effect sizes?

[jellie1410]

I wondering how one normally would obtain effect sizes (e.g., partial eta) for individual variables in multilevel models? Given that MLwiN does not provide them, I struggle with how to include them in my papers when running my models in MLwiN.
Thanks for your help.

[billb]

Effect sizes come in many shapes and sizes and are often some function of the parameters of a statistical model or even simply the parameters themselves - for example Cohen's d is a measure based on the parameters for a difference scaled by standard errors and is very similar to a standardised regression coefficient when formulated as an intercept and group difference. MLwiN gives the equivalent of regression coefficients for the multilevel model and these could clearly be scaled (by dividing by the ses) to be standardised. Most measures however like Cohen's d (or indeed the partial eta (squared?) which you refer to) are developed with specific models in mind and not usually multilevel models.
Hope this helps a little.
Bill.

[jellie1410]

Thanks! Just one follow up: would you advice me to therefore report standardized beta's in a manuscript as reflecting the size of the effect?

[billb]

Hi jellie1410,
I guess your best bet is to look for examples where multilevel models have been used in the journal in question and copy the conventions used. Some journals simply have tables with coefficients, some with binary responses use odds ratios etc.