mixed model – Which test statistic to report for (G)LMMs with categorical variables with >2 levels and/or interactions?

When building a glmm model with a binary outcome variable in R using the lme4 package I usually use something like the following structure:

model_1 <- glmer(binary_value ~ condition + variable + continuous_cov + (1 | ParticipantID), family = binomial, data = df)

model_interaction <- glmer(binary_value ~ condition + variable + continuous_cov + condition*variable + (1 | ParticipantID), family = binomial, data = df)



anova(model_1, model_interaction)

This gives me the test statistics (z value, or t-values for LMMs), standard errors, confidence intervals and p-values for the random effects and continuous predictors, or predictors with only 2 levels. However for predictors with >2 levels I am somewhat confused about what values to report. Should I report the LRT value given by drop1? If so, how should I denote this in text? As LRT or as Chi^2 ? I could also use anova(model_1), which gives me an F value, but I am pretty sure this is wrong as it doesnt account for the random effects. Or is there another “global” test which specifies the fixed effects of categorical variables with more than 2 levels?

Similarly for the interactions, is reporting the LRT given by drop1 correct? I then normally would compare the models using AIC, log likelihood values etc.

Read more here: Source link