Post hoc in generalized mixed models

Gozdem · Post by **Gozdem** » Mon Aug 29, 2022 8:03 am

Hiya,

I am trying to bring my head around interpreting the posthoc output from generalized mixed models. I am not sure what "difference" exactly means and some numbers are super large! I assume it is due to

one of my factors having a very large effect but since I do not exactly understand what the "difference" score exactly means&how it is calculated (based on marginal means of clustred variables or all of the observations) in posthoc (with bonferroni correction), I am a bit confused. Can someone help please ?

Some short info on my design:
3 factors, with a 2x3x3 design.
1 binomial outcome: 1 and 0
clustered by participants
factors coded with helmert contrast
the model with binomial distribution and logit link function
I am trying to interpret two and three way interactions as well as factor effects that came out from the model via post hoc analysis.
there are no covariates.

Thanks in advance !
G

mcfanda@gmail.com · Post by **mcfanda@gmail.com** » Mon Sep 19, 2022 7:40 am

If you used GAMLj generalized linear model, and selected the logistic model, in the post-hoc table you should see a column labeled as "Exp(B)", and no "Difference". If you see a "Difference" column, you did not select a logistic model. The "Exp(B)" column found for the logistic model is the odd ratio, so it would be the ratio between the two groups' odds.

Gozdem · Post by **Gozdem** » Fri Sep 23, 2022 8:57 am

Thanks for the reply. I actually used generalized mixed models not generalized linear models. Is the "difference" then the difference in the ratio of groups' odds?

jamovi