PG851: GLM--ANOVA

This blog is for the use of students enrolled in either section of PG851. It is intended to be a forum for asking, discussing, clarifying, and helping.

Monday, January 31, 2011

ANOVA Homework

I noticed in the factorial homework that in the case of the life quality univariate analysis that while we had a significant ANOVA result regarding region (p=.036), there was no significant result in the post-hoc Bonferroni analysis in any of the 3 comparisons. Is this possible or is my analysis faulty?

Thanks!

posted by Rebecca Burnside @ 4:38 PM  

6 Comments:

  • At 5:06 PM, Blogger Ryan T said…

    I ran into the same problem. Not sure what that means.

     

  • At 7:51 PM, Anonymous Anonymous said…

    I had the same thing too.
    - Ashli

     

  • At 8:03 PM, Anonymous Nikki said…

    Sorry for the wait guys. We'll be answering this question very soon on the blog!

     

  • At 8:45 PM, Blogger David said…

    I haven't run the data myself yet, but here is one of Mari's responses to a blog post last year that may be relevant.

    "The omnibus F (i.e., the F in the Test of Between Subjects Effects) tells you that, when you look at all groups together, there is a systematic group effect for the IV you are testing.

    The post-hocs in the Multiple Comparison Procedures tests tell you whether or not there are significant pairwise differences between groups. Thus, the post-hocs do NOT consider all groups together, but one pair of groups at a time.

    It is, in fact, possible to have a significant overall (or omnibus) F but not have significant results on your post-hocs. (This is more common when testing interactions, but can occur in oneway analyses.)

    You should NEVER, NEVER, NEVER run post-hocs in the absence of a significant omnibus F, but it is possible to have pairwise comparisons that reach significance even when the overall effect is not significantly different from chance. Note that examining these post-hocs in the absence of a significant omnibus F is akin to running multiple t tests. The actual alpha level in this case is not < .05."

     

  • At 9:08 PM, Anonymous Anonymous said…

    Hi David,
    That doesn't quite clear up the problem. (Or if it does I don't get it). The omnibus F is significant but the Bonferroni post-hoc doesn't show significance at any level. The LSD (which we were told never to use) shows 2 significant interactions. We were specifically told to use those post-hocs so what do we do?
    - Ashli

     

  • At 10:59 PM, Blogger Mari Clements said…

    David's response is actually spot on, and does address exactly what you were asking about--an ANOVA with an overall (I.e., omnibus F) significant analysis but NO significant pairwise comparisons.

    If this still doesn't make sense, please see my response to a similar question above.

     

Post a Comment

<< Home