How to do a Post hoc of a nested ANOVA

Hi, I did a nested ANOVA and I would like to do a multiple comparison test as the Tukey's post hoc analysis.

I saw that the multcompare doesn't accept stats from nested ANOVAs.

I would appreciate any suggestion on how to permorm a post hoc for this kind of stats.

[Lu Zhang]

I'd like to know, too.

[Scott MacKenzie]

@Lu Zhang, you might consider posting this as a separate question, since the original posting is from 2013. However...

By "nested" do you mean a design with >1 factor? If so, consider treating the data as a single factor then using one of the anova functions followed by multcompare. For example, if you have data from a 3 x 2 within-subjects design, there are 6 columns of data. Treat the data as single factor with 6 levels, do the anova, then pass the stats object to multcompare. There will be 15 pairwise comparisons, performed using Tukey's method (default).

[Lu Zhang]

@Scott MacKenzie Thank you for your answer. I have one factor with 4 different levels in my design. But I wanted to clear the effect of subject. So I used anovan with subject as random varible. In this case, I can't use multcomapre. If I use multcompare, the random effect were treated as fixed effect according to the warning from matlab.

[Scott MacKenzie]

OK, so you have a single factor with four levels. So, there are six pairwise comparisons of interest. Seems simple enough. I'm not sure about the warning you are seeing. Perhaps post your code and data (or a subset of the data).

[Lu Zhang]

Now I used paired-sample t-test between each pair, in this case, six pairs. Then I perfomed Boferroni correction (0.05/4, i.e., four different levels). Is it ok?

[Scott MacKenzie]

Yes, this is reasonable, except the denominator, n, is the number of comparisons, not the number of levels. With 4 levels yielding 6 comparisons, use 0.05 / 6 = 0.0083 as the threshold for significance to maintain the family-wise alpha of .05.

[Lu Zhang]

Thank you so much.

Answer 1

[An Ko]

I'd like to know too!