Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Supported by

Repeated measures ANCOVA - post-hocs ?

edited May 2016 in JASP & BayesFactor

Dear Jasp community,

I was wondering if it is possible to do post-hocs on repeated measures ANOVA with a covariate.

Here are my analysis :

I have a within factor (A, 2 levels), between factor (B, 3 levels) and a covariate (C, continuous).

  • When I do the repeated measures Anova (A~B), I have no effect of factors A and B, and no AxB.
    But...

  • When I control with the covariate C, my within factor A gets significant, and the rest is non-significant.

I don't know how to interpret those results and what kind of post-hocs analyses I should do.

Thanks a lot,

Best regards,

Juliane

Comments

  • EJEJ
    edited 8:33PM

    Hi Juliane,

    You have to be careful with covariates. Suppose you have different groups and you suspect your DV is influenced by IQ. You have measured IQ and the distribution of IQ is approximately the same in the different groups. In that situation you can add the covariate. But now suppose IQ differs between the groups (e.g., one group is normal controls, the other is AD patients). Adding IQ as a covariate is now problematic, because the effect of group is confounded with the covariate. So if you are in that situation, just don't add the covariate.

    Suppose you can add the covariate. Did you think of this before you saw the data? When you explore the results by adding one or more covariates you can no longer interpret the p-value. I would just be honest about this and indicate the results both ways (in fact this is what Simonsohn et al also recommend -- reporting the analysis both with and without the covariate; this makes sense to me)

    Cheers,
    E.J.

  • edited 8:33PM

    Hello,
    Thanks for your very quick reply.

    The covariate seems to be equally distributed between groups so I think I am not in the situation you describe.

    I will definitely report both analyses. If I decided to run an ANCOVA, it is because I ran correlations between C and my 2 levels of factor A (let's say A-1 and A-B) and I found that A-1 correlated with C, but A-2 did not. This would actually be congruent with the rest of my analysis and expected.
    I thought that maybe the ANCOVA would say the same think but in a different way.

    Thanks,

    Juliane

Sign In or Register to comment.