should be using an omnibus test (a nonparametric repeat mesures aov ) which appear to be Cochran's Q.

The problem is that Q tests the null hypothesis that there are no differences in proportions identified- In MY CASE, this is my hypothesis, that is the six methods of identification do identify significantly different proportions of participants. It seems like a non significant Q would be more likey (too liberal). Is Cochrans Q the correct statistic to use for this problem or is there a statistic that tests the null hypothesis that proportions identified are not equal?