We've been wrestling with this problem, not so much at the level of
computing eta-sq, but regarding which of standard eta-sq or partial
eta-sq is generally more appropriate. We came across an interesting
paper, Olejnik & Algina (2003), Generalized eta and omega squared
statistics: measures of effect size for some common research designs,
Psychological Methods, 8, 434-447. What they propose is an algorithm
for figuring out which individual sums of squares are added into the
denominator sum of squares; the result ranges from a value equal to
that used in partial eta-sq to one equal to that used in standard eta-
sq, with possible intermediate values, depending on the design and
upon which effect or interaction you're looking at. The algorithm
depends on the user identifying blocking / measured factors like age,
gender, handedness, and so on, as opposed to measured factors such as
experiment conditions and within-subjects factors. The approach is
based on the twin observations that (1) unlike standard eta-sq,
partial eta-sq allows comparison of similar factors within different
multifactorial designs, and (2) blocking factors can reduce the
subjects-within-cells sum of squares and inflate partial eta-sq. I
wonder if anyone else has considered their generalized eta-sq (or
omega-sq) for actual use instead of standard or partial eta/omega-sq,
or perhaps some other, different approach?
Greg Shenaut
On Jul 20, 2007, at 5:06 AM, kimberley wrote:
> I have run into this same problem, and would be interested in
> knowing how it
> can best be solved.
>
> Thanks,
> Kim
>
>
>
> Veronica Gomez wrote:
>>
>> Dear list members
>>
>> I don't know if this has already been discussed in this mailinglist,
>> but I hope that you can help me out of this: When running MANOVA the
>> output only reports 'Partial Eta Squared' for the main effects and
>> interactions. How do I get SPSS to compute Eta Squared instead? It
>> seems not to be possible through the drop down menues, but is there a
>> syntax command that I can add to my GLM-syntax? Is it even possible
>> or do I have to compute it by hand dividing the Sum of Squares and
>> then multiplicate with 100%?
>>
>> Any help gratefully appreciated!
>> Veronica
>>
>>
>
> --
> View this message in context:
http://www.nabble.com/Compute-Eta-> Squared-in-SPSS-tf3759274.html#a11706945
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
Locked
