0

Yes, another question asking if my hierarchical general linear model is written correctly, and if the question I am asking is being tested.

Basically, we have 63 subjects answering 48 (Recog) questions. Each question is split into group one or group two (Format). I want to ask if the effect of Format still holds when the effect of a third variable, Time, is included.

I am doing my analysis in R. I want the variability within Subject to be taken into account, so I include it as a random effect. My model looks like this:

glmer(Recog ~ Time + Format + (1|Subject),family=binomial)

Does this appropriately answer my question? If Format is still significant, can I say that the effect of Format on Recog is still significant when controlling for Time, or is that too far? Maybe significant when including Time?

Thank you!

Mark White
  • 10,252
Anthony Resnick
  • 1

1 Answers1

1

Yes, another MLM question indeed! My favorite.

Your model looks right to me. If Format is still significant in this model, you can say that the effect of Format on Recog is still significant when controlling for Time.

The only additional thing I would suggest doing is including random slopes. Right now, your random effects structure only contains 1, the random intercept. Since Format is measured at Level 1, the effect of Format might be different for each subject. This variation could be modeled by specifying the random effects structure as: (1 + Format|Subject)

You could save the model with that random effects structure as one model and then do a nested model comparison using the likelihood ratio test to see if the random slope is significant. See my post here on how to do that.

If Time is also a Level 1 variable, you could consider using that as a random effect, as well. This would tell you that the influence of Time on Recog varies by person.

Mark White
  • 10,252