I'm doing my analysis with R 4.0.2 and was using an linear model with the lm() function (default setting) and omega squared as a measure of effect size (omega_sq() function).
After reading the following paper :Article Regression Models for Log-Normal Data: Comparing Different M...
As my data fits the criteria, I decided to try the general linear model with a gamma distribution an log link method. I was able to fit the model with the glm() function, but I'm not able to use the omega_sq() function on it and I am not quite sure that this is the correct effect measure for this approach (from the documentation I read it seems like omega_sq() coerces the model to an ANOVA).
Any insights on what the correct effect size measure for this model is and what R function can I use to calculate it?
This is a complicated issue. Omega squared is a standardised effect size measure that estimates the proportion of variance in the population accounted for by your model. It is therefore similar to R^2 which is the proportion of variance in your data (not the population) accounted for by your model. However, both assume a linear model in which the variance accounted for is a simple additive function of the predictors.
This might not be true. In particular if you have a model in which the variance explained and total variance are a function of the predictors in the model and the values you take then simple quantities like R^2 or omega^2 can't be computed. As you have selected a generalized linear model it seems likely that simple variance explained statistics can't be computed. In these cases there are a bunch of pseudo-R^2 like measures than have been proposed. None will capture all the properties of R^2 (or omega^2) but they might be useful in some contexts.
See the attached for a discussion of some popular pseudo-R^2 measures (but note there are more recent papers on this).
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling