Effect size

73 views
Skip to first unread message

Stefano Rastelli

unread,
Aug 17, 2016, 1:23:48 AM8/17/16
to statforli...@googlegroups.com
Hi all

I am about to send a paper to a top ranked journal. In the submission guidelines, they wrote that reporting effect sizes is mandatory. I m using multinomial linear regression (for percentages as outcome variable) and Kruskal Wallis an alternative to anova (my continuous data violate all normality assumptions)

I have no idea of how to calculate effect size for MLR. Any suggestion?

Is there any method to calculate effect sizes for Kruskal?

Thank you in advance

Stefano 

Stefan Th. Gries

unread,
Aug 17, 2016, 10:50:48 PM8/17/16
to StatForLing with R
> I am about to send a paper to a top ranked journal. In the submission guidelines, they wrote that reporting effect sizes is mandatory. I m using multinomial linear regression (for percentages as outcome variable) and Kruskal Wallis an alternative to anova (my continuous data violate all normality assumptions)
Are you sure you're using a "multinomial linear regression" - what is that even?


> I have no idea of how to calculate effect size for MLR. Any suggestion?
If it's a linear regression model you're talking about, you can just use 'raw effect sizes', i.e. differences between (predicted) means or slopes. I know that many people use "effect sizes" to refer to 'standardized effect sizes' but like Baguley, for instance, <subjective>I myself have begun to favor raw ones some time ago</subjective>.


> Is there any method to calculate effect sizes for Kruskal?
See above, I would go with differences between medians etc.

HTH,
STG
--
Stefan Th. Gries
----------------------------------
Univ. of California, Santa Barbara
http://tinyurl.com/stgries
----------------------------------

Matías Guzmán Naranjo

unread,
Aug 17, 2016, 11:23:03 PM8/17/16
to statforli...@googlegroups.com
For percentage data Beta regression is better than linear regression.

--
You received this message because you are subscribed to the Google Groups "StatForLing with R" group.
To unsubscribe from this group and stop receiving emails from it, send an email to statforling-with-r+unsub...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Stefano Rastelli

unread,
Aug 18, 2016, 1:24:17 AM8/18/16
to statforli...@googlegroups.com
I meant multinomial logistic regression, of course. I fond that McFadden pseudo R squared is often used to calculate the effect size, whereas epsilon squared is used for Kruskal Wallis- Any advice?

2016-08-18 4:50 GMT+02:00 Stefan Th. Gries <stg...@gmail.com>:

--

Stefan Th. Gries

unread,
Aug 18, 2016, 2:56:10 PM8/18/16
to StatForLing with R
For multinomial logistic regression, you can also use exponentiated
coefficients, because those correspond to changes in odds ratios.
Reply all
Reply to author
Forward
0 new messages