rmixed-models

Omnibus effect size r2glmm


I have a dataframe containing four columns:

Participants are crossed with both factors.

Here is a very limited simulated example for structure, not simulating expected (random) effects:

df <- data.frame(ID = rep(1:10, each = 6),
                 F1 = factor(rep(c("A","B"), n = 30)),
                 F2 = factor(rep(c("C","D","E"), each = 2, n = 10)),
                 score = rnorm(60, mean = 4, sd = 0.8))

   ID F1 F2    score
1   1  A  C 4.066147
2   1  B  C 3.613192
3   1  A  D 3.897367
4   1  B  D 3.612968
5   1  A  E 3.157702
6   1  B  E 3.707172
7   2  A  C 2.984475
8   2  B  C 4.488807
9   2  A  D 4.711186
10  2  B  D 3.774878

Considering that this is just an example and that the real dataset is fitted for mixed modelling, I want to look at the F1 * F2 interaction using the lme4 and lmerTest packages, using contrast coding:

contrasts(df$F1) <- c(-1/2,1/2)
contrasts(df$F2) <- matrix(c(-2/3,1/3,1/3,
                           0,-1/2,1/2),
                           ncol = 2)

library(lme4)
library(lmerTest)
model <- lmer(score ~ F1 * F2 + (1 | ID), data = df)

From here, I can look at the results using anova(), giving me the omnibus effects:

anova(model)

       Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
F1    1.59711 1.59711     1    45  2.3893 0.1292
F2    1.17285 0.58642     2    45  0.8773 0.4229
F1:F2 0.39331 0.19666     2    45  0.2942 0.7466

However, when trying to get the effect sizes with r2beta() from r2glmm, I only get the effect sizes of specific contrasts:

library(r2glmm)

r2beta(model)

   Effect   Rsq upper.CL lower.CL
1   Model 0.119    0.435    0.052
2     F11 0.064    0.286    0.000
3     F21 0.042    0.249    0.000
6 F11:F22 0.012    0.181    0.000
4     F22 0.006    0.163    0.000
5 F11:F21 0.005    0.157    0.000

Is there a way to get effect sizes for the omnibus tests using r2glmm? I know it is possible with effectsize and the eta_squared() function, but both packages have a very different approach to compute effect sizes. I would like to stick with r2glmm as much as possible to produce effect sizes in a similar fashion to previous work, but am open to other suggestions.

EDIT: It actually is possible to get omnibus effect sizes with r2glmm when using the GitHub version - the CRAN version has not been updated yet. See https://github.com/bcjaeger/r2glmm/issues/16#issue-1236836438


Solution

  • It actually is possible to get omnibus effect sizes with r2glmm when using the GitHub version - the CRAN version has not been updated yet. See https://github.com/bcjaeger/r2glmm/issues/16#issue-1236836438