ANOVA with random factor
Hello JASP-ers,
For my study I conducted Bayesian univariate ANOVA with fixed factors and subject as random factor. That worked fine, but I am asked to report the classical ANOVA results too. However, I see that the classical univariate ANOVA in JASP do not have the option for "random" factor, but has WLS. I am definetly no expert how to properly use WLS in regressions (or ANOVAs) and my question might be quite naive, but I was wondering why you opted for the WLS instead of the random factor option, and can the WLS be used in such a way (e.g., subject as WLS)?
Thank you in advance,
Mila
Comments
Hi Mila,
I'll ask our expert.
Cheers,
E.J.
Hi Mila,
In the Bayesian ANOVA, these terms have a bit of a different meaning - this is from the JASP help file:
The difference here is that the random factors are added to the null model by default (and which interaction terms are added to the model by default), and are assigned a slightly wider prior distribution by default (r scale parameter set to 1 instead of 0.5, which is the setting for fixed factors). Aside from this, they are analyzed the same. For an illustration, see the example below, where I get the same results for 2 analyses: one where I added "facFive" as a random factor, and one where I added "facFive" as a fixed factor. I then set the prior distributions for random and fixed to be the same, and added identical model terms for both analyses:
As for adding WLS - this is a very different concept from random/fixed. By specifying these weights, you are assigning more weight to certain observations. For instance, if you set the WLS weight for one specific observation to 2, and all others to 1, you will count that one observation as having occurred twice. I have to say I have never used these weights myself, and them seem to be a fairly exotic tool as far as I know.
Cheers,
Johnny
Hi Johnny,
Thank you very much for this detailed and very clear explanation!
All the best,
Mila