rlme4p-valuesjplot

Why do p-values from sjPlot's tab_model() and Anova() not match for LMM output?

I have a question regarding p-value estimations for an lme4 LMM model output:

I compared p-values produced by the functions tab_model and Anova and they don't seem to match. When I run Anova, I get significant results for a fixed effect I don't get significant results for when running tab_model. I'm aware computing p-values for LMMs is a controversial topic, but I would like to do this nonetheless.

I ran tab_model with both the Wald and the Satterthwaite approximation for the argument df.method, and with setting p.adjust = NULL so I get uncorrected p-values, but it still produces different p-values than Anova.

My code looks like this:

tab_model(my_lmm,
          seed = 42,
          show.se = TRUE, 
          show.stat = FALSE,
          show.intercept = FALSE, 
          show.df = FALSE,
          show.re.var = FALSE,
          show.icc = TRUE,
          show.obs = FALSE,
          p.style = "scientific",
          p.threshold = c(0.05),
          p.adjust = NULL,
          df.method = "satterthwaite",
          digits = 6, 
          digits.p = 3)

Anova(my_lmm)

Thanks in advance for your help!

Merle

Solution

  • By default, Anova() uses type = "II". If you set type = "III", it produces the same results as tab_model. So instead of running Anova(my_lmm), the code should look like this: Anova(my_lmm, type = "III").

    Here's an explanation concerning these different ANOVA types: https://md.psych.bio.uni-goettingen.de/mv/unit/lm_cat/lm_cat_unbal_ss_explained.html