using Bayesian statistics to take apart the congruity effect?

labovich

Hi all,

When using congruent, neutral and incongruent trials in a congruity task, one can calculate facilitation (RTneutral-RTcongruent) and interference (RTincongruent-neutral), together facilitation+interference=congruity.

It's important for me to separate facilitation from interference, becasue they might be the result of different cognitive mechanisms. But they do share a variance because they are both related to neutral trials.

Let's say that I want to test for the correlation of facilitation and some continuous measure, while controlling for differences in interference. I think it would be impossible in regular correlation,
but what about Bayesian statistics? is there any way to go around this shared variance of facilitation and interference.

Hopefully it was clear enough

Thanks for reading! any help would be much appreciated.

Tali

EJ

Hi Tali,
I think this is a modeling issue that surpasses the distinction between the statistical paradigms. Suppose facilitation and inference are perfectly correlated, so that people who have large facilitatory effects also have large interference effects. Then it is not clear that you wish to do when you "control" for interference effects. This has been called "throwing the baby away with the bathwater". But if you have three conditions and a continuous variable, and you come armed with a specific hypothesis, then some (perhaps non-standard) modeling could be applied. But perhaps it requires a probabilistic programming language such as JASP or Stan.

Cheers,
E.J.