R-Square for Individual Fixed Effect Estimates in Proc Mixed

[bigjer]

Hello, 

I am trying to get % variance explained for the fixed effects from my proc mixed code for a mixed linear regression. Unfortunately, I do not know how to request this in my output. Here is an example of my current code:

Proc Mixed Data=Set covtest;
Class ClassVar;
Model Outcome = Predictor1 Predictor2 Predictor3/solution ddfm=kr;
Random intercept/subject=pedid;
Run;

If there is anyway to get this information in my output that would be great. 

Thanks, 

Jer

[PaigeMiller]

There is no such thing as an R-squared value for fixed effects. You get an R-squared value for an entire model being fit (and as far as I remember, you can't get this from PROC MIXED).

[StatsMan]

Paige is correct.  There have been pseudo r-squares proposed for mixed models, but none have really been accepted in the statistical literature.  If someone has a reference for a good one, post it here.  

You can get r-squares for effects if you use type 1 statistics in a GLM, but again these r-squares are not available in a mixed model.

Random effects are a different story.  There have been ways proposed to evaluate the relative variation explained by each random effect in the overall error variance, if the covariance structures involved are simple.   

[Kip1]

All,

I want to resurrect this discussion of R-square for mixed effects models, in reference to a number of recent discussions that have surfaced re implementation of Nakagawa and Schielzeth 2013 http://dx.doi.org/10.1111/j.2041-210x.2012.00261.x   'marginal' and 'conditional' R-squares for GLMMs in the MuMIn package in R https://cran.r-project.org/web/packages/MuMIn/MuMIn.pdf

Is there a way to compute similar values via SAS Proc Mixed? If so, I'd like to see some code examples. The only SAS reference I have found is: Rsquare = 1-(SSResidual_Model/SSResidual_InterceptOnlyModel) which can be computed in Proc Mixed and given the interpretation "percent reduction in variance due to model'. However, I don't know if or how this relates to Nakagawa and Schielzeth.

Any (new) thoughts?

[PaigeMiller]

---

@Kip1 wrote:

All,

I want to resurrect this discussion of R-square for mixed effects models, in reference to a number of recent discussions that have surfaced re implementation of Nakagawa and Schielzeth 2013 http://dx.doi.org/10.1111/j.2041-210x.2012.00261.x   'marginal' and 'conditional' R-squares for GLMMs in the MuMIn package in R https://cran.r-project.org/web/packages/MuMIn/MuMIn.pdf

Is there a way to compute similar values via SAS Proc Mixed? If so, I'd like to see some code examples. The only SAS reference I have found is: Rsquare = 1-(SSResidual_Model/SSResidual_InterceptOnlyModel) which can be computed in Proc Mixed and given the interpretation "percent reduction in variance due to model'. However, I don't know if or how this relates to Nakagawa and Schielzeth.

Any (new) thoughts?

---

Your first link doesn't work. The second link (to the R package MuMin) doesn't really explain what MuMin does. So, no new thoughts from me.

[Kip1]

Try this https://doi.org/10.1111/j.2041-210x.2012.00261.x

"A general and simple method for obtaining R^2 from GLMMs" Nakagawa and Schielzeth 2013 Methods Ecol. Evol 4: 133-142

[PaigeMiller]

Okay, that works. But as far as I know, none of this is programmed in SAS, you'd have to program it yourself, or get the R-code to work.