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.

Tuesday, March 15, 2011

Comparing Cell Sizes for Fmax

Sorry to have so many questions :( ... When comparing cell sizes for Fmax, do we look at the initial cell size totals: for example there are 5 initial cells in a 2x3 factorial anova, or the cells once they are compared: 6 cells?

Also, if we have a cell size ratio of 7:1, is it okay to say that a Fmax of 3.22 is statistically robust since 3:1 requires less than or equal to 10 and 9:1 less than or equal to 3?

posted by bfullmer @ 6:40 PM  

8 Comments:

  • At 7:04 PM, Blogger David said…

    I'm a bit confused by your first question. Are you talking about determining Fmax or finding the ratio of cell sizes?

    Fmax is the largest variance (of all 6 cells) divided by the smallest variance (of all 6 cells).

    Cell size ratio is the ratio between size of largest cell and the size of the smallest cell.

    If your ratio is 7:1, it looks like you're entering the grey zone where there are no hard and fast rules. You could try Levene's test and hope that comes out nonsignificant. Otherwise, I'm not sure what to say except to provide the best rationale you can. You might want to include a Discussion section and address this limitation to be on the safe side (as mentioned in the earlier post about cell sizes).

     

  • At 7:26 PM, Blogger bfullmer said…

    First I was talking about cell size ratio... specifically, which cell sizes do you look at? Under my between-subject factors I have 5 cells: yes, no, happy, unhappy, or very happy. But under my descriptive stats I have six cells: yes-happy, yes-unhappy, yes-very happy, etc. Do I use the five or six cell version when figuring out my cell size ratio?

    Second, if Levine's is significant and my cell size ratio is 4.7:1 and fmax = 3.62, is that okay since it's close to the 4:1, < or equal to 10 distinction? I plan on reporting... do I need to do a discussion section or can I just include in the results section?

    Thanks!

     

  • At 7:54 PM, Blogger bfullmer said…

    Also, David, if deleting outliers (3) does not change significance of F values or Levine's and makes Kurtosis worse, but does improve (slightly) skew and Fmax would you do so?

    Thanks!

     

  • At 8:33 PM, Blogger David said…

    Oh I understand now. You should look at the 6 cells when calculating ratio and Fmax. The 5 "cells" are not really cells because you would then have some participants in more than one "cell."

    I don't know what the exact function between cell size ratio and acceptable Fmax is (there probably isn't one). Strictly speaking, assumption testing wouldn't fall under the results section (which is why Dr. Clements wants a discussion section for "glaring inadequacies"). Do your best from the information you have so far regarding how you report this.

    Outliers can be tricky to deal with. I would be inclined to leave them in if they do not affect significance.

     

  • At 9:26 PM, Blogger bfullmer said…

    Do outliers go in the results section or discussion? Do you report if you leave them in?

     

  • At 9:52 PM, Blogger David said…

    If you altered your data, describe that in your results (preliminary analyses).

    If you violated some assumptions in a "glaring" way, you'd discuss this limitation in a discussion section.

    If you're on the fence, writing about the limitations would be the safer route.

     

  • At 10:14 PM, Blogger bfullmer said…

    Thanks!

     

  • At 11:55 PM, Blogger bfullmer said…

    If I write about my homogeneity of variance issues in a discussion section for this project, do I need to do the whole discussion section as I would for a real paper, or just include the one limitation?

     

Post a Comment

<< Home