612 HUMAN COMMUNICATION RESEARCH /October 2002
Human Communication Research, Vol. 28 No. 4, October 2002 612–625
© 2002 International Communication Association
Eta Squared, Partial Eta Squared, and
Misreporting of Effect Size in Communication
Research
TIMOTHY R. LEVINE
Michigan State University
CRAIG R. HULLETT
University of Wisconsin-Madison
Communication researchers, along with social scientists from a variety of disciplines, are
increasingly recognizing the importance of reporting effect sizes to augment significance tests.
Serious errors in the reporting of effect sizes, however, have appeared in recently published
articles. This article calls for accurate reporting of estimates of effect size. Eta squared (η
2
) is
the most commonly reported estimate of effect sized for the ANOVA. The classical formula-
tion of eta squared (Pearson, 1911; Fisher, 1928) is distinguished from the lesser known par-
tial eta squared (Cohen, 1973), and a mislabeling problem in the statistical software SPSS
(1998) is identified. What SPSS reports as eta squared is really partial eta squared. Hence,
researchers obtaining estimates of eta squared from SPSS are at risk of reporting incorrect
values. Several simulations are reported to demonstrate critical issues. The strengths and
limitations of several estimates of effect size used in ANOVA are discussed, as are the impli-
cations of the reporting errors. A list of suggestions for researchers is then offered.
N
ull hypothesis testing with standard tests of statistical signifi-
cance have long been the decision rules of choice in most quan-
titative communication research. There is, however, a growing
recognition of the limitations associated with significance testing and p-
values as the sole criterion for interpreting the meaning of results (e.g.,
see Boster, this issue). As a consequence, many communication journals
have adopted editorial policies that require estimates of effect sizes and
statistical power be reported in addition to significance tests. For example,
the current editorial policies of both Communication Monographs (CM) and
Human Communication Research (HCR) require authors to report the effect
sizes for statistical tests. As we argue here, there are good reasons to re-
port estimates of effect size in addition to p-values.
1
Timothy R. Levine (Ph.D., Michigan State University, 1992) is a Professor in the Department
of Communication at Michigan State University. Craig R. Hullett (Ph.D., Michigan State Uni-
versity, 2000) is an Assistant Professor in the Department of Communication Arts at the
University of Wisconsin-Madison. Electronic correspondence can be directed to the first
author at levinet@msu.edu.

Levine, Hullett / EFFECT SIZE 613
The analysis of variance (ANOVA) is one of the most frequently used
statistical analyses in quantitative communication research. Although a
number of different estimates of effect size are available when using
ANOVA (e.g., omega squared, epsilon squared, eta squared; Keppel, 1982),
eta squared seems to be, by far, the most frequently reported. For example,
a brief perusal of recent issues of CM (Volume 66, No. 3) and HCR (Vol-
ume 25, No. 3) showed that ANOVA was used in 6 out of the 10 articles
reporting statistical analyses of data. Eta squared was reported as the es-
timate of effect size in five out of the six of these cases (the sixth case did
not report effect sizes at all). Thus, eta squared is currently a frequently
reported and important descriptive statistic in communication research.
The current authors have become aware of published errors connected
with the reporting of eta squared. Specifically, impossibly large estimates
of eta squared have been appearing in submitted and published reports
of communication research. Although we do not know how widespread
these errors are, we suspect they are quite common and most often are
likely to go unnoticed. We further believe that these errors may be attrib-
utable to a reporting error that we have discovered on the statistical soft-
ware SPSS for Windows printouts. Values labeled as eta squared on (at
least some) SPSS printouts are really partial eta squared (Cohen, 1973),
and consequently, researchers may often be unknowingly reporting par-
tial eta squared values as if they were eta squared. Because partial eta
squared values may, in some cases, be widely discrepant from the values
of omega squared, epsilon squared, and eta squared, these reporting
errors may lead to serious substantive errors in the interpretation of
results. For these reasons, a closer look at eta squared and partial eta
squared is warranted.
To further compound the problem, most popular statistics texts for the
social sciences are, at best, of little help on this issue. For example, al-
though the issue has been recognized for at least 30 years (see Kennedy,
1970), an examination of the more than 20 statistical texts showed that
partial eta squared is mentioned by name in only one (Pedhazur, 1997).
Further, not all texts provide equivalent formulas for the more commonly
mentioned eta squared. For example, Rosenthal and Rosnow (1985, 1991)
provide a formula for eta squared that is equivalent to Cohen’s (1973)
formula for partial eta squared and discrepant from the formulas for eta
squared given in Keppel (1982), Kerlinger (1986), and Kirk (1995). In short,
there seems to be much confusion in the literature regarding eta squared
and partial eta squared.
The primary aim of this paper is to alert communication researchers to
potential errors stemming from the use of SPSS to obtain estimates of eta
squared in ANOVA. As a secondary goal, this paper strives to clarify is-
sues concerning the development and appropriate use of eta squared and