Stats Question

[cmuhooligan]

---

Members don't see this ad.

So I recently received back a review on an article and one of the reviewers asked me to perform an ANCOVA and I don't believe the rationale calls for it. Here is the situation:

I am testing moderated mediation models, and I have 1 continuous predictor, 1 continuous mediator, 1 dichotomous moderator, and 1 continuous DV.

I ran correlations between demographic variables and the DV and found no significant relationships. Thus, I did not include any covariates in the mediational models, b/c my understanding is that if there is no evidence (via correlations) that these variables covary with one another then you would have no need to include them as covariates.

The other issue is that this reviewer suggested that conducting an ANCOVA is the only way to test for covariates...my understanding is that ANCOVA doesn't test for covariates, per se (as in testing if a variable is a covariate or not) but rather statistically controls for variables when you're interested in group differences on some DV.

So, am I right to object to this, or am I way off base?

 

[ClinicalTrainee]

I think your rationale for not including the demographic variables as covariates is acceptable. However....you said you ran correlations between the demographic vars and the DV, did you also run them with the IVs?

An ANCOVA will give you a statistic for the effect of a covariate on the DV, so the reviewer isn't wrong (other than there are certainly other ways to test for covariates besides an ANCOVA). Also, if you ran correlations for these demographic variables and none were significant, including them in an ANCOVA shouldn't change your results anyway (except by altering degrees of freedom, which could affect significance levels).

I am also wondering how you tested the mediated moderation in the first place, because the answer to that may influence whether or not this ANCOVA is a good idea.

 

[cmuhooligan]

I am also wondering how you tested the mediated moderation in the first place, because the answer to that may influence whether or not this ANCOVA is a good idea.

Well, that's another issue. I used non-parametric bootstrapping to test for the moderated mediation (Preacher, Rucker, & Hayes, 2007). So, running an ANCOVA is not sophisticated enough to answer my research questions. I suppose I could include these demographic variables as covariates in the moderated mediation, but it seems superfluous to do so, given that there is not a significant relationship between them and the DV.

 

[ClinicalTrainee]

Yeah, I think you have a decent argument to not add them, as there is no relationship with the DV. Either say that in your response to the reviews (you don't have to take every suggestion the reviews have, of course) or just run it with them in and note the superfluousness. 🙂