Date: Thu, 4 May 2000 17:46:50 -0500
Reply-To: "Nichols, David" <nichols@SPSS.COM>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Nichols, David" <nichols@SPSS.COM>
Yes, there is a problem with the way the effect size measure is reported and
some ambiguity in the way it's documented in the online Help in Release 9.0.
The measure is indeed a partial eta squared, the same measure we report with
the default UNIQUE sums of squares in the older MANOVA procedure. The output
simply labels it as "Eta Squared."
In the 9.0 Help looking under eta squared in the Index one finds: "Estimates
of effect size gives a partial eta-squared value for each effect and each
parameter estimate. The eta-squared statistic describes the proportion of
total variability attributable to a factor." The right button popup from the
dialog in 9.0 reads: "Partial eta squared. Eta squared is interpreted as the
proportion of the total variability in the dependent variable that is
accounted for by variation in the independent variable. It is the ratio of
the between groups SS to the total SS."
The output is mislabeled; it needs to have the word "Partial" added to the
front of "Eta Squared." This has actually been reported in this way since
the GLM procedure was implemented in Release 7.0. I didn't hear a complaint
about it until after Release 9.0, and at that time I filed bug reports
against both the output labeling and the Help system. The output labeling
was not fixed for Release 10.0, but is now fixed for the upcoming 10.0.7
release. The popup Help was changed for Release 10 to read: "Partial eta
squared. Partial eta squared is the ratio of the variation accounted for by
an individual independent variable (SSH) to the sum of the variation
accounted for by the independent variable and the variation unaccounted for
by the model as a whole (SSH+SSE)."
The manuals have been correct and very clear that the only effect size
measure being offered is a partial eta squared going all the way back to the
SPSS Advanced Statistics 7.0 Update manual. We apologize to anyone misled by
the incorrect labeling and ambiguous Help definitions.
Principal Support Statistician and
Manager of Statistical Support
> -----Original Message-----
> From: William P. Eveland, Jr. [mailto:eveland@SSCF.UCSB.EDU]
> Sent: Tuesday, May 02, 2000 3:02 PM
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: [SPSSX-L]
> I just read an article, to be presented next month at the
> Communication Association annual meeting, that states that SPSS 9
> misrepresents eta squared in the GLM procedure. That is,
> while the online
> help correctly describes the calculation of the concept "partial eta
> squared" and describes it as such, the output labels the coefficients
> simply "eta squared." Calculations by the authors suggest
> that SPSS is
> really reporting partial eta squared in this procedure.
> This, the authors
> point out, had lead to numerous misreports of effect size in
> journals, and surely in other fields as well, since eta
> squared and partial
> eta squared are considerably different measures of effect
> size that should
> be interpreted differently. In the article, this problem was
> identified in
> SPSS 9.
> I'm about to present an article at this same conference using the eta
> squared effect size estimates from GLM, and I'd like to know:
> (a) has this
> problem been remedied in later versions of SPSS; and if not,
> what are the
> thoughts of others on the list, and from the folks at SPSS?
> The article to which I'm referring is:
> Levine, T. R., & Hullett, C. R. (2000, June). Eta-square, partial
> eta-square, and misreporting of effect size in communication research.
> Paper prepared for presentation at the annual meeting of the
> Communication Association, Acapulco, Mexico.
> Thanks for your input.
> William "Chip" Eveland
> Director, Benton Survey Research Laboratory
> & Assistant Professor, Department of Communication
> University of California
> Santa Barbara, CA 93106
> Email: firstname.lastname@example.org
> Voice: (805) 893-7223
> Fax: (805) 893-7102