[FieldTrip] Antw: effect sizes in M/EEG data

Gopakumar Venugopalan venug001 at crimson.ua.edu
Tue Sep 6 14:14:52 CEST 2011


Dear All, Stevens (1996) in his multivariate statistics talks about the use
of  eta squared and partial eta squared for use in F and t tests. He cites
Cohen (1977) while saying this: The partial Eta square (η2) has statistical
benchmarks (small = .01, medium = .06, and large = .14) (Stevens, 1996).

Most stat packages should come with a effect size calculator. In SPSS it is
as simple as checking a box in the GUI.

Like Gregor and Stanley say D2 (also called the Mahalanobis distance)  is
used for multivartiate situations. An earlier edition of Stevens (1980)
comes with a table of power values for 2 to 7 groups means and alpha of .05
and .10.

I believe Gregor is right on the use of this statistic in clinical
situations, but is psych not moving in that direction, as I know most psych
folks (and all grant applications!) ask for those statistics.

Warm regards
gopa



On Tue, Sep 6, 2011 at 1:16 AM, Stanley Klein <dualitystan at gmail.com> wrote:

> Dear Nina,  Gregor is correct. A simple way to think of it is that effect
> size (d) is the distance between two means divided by the standard deviation
> of the data (like IQ). For determining significance one uses t, which is the
> same difference of the means, but now divided by the standard error (the
> standard deviation of the means).  SD doesn't depend on the number of
> trials, but SE does. I often advocate reporting d plus its standard error.
> That way one can avoid reporting an embarrassing naked statistic (a number
> without its error bar).
> Stan
>
>   On Tue, Sep 6, 2011 at 1:21 AM, Gregor Volberg <
> Gregor.Volberg at psychologie.uni-regensburg.de> wrote:
>
>>   Dear Nina,
>>
>> dependend samples t values can easily be transformed into Cohen's d as d =
>> t / sqrt(df) [taken from Rosnow & Rosenthal, Effect sizes for Experimenting
>> Psychologists, Canadian Journal of Experimental Psychology, 2003, 57:3,
>> 221-237; you should find a pdf copy with a search on Google scholar]. For
>> example, a t value of 5.15 with 19 degrees of freedom corresponds to a d of
>> 5.15/sqrt(19) = 1.18. So just compute depsamplesT and divide the resulting
>> vector or matrix by the number of dfs - no further FT functions are needed.
>>
>> My two cents are that effect sizes are especially useful for integrating
>> results across studies where the very same design is applied to different
>> samples, e.g., in clinical trials on pharmaceutical products. In EEG/MEG
>> studies, even when they are on the same topic, there will be very divergent
>> experimenting protocols (timing, task, etc), and also the dependent variable
>> will differ from study to study,  depending on the chosen time points /
>> frequencies / channels of interest. A common effect size metric like Cohen's
>> d does not improve the comparibility much in this case. I would therefore
>> simply report the t
>>
>> statistic along with the corresponding  df and p, and leave it for the
>> interested reader to compute the corresponding d if there should be a need
>> for that.
>>
>> Best regards,
>>
>> Gregor
>>
>>
>>
>>
>> --
>> Dr. rer. nat. Gregor Volberg <
>> gregor.volberg at psychologie.uni-regensburg.de> ( mailto:
>> gregor.volberg at psychologie.uni-regensburg.de )
>> University of Regensburg
>> Institute for Experimental Psychology
>> 93040 Regensburg, Germany
>> Tel: +49 941 943 3862
>> Fax: +49 941 943 3233
>> http://www.psychologie.uni-regensburg.de/Greenlee/team/volberg/volberg.html
>>
>>
>>
>> >>> Nina Kahlbrock <Nina.Kahlbrock at uni-duesseldorf.de> 9/5/2011 2:45 PM
>> >>>
>>
>> Dear statistic experts,****
>>
>> ** **
>>
>> I have a rather general question concerning effect sizes. I understand
>> that t-values give me an approximation of the effect size, as they state how
>> big the difference between two conditions is by taking into account the
>> variance and the number of observations. However, as I recall, effect sizes
>> (like Cohen’s d) are computed in order to find out how important the effect
>> is, independent of the number of observations. However, if I understand the
>> code correctly, dependent samples t-values are computed in a very similar
>> way as Cohen’s d (which is “the” value for effect sizes), both including the
>> variance and thus the number of observations in their functions.****
>>
>> In order to draw conclusions about the importance of the observed effect,
>> is it necessary to compute effect sizes like Cohen’s d or is it sufficient
>> to compute t-values as effect sizes for M/EEG data? If important to compute
>> effect sizes in a way different from t-values, does anyone know of a
>> function that computes effect sizes like that in FT?****
>>
>> ** **
>>
>> Thank you in advance for your answer.****
>>
>> ** **
>>
>> Best Regards,****
>>
>> ** **
>>
>> Nina****
>>
>> ** **
>>
>> - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
>> - - - - - - - - - - - - - - - - - - - - -****
>>
>> ** **
>>
>> Nina Kahlbrock****
>>
>> Institute of Clinical Neuroscience and Medical Psychology ****
>>
>> Heinrich Heine University Duesseldorf****
>>
>> Universitaetsstr.  1****
>>
>> 40225  Düsseldorf****
>>
>> ** **
>>
>> Tel.:      +49 211 81 18075****
>>
>> Fax. .:   +49 211 81 19916****
>>
>> ** **
>>
>> Mail:     Nina.Kahlbrock at med.uni-duesseldorf.de****
>>
>> http://www.uniklinik-duesseldorf.de/medpsychologie****
>>
>> ** **
>>
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl
>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20110906/b96bf083/attachment-0001.html>


More information about the fieldtrip mailing list