Repeated Measures ANCOVA

MaximusLuminis

I am trying to do an ANCOVA per my advisor's recommendation, but I am not sure how to proceed here:

I have two treatments (A and B). All participants completed both treatments on separate days (within-subjects design). Participants completed the same cognitive test under both treatments once at the start (pre) and at the end (post).

My advisor has now suggested calculating an ANCOVA since the pre-values differ between the two treatments. To do this, I tried to do the following:

1. Open Repeated Measures ANOVA
2. Add both treatments as levels of RM factor 1
3. Add both post-measurements as outcome variables (repeated measures cells)
4. Add both pre-measurements as covariates

Is this correct? I am not sure if this is correct, since here, the pre-measurement of treatment A would also serve as a covariate for the post-measurement of treatment B (and the other way around). This sounds like a problem to me, but I am not sure...

So basically, my question is: Is it possible to do an ANCOVA in a pre-post within-subjects design with the pre-measurement as a covariate?

Thank you in advance!

EJ

Hi MaximumLuminis,

Can you explain the difference in the pre-values through the order in which the treatments are administered? In other words, there may be a training effect, in which case you might consider adding "order" as a co-variate. Anyway, I think what you are doing makes sense...

Alternatively, it seems that you can also view this as a 2x2 ANOVA, with treatment type (A or B) as one factor and treatment effectiveness / time (pre vs post) as the other factor. You could visualize the results by "pre" vs "post" on the x-axis and "A" vs "B" as separate lines. I might be missing something but that seems easier to interpret?!

Cheers,

E.J.

andersony3k

MaximumLuminis,

It seems that you are doing a 2 X 2 repeated-measures ANOVA, but with the addition of two covariates--the two data columns containing the "pre" values. I think what you are doing would be correct if you were not seriously violating any ANCOVA assumptions. However, you might re-consider whether ANCOVA is even a good idea for your dataset. See:

https://www.frontiersin.org/articles/10.3389/fpsyg.2015.00474/full

MaximusLuminis

Thank you both for your responses!