Date: Thu, 8 Sep 2005 09:53:36 +0930
Reply-To: kylie.lange@flinders.edu.au
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Kylie Lange <kylie.lange@flinders.edu.au>
Subject: Re: About effect sizes
In-Reply-To: <6803.1126102652@www6.gmx.net>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Hi Karl,
Another reference you may want to check out:
Charles A. Pearce, Richard A. Block & Herman Aguinis, Cautionary Note
on Reporting Eta-Squared Values from Multifactor ANOVA Designs.
Educational and Psychological Measurement, Vol 64(6), pp 916-924, Dec 2004.
Kylie.
On 7/09/2005 11:47 PM, Karl Koch wrote:
>Hello Paul,
>
>thank you for the information. I will follow the links and have a look to
>the tools as well. I will also see if I can hold of this book. I have
>general problems with the libraries here in order to get fast and cheap
>access to books of this magnitude. This is my mean reason, why I look for
>papers and examples or documentens that are freely available (e.g. written
>for tutorials at universities).
>
>My general problem here is really how to calculate effect size values. I
>have already calculated partially eta squared (with SPSS), eta squared and
>Cohen's f value for my design. However, I stuck at the point where I want to
>know how much do factorial interactions affect the value of the main
>effects. Also on the decision which effect size would be "best" (if this is
>not a research question by itself)...
>
>I would like to make an example:
>
>I have three factors which are all significant. They all have non-trivial
>(Cohen f / eta squared / partically eta squared) effect sizes (which is
>good). However, I also have ongoing interactions between the effects.
>Natuarally I wouild assume that I need to correct somehow those effects
>sizes of the main effects in order to arrive at the real values that
>represent the strengh of the factor. Those values would then allow to
>estimate the impact of each of the three factors on the DV. I could then
>say, for example, factor 1 was the strongest of my factors in terms of
>increasing the score in my DV, factor 3 was second and factor 2 was third.
>etc.
>
>What really could help here would be a paper that used a factorial design
>(preferably with more than two levels and more than two factors, but not
>necessary) that applied effect sizes.
>
>Kind Regards,
>
>
>
>
>
>
>
>>--- Ursprüngliche Nachricht ---
>>Von: Paul Ginns <p.ginns@itl.usyd.edu.au>
>>An: SPSSX-L@LISTSERV.UGA.EDU
>>Betreff: Re: About effect sizes
>>Datum: Tue, 6 Sep 2005 14:28:47 +1000
>>
>>Hi Karl,
>>
>>a few suggestions for ANOVA effect sizes:
>>
>>Cohen's f is one effect size you can use, but there are others. The
>>standardised mean difference (Cohen's d) can be scaled up to factorial
>>
>>
>designs.
>
>
>>Freeware which can be used to produce factorial d values, and confidence
>>intervals around these d values, is available at
>>http://www.psy.unsw.edu.au/research/PSY.htm and Kevin Bird has written a
>>
>>
>book about this approach to
>
>
>>ANOVA called
>>
>>Bird, K.D. (2004). Analysis of variance via confidence intervals. London:
>>Sage Publications.
>>
>>Note that the above webpage also includes some SPSS syntax for calculating
>>studentized maximum root product interaction contrasts.
>>
>>Another effect size is Epsilon squared. Jim Jaccard's Zumastat programme
>>www.zumastat.com has a module that allows calcualtion of e-sq. and its CI
>>using F values and degrees of freedom. (Note: I have no commercial
>>
>>
>interest
>
>
>>in Zumastat.)
>>
>>cheers,
>>
>>Paul
>>
>>
>>
>>Date: Mon, 5 Sep 2005 16:27:02 +0200
>>From: Karl Koch <TheRanger@gmx.net>
>>Subject: About Effect Sizes...
>>
>>Hello group,
>>
>>I have done a 3x3x2 factorial experiement which has significant main
>>effects
>>and interactions. The DV was a score between 1 (not useful) and 6 (very
>>useful) on 6 levels. To know more about the magnitude of the three
>>factors,
>>I was thinking of exploring the possibility to apply effect sizes. This
>>would also spice up the experiment documentation.
>>
>>I found a, in my oppion good article in the Information Technology,
>>Learning, and Performance Journal which is available online
>>at:
>>
>>
>>
>>
>http://education.nyu.edu/alt/beprogram/osrajournal/kotrlikwilliamsspring2003.pdf
>
>
>>The article states that for Analysis of Variance (which applies to me) I
>>should use Cohen's f effect size measure. Now my questions:
>>
>>1) Can somebody here confirm that or does somebody here know other books
>>or
>>articles (preferable also available online) which can help me to make a
>>decent decision on that? I know that the entire topic is pretty strong
>>discusses and some people disagree on what to choose. However, I am
>>looking
>>for a decent narrative what to use. Some examples (papers?) where people
>>have used effect sizes in similar experiments would be helpful, too.
>>
>>2) I have not only significant main effects but also significant
>>interactions. How does this influence the meaning of the effect size
>>measures of the meain effects.
>>
>>3) If I want to find out the following: What is the most influcial factor
>>(the one that influenced the DV most) ? Can I use the effect size measure
>>to
>>determine that e.g. by determining that by choosing the main effect with
>>the
>>greatest effect size?
>>
>>4) Is there any other way to rank the three factors according their
>>importance (in terms of influencing the DV by increasing the score)?
>>
>>Any help would be greatly appreciated.
>>
>>Kind Regards,
>>Karl
>>
| 
