Homogeneity of Variance
So, I ran a Factorial RG, Fixed Effects and found a significant interaction and no significant main effects but I have a substantial violation of homogeneity of variance. When I computed the Fmax, it came out to 15.54. First of all, is this considered moderate or severe (which would impact my significance at p = .016)? Secondly, in looking at the issue of nonorthogonality, since my highest order effect, my 2-way interaction, is significant but none of the others are, can I ignore the issue of heterogeneity, or is the most substantial issue going to be in my comparison of marginal means?
posted by Kris Thomas @ 8:30 PM
3 Comments:
At 8:53 AM,
Mari said…
There are no hard and fast cutoffs for what makes the violation moderate or severe. Having said that, compare your obtained value to the critical value for your situation. If the Fmax critical for this analysis should be 3, 15+ is very different. If the Fmax critical is 10, not as different.
Second, you said you had a significant interaction, which I assume you followed up with an appropriate post-hoc (i.e., one that did not assume equal variance)?
At 7:06 PM,
Kris Thomas said…
Well, originally I ran the data with an ordinal IV, but I realized its a better model if I combine two of the factors, which really makes better sense anyway (putting the happy and very happy couples together into one). This results in better cell sizes where I am not dealing with an Fmax issue. When I did this I found a significant interaction and main effect, so I will do t-tests to determine the relationships b/c there are not enough levels to do post-hoc.
At 12:29 PM,
Mari said…
If you have a significant interaction, that means you have an effect that includes (at minimum) 4 cells. Is there a reason you are not creating a new variable with (at minimum) 4 levels and testing via regular MCPs? That is, you only use t-tests when you can't use the other options.
Post a Comment
<< Home