Skip to content
Prev 207274 / 398506 Next

Equality between covariance matrices?

Greg Snow 
One approach is to do a permutation test on the covariances.  You need a measure of how different they are (maximum difference, mean difference).  Calculate that for the original data.  Then randomly permute observations as to which group they are in and recalculate your difference measure.  Repeat that a bunch of times (9999 maybe) then see where your difference for the original data falls in the distribution of all the differences.  If the population covariances are the same, then the observed should fall somewhere in the middle.  If they are different, then the observed should be in a tail (the proportion more extreme than the observed is the p-value).

Hope this helps,

Thread (2 messages)

Robert Lonsinger Equality between covariance matrices? Jan 22 Greg Snow Equality between covariance matrices? Jan 22