Timeline for Which R functions are correct for estimating partial Eta Squared for effects in multiple linear regression with interaction terms?
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 24, 2022 at 7:40 | comment | added | Tom Wenseleers | Yes indeed effectsize::eta_squared(fit) and effectsize::eta_squared(anova(fit)) both return type I (which is not very useful) but effectsize::eta_squared(Anova(fit, type=2 or 3))) can be used to get partial eta squared with type II or type III tests. Only thing is if you use type III that you have to remember to use sum contrasts / effecting coding & center all covariates. A function effectsize::eta_squared(fit, type=XX) could have been more logical... | |
| Oct 24, 2022 at 7:30 | comment | added | dipetkov | @TomWenseleers It sounds like I should look into the effectsize as well. Makes sense that a package specially for effect size calculations provides more options. I've used DescTools before, so that's why I went with it. Also, on the models I tried it on, the effectsize::eta_squared default is type 1, which doesn't make sense to me. | |
| Oct 24, 2022 at 7:10 | history | bounty ended | Tom Wenseleers | ||
| Oct 24, 2022 at 7:10 | comment | added | Tom Wenseleers | Many thanks - makes sense! I had suspected it had to do with using different types of SSs, but hadn't dug deep enough in the documentation! effectsize::eta_squared() is the only option available that also returns confidence intervals on the partial eta squared. To get partial eta squareds using type 2 or type 3 SSs one can apparently use effectsize::eta_squared(Anova(fit, type=3))). But one has to use effect coding using afex::set_sum_contrasts() and center all covariates using scale() in the mode formula. | |
| Oct 23, 2022 at 0:18 | history | answered | dipetkov | CC BY-SA 4.0 |