ANOVA interactions,Tukey HSD

[valgo3117]

Hello
I am analyzing behavioral data on face processing.
This is a within-subject 3x2x2 design.
When I run repeated-measure ANOVA, I obtain significant main effects and 2 way interactions, but no 3-way interactions.
In order to explore the data further (and also because I have strong a priori assumptions about how individual conditions may differ), I compare means 2 by 2 using Tukey HSD tests.
One reviewer states that since the 3 way interaction is not significant, I cannot compute such Tukey tests.
Another one says that Tukey is a non-parametric test and should not be used in conjunction with ANOVA!

This is not what i thought...
I thought the (parametric!) Tukey HSD controls the probability of false rejection among all comparisons made, and so a significant effect or interaction is not necessary to compute it.

But I do not know what bibliographical reference or textbook to support this claim.
Do you confirm that aspect of Tukey HSD tests? And could you advise me an article or textbook to cite in support for this?

Many thanks in advance!

[TheAnalysisFactor]

Hello
I am analyzing behavioral data on face processing.
This is a within-subject 3x2x2 design.
When I run repeated-measure ANOVA, I obtain significant main effects and 2 way interactions, but no 3-way interactions.
In order to explore the data further (and also because I have strong a priori assumptions about how individual conditions may differ), I compare means 2 by 2 using Tukey HSD tests.
One reviewer states that since the 3 way interaction is not significant, I cannot compute such Tukey tests.

It's true that you should not compute Tukey tests to compare means after a non-significant effect in ANOVA.

If you have strong a priori predictions about how means differ, you would use planned contrasts. However, my experience is that a lot of reviewers view planned contrasts as post-hoc ways to find significance when the overall ANOVA is not. You can't really run both the factorial ANOVA and the planned constrasts.

Another one says that Tukey is a non-parametric test and should not be used in conjunction with ANOVA!

This is not what i thought...
I thought the (parametric!) Tukey HSD controls the probability of false rejection among all comparisons made, and so a significant effect or interaction is not necessary to compute it.

But I do not know what bibliographical reference or textbook to support this claim.
Do you confirm that aspect of Tukey HSD tests? And could you advise me an article or textbook to cite in support for this?

Many thanks in advance!

I've never heard Tukey called nonparametric. I don't remember exactly what each of these books says about Tukey, but these three books are all excellent ANOVA books: http://www.analysisfactor.com/resour...oks.html#anova.

Karen