Question regarding RM ANOVA

#1
Dear readers,

I have a question regarding my output from the repeated measures ANOVA.
For my paper I have hypothesized that Dutch, masculine voices in men were more attractive than foreign, masculine voices in men for a short-term relationship.
My results from the three-way RM ANOVA were that the variables group*masculinity*strategy were not significant (P-value= .572).

From the estimated marginal means table, there can be seen that Dutch, masculine voices are more attractive (M=3.6) than the foreign masculine voices on short-term (M=2.6).

My question was I have to report this? Because I have learned that after a not significant interaction you have to interpret the main effects, and not the means from the estimated table from the output.

Can someone help me how to interpret this?
Thanks in advance!

Kind regards,
Naomi
 
#2
Hi, Naomi:

You haven't told us if any of the main effects or two-way interactions were significant. It appears that your primary hypothesis is about a main effect, so if the interactions are NOT significant I would interpret the main effect.

You don't have any covariates (you don't say you did an ANCOVA) and if you have a balanced design, wouldn't the marginal means (from the main effect) be the same as the estimated means?
 
#3
Hi EdGr,

Thanks for your reply, but from the output, only the 2-way interaction from group*masculinity is significant, meaning that Dutch masculine voices are more attractive than the foreign, masculine voices. Also the three main effects are significant, for example the main effect for strategy is significant, meaning that a short-term strategy is more attractive for women than a long-term strategy.

But now I only reported that the three-way is not significant, I already disucssed the main effects, but do I have to give the means also?
 
#4
So if I understand correctly, the group x masculinity interaction is significant. By itself that does not indicate what you wrote. It just indicates that the effect of group (Dutch versus foreign) is different for masculine and non-masculine voices. You could look at the pattern of means to see how the differences went, or do simple effects (as you appear to have done). Did you find that group was significant for masculine voices, but NS for other voices?

You then say that the 3 way is NS, as are the other 2-way interactions. But all 3 main effects are significant, right?

Is your question really just whether to present the means at the 2 x 2 x 2 level? In my view, it depends. If this is the main analysis in your study, it would probably be interesting to readers to see the fine breakdown. But if you have 19 other dependent variables with similar design, it would be too complicated to present them all. Do you HAVE to present all the cell-level means? No. Only enough to show and support the significance tests you report. IMHO!

Interesting that you feel Dutch accents are "not foreign sounding". I know several people who speak English with a Dutch accent, and they sound as foreign as any other accent. The only place a Dutch accent isn't forign is in Holland. Oh wait, are you from the Low Country???