LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous messageNext messagePrevious in topicNext in topicPrevious by same authorNext by same authorPrevious page (April 2002)Back to main SPSSX-L pageJoin or leave SPSSX-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Fri, 5 Apr 2002 10:24:22 -0500
Reply-To:     "David M. Fresco" <>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "David M. Fresco" <>
Subject:      Re: Cohen's f^2 ???
Comments: To: "Phillips, Henry L." <>
In-Reply-To:  <07DFA89F0B7DD411A7780090277389E788B0E2@NOMI-PSFL-EXCH1>
Content-Type: text/plain; charset="us-ascii" ; format="flowed"

Thanks for the quick and helpful reply. You may be on to something, because I am interested in the ES associated with an interaction term. The main effects in the model have zero order correlations with each other and with the DV in the .2 to .3 range, FWIW.

At 9:04 -0600 05/04/2002, Phillips, Henry L. wrote: >David, > >While I don't have a copy of Cohen handy, I have some ideas about why >semipartial correlations are more appropriate indices of IV-DV associations >than are partials. > >If you partial out a variable highly correlated to both your IV and DV, a >high partial r between IV and DV may only mean that a large proportion of >the tiny sliver of DV variance remaining is explained by your IV. The true >overlap between the IV and DV might be tiny, while the partial r describing >the relationship could be enormous. > >A semipartial correlation represents the proportion of total DV variance >accounted for by your residualized IV. It offers the advantage of being a >proportional representation of variance accounted for in the DV. You could >safely interpret differences between the semipartial correlations of two IVs >with a given DV, since all semipartial correlations involving a given DV >represent explained proportions of total DV variance. > >The same cannot be said for the partial correlations of two IVs with the >same DV. Assuming covariates C1...Cn, a strong partial r for predictor X1 >and weak partial r for predictor X2 does not necessarily mean that >residualized X1 has a stronger relationship with Y than does residualized >X2. > >Hope this makes sense. > >Hank Phillips > > >-----Original Message----- >From: David M. Fresco [] >Sent: Friday, April 05, 2002 8:45 AM >To: SPSSX-L@LISTSERV.UGA.EDU >Subject: Cohen's f^2 ??? > > >I have read and re-read Cohen and Cohen (1983) as well as Cohen >(1977, 1988). I am trying to understand his rationale for computing >f^2 as sr^2/1-R^2 as opposed to pr^2/1-R^2. > >It seems to me that partial r is a more pure measure of the >association of the IV of interest to the DV than is the semi-partial >r. > >On page 118 of C&C, there is the equation f^2 as sr^2/1-R^2. >However, on page 154, he seems to discuss Effect Sizes "whether >indexed by sr^2, pr^2, B, or Beta." > > >TIA and please cc my email as I only take the digest. >-- >David M. Fresco, Ph.D. >Department of Psychology Voice: (330) 672-4049 >Kent State University Fax: (330) 672-3786 >315-A Kent Hall >P.O. Box 5190 >Kent, OH 44242-0001 > >ListOwner of Helplessness >

-- David M. Fresco, Ph.D. Department of Psychology Voice: (330) 672-4049 Kent State University Fax: (330) 672-3786 315-A Kent Hall P.O. Box 5190 Kent, OH 44242-0001

ListOwner of Helplessness

Back to: Top of message | Previous page | Main SPSSX-L page