Proportion of explained variance in a mixed-effects model

Question

I do not know if this has been asked before, but I do not found anything about it. My question is if anyone can provide a good reference to learn how to obtain the proportion of variance explained by each one of the fixed and random factors in a mixed-effects model.

Answer 1

I can provide some references:

Xu, R. (2003). Measuring explained variation in linear mixed effects models. Statistics in Medicine, 22, 3527-3541. DOI:10.1002/sim.1572

Edwards, L. J., Muller, K. E., Wolfinger, R. D., Qaqish, B. F., & Schabenberger, O. (2008). An $R^2$ statistic for fixed effects in the linear mixed model. Statistics in Medicine, 27, 6137-6157. DOI:10.1002/sim.3429

Hössjer, O. (2008). On the coefficient of determination for mixed regression models. Journal of Statistical Planning and Inference, 138, 3022-3038. DOI:10.1016/j.jspi.2007.11.010

Nakagawa, S., & Schielzeth, H. (2013). A general and simple method for obtaining $R^2$ from generalized linear mixed-effects models. Methods in Ecology and Evolution, 4, 133-142. DOI:10.1111/j.2041-210x.2012.00261.x

Happy reading!

Answer 2

According to this blog post from 2013, the MuMIn package in R can provide R$^2$ values for mixed models ala an approach developed by Nakagawa & Schielzeth 2013$^1$ (which was mentioned in a previous answer).

#load packages
library(lme4)
library(MuMIn)

#Fit Model
m <- lmer(mpg ~ gear + disp + (1|cyl), data = mtcars)

#Determine R2:
r.squaredGLMM(m) 

       R2m       R2c 
 0.5476160 0.7150239  

The output for functionr.squaredGLMM provides:

• R2m: marginal R squared value associated with fixed effects

• R2c conditional R2 value associated with fixed effects plus the random effects.

Note: a comment on the linked blog post suggests that an alternative Nakagawa & Schielzeth inspired approach developed by Jon Lefcheck (using the sem.model.fits function in the piecewiseSEM package) produced identical results. [So you have options :p].

• I did not test this latter function, but I did test the r.squaredGLMM() function in the MuMIn package and so can attest that it is still functional today (2018).

• As for the validity of this approach, I leave reading Nakagawa & Schielzeth (2013) (and follow-up article Johnson 2014$^2$) up to you.

---

^{1: Nakagawa, S., and Schielzeth, H. 2013. A general and simple method for obtaining R2 from generalized linear mixed‐effects models. Methods in Ecology and Evolution 4(2): 133-142.}

^{2: Johnson, P. C. D. 2014 Extension of Nakagawa & Schielzeth’s R2GLMM to random slopes models. Methods in Ecology and Evolution 5: 44–946.}