FAQ/Escomp - CBU statistics Wiki

Revision 60 as of 2012-07-19 15:42:55

Clear message
location: FAQ / Escomp

Computing effect sizes

  • Cohen's d which is used for t-tests may be computed [http://www.uccs.edu/~faculty/lbecker/ with a calculator] or using free [http://www.swin.edu.au/victims/resources/software/effectsize/effect_size_generator.html PC downloadable software.] This can also be calculated in EXCEL (see [:FAQ/mse: here]).

  • SPSS computes partial eta-squared, $$\mbox{Partial } \eta^text{2}$$, on request using ANOVAs. If using General Linear Model>univariate or General Linear Model>Repeated Measures click options and select Estimates of Effect Size. An extra column in the outputted anova tables is produced showing partial eta-squareds of terms in the anova table. Partial eta-squared represents the proportion of variance not attributable to any of the other observed factors which is explained by the factor of interest.

  • An alternative for anovas, Eta-squared ($$\etatext{2}$$), can also be calculated for anovas. $$\etatext{2}$$ is defined as the sum of squares for a particular effect divided by the total of all the sums of squares of effects in the analysis of variance table. [http://www.jalt.org/test/bro_28.htm It is suggested], however, that partial eta-squared be used for repeated measures analysis of variance and eta-squared for between subjects anovas (which feature just one error term in the anova). [:FAQ/rField:Field(2005) advocates] only using effect sizes when comparing a difference between two groups in repeated measures anova.

  • [:FAQ/power/prop1sn:An EXCEL spreadsheet calculator] computes the one sample chi-square effect size measure, $$\omega$$.
  • The Pearson correlation is, itself, an effect size.
  • Field (2005)(pp. 222-223) suggests evaluating a correlation based upon output from a logistic regression. This is based upon the Wald statistic [:FAQ/infmles:which can give misleading] results.
  • Field also suggests a nonparametric effect size for comparing two groups [:FAQ/nonpz: (see here).]

You might find [:FAQ/effectSize:A guide to magnitudes of effect sizes] and [http://www.depts.ttu.edu/aged/effect%20size.pdf Calculating, Interpreting and Reporting Estimates of "Effect Size"] useful. The following [attachment:effconvert.xls spreadsheet] by Jamie DeCoster converts a single effect size, such as Cohen's d, to several others including Odds Ratios using references mentioned in the spreadsheet. One could also convert a partial eta-squared to a Cohen's d by regarding the partial eta-squared as a squared correlation. It follows square rooting the partial eta-squared and entering it in Jamie's spreadsheet as a r will then allow you to read off the Cohen's d. Jamie has written other EXCEL spreadsheet calculators [http://www.stat-help.com/spreadsheets.html here.] Howell (2013) p.627-8 gives formulae for conversions of effect sizes such as odds ratios and correlations to Cohen's d.

References

Baguley T (2012) Serious Stats. A guide to advanced statistics for the behavioral sciences. Palgrave MacMillan:New York. R code and formulae for a range of commonly used effect sizes are in Chapter 7 on pages 235-276.

Field A (2005) Discovering statistics using SPSS Sage:London.

Howell DC (2013) Statistical methods for psychologists. 8th Edition. International Edition. Wadsworth:Belmont,CA.

Vacha-Haase T & Thompson B (2004) How to estimate and interpret various effect sizes. Journal of Counseling Psychology 51(4) 473-481. Details computing effect sizes in SPSS for methods including ANOVA and regression as well as showing conversion formulae expressing one effect size in terms of another.