To provide you with the context, we perform these statistical tests following the realization of an experiment in which we wanted to know the impact of two feedbacks, one visual and the other visual and tactile, on the evolution of performance in brain-computer interfaces. These interfaces in particular make it possible to control different digital applications by performing only mental imagery tasks, e.g., imagining an object in rotation, mental calculation, etc.
Our participants (n = 16) learned to use a brain-computer interface with visual feedback in 5 sessions, and visual and tactile feedback in 5 other sessions. The order of presentation of these feedback is randomized. We also want to know the influence of the order of presentation on the performances.
We obtained 10 BCI performance measures, 5 per feedback. It turns out that the groups formed by the feedback modality with which the participants had started to train with, had different initial mental imagination abilities (measured using a specific test), i.e., "KVIQ_S1_ImS". Mental imagination can have an impact on BCI performances. Therefore, we want to take them into account in our analysis.
Using SPSS I performed a 3-way repeated measure ANCOVA with the session, the feedback modality as intra-subject independent variables and the order of presentation as inter-subject independent variable, the BCI performance as dependent variable and the initial mental imagination abilities as covariate.
The table reporting intra-subject effects reports one line for each single and interactive effect with another line reporting the impact of the covariate on this effect. I thought that the results of the table of intra-subject effects already took into account the influence of the covariate. Do they not? If not, why are the results of the single and interactive effects different from the ones of this same analysis without the covariate?
Do you know how to interpret these results? Should we first look at the line reporting the influence of the covariate before looking at the differences obtained compared to a covariate-free ANOVA?
Sorry for the length of this description and if the latter is not clear enough. I’d be happy to provide you with more information if needed.
Thank you very much for all the help you will be able to provide !
Yes, the intra-subject effect reported has taken into account effects of the covariable. I think your confusion arises from that you see the intra-subject effects as those without being affected by the covariate, and the interaction effects beween intra-subjects and the covariate as the impact to the effects, which is not correct. When you add a covariate, the intra-subject effects become the ones that can not be explained by the variance casued by the main effect of the covariate. For the other question, if the adding of a covariate is reasonable based on previous theories/findings, you don't need to compare both models. It is usually in the situation when you find a main effect but need to rule out the possiblity that it can be explained by a covariate that you do both.
Thank you very much Guanmin for your answer ! I am in the second situation where I found a main effect but need to rule out the possiblity that it can be explained by a covariate.
I have a last question to be sure that I understood things correclty, if you do not mind. As I understand it, I first need to check if the interaction with the covariate is significant. Only if it is, I check if the effect of the main effect is different from the one found without the covariate. Is that correct?
Thank you again for you help.
It depends on your hypotheses. It seems that you knew from previous theories/findings that you needed to add a covariate: "Mental imagination can have an impact on BCI performances." In that case, you only need to do analysis with the covariate. If that is not the case, you may want to do one without the covariate frist to see whether the intra-subject effects are significant, and then do the one with the covariate to see whether the effects still hold, especially when you find a significant main effect of the covariate.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA