If you have a 2x2 factorial design evaluated with a mixed effects model, is there a method to estimate the r-squared of the independent factors and interactions?
I have found methods to estimate the r squared of the full model, the fixed factors of the whole model, and the random factors of the model. However, I would like to know the contribution of the factors independent of the interaction.
You could simply leave out the interactions from the model and just fit the main effects, but I am probably missing something vital in your question.
You may also want to look at the answers that I got to my question; there were some useful refences:
https://www.researchgate.net/post/Recent-developments-in-pseudo-R-squared-statistic-for-logit-multilevel-models
Further to Kelvin's post I think its worth emphasizing that there is no such thing as R^2 for a multilevel model (except in simple cases where there is equivalence to standard linear models).
For any model where variance is not constant across values of x (predictors) or sets of predictors there is no version of of R^2 that has all the properties of classical R^2. Hence these statistics are usually referred to as pseudo-R^2. So the 'correct' choice depends on the properties of classical R^2 that one wants to preserve.
You might be able to adapt partial eta-squared for your situation. It can be calculated from the F value, degrees-of-freedom for the effect, and degrees-of-freedom for the residuals. Alternatively, it can be calculated from the sum-of-squares for the effect and the sum-of-squares for the residuals. I don't know that I've ever seen this in this application, and it may depend on what information is available in your output.
- Badges
- Science topic
- Similar topics
- Biological Science
- Histology
- Fixatives