I think you can. Run the same statistical analysis on the residuals. In attachment an example of an ANOVA with R.

I think you need to explain why and what you are going to do with these adjusted values. Generally it is better undertake one overall modelling rather than feeding through stuff from one analysis to another - remember predictor variables specified in a regression are supposed to be measured without error. Otherwise attenuation of the size of the effects will occur. I do not see what continuous variables have to do with it.

I fully agree with Kelvyn. In the example I gave (ANOVA on residuals), this corresponds to an ANCOVA. It is much better to run 1 ANCOVA than 2 ANOVA's. I addapted the R script (in attachment) adding an ANCOVA.

Thank you very much for your answers and for the R file!

As you said, it is better to do one overall model instead of many analyses.

Initially, I wanted to do an ANCOVA, however I think that the design of my work seems to not fit this model.

Indeed, I have two groups (Group A and Group B) and, for the subjects of each group, we assessed memory performances twice:

-> first time during one session (called Session 1),

-> second time during another session, one week after the first one (called Session 2).

Before each memory task, all of our subjects had to complete a scale to evaluate their anxiety level.

So, in this work, I have two pairs of variables for each group:

-> memory performances and anxiety scores for Session 1,

-> memory performances and anxiety scores for Session 2.

I wanted to compute an ANCOVA with two factors: the Group and the Session (Group x Session), and I wanted to add the anxiety scores for each session as covariates.

However, I do not think that I can compute an ANCOVA with these variables because I have two different datasets for anxiety levels (one for each session).

In this case, do you think that I can get the residuals for each memory performances by computing two regressions separately, with the anxiety scores, and then do an ANOVA with the residuals? Or is there another model which allows me to add the anxiety scores for both session?

Thank you in advance!

Stéphane Rehel

I think you can run a repeated measures an(c)ova, i.e. on the residuals over the (between) groups and (within) times, as illustrated in the R script.

Sorry, I made a mistake in the R script. I ran the analysis on the Memory Performance scores instead of on the residuals from Anxiety on these scores. In attachment the corrected script and the image of the interaction on the fabricated data.

I fabricated 'worst-case-scenario' data, in which -in the anova- there are sign. effects of group, time and the interaction on the raw Memory Performance scores; but, when analyzing the residuals, only the time effect remains. This illustrates that it can indeed be necessary to correct for Anxiety.

Hello Jos,

To begin, I would like to apologize because I did not answer since two months... I was working on other analyses and I have just come back on them.

Thank you very much for all of your files and explanations, I think that the repeated measures ANCOVA will fit on my data! I will run these scripts on R and try it!

Thanks!

Stéphane Rehel