Repeated Measures ANOVA Effect Size

Comments

• RaviSelker

  April 2016 edited 4:40AM

  Can you please provide an example. The only reason for them to have the same value would be if you only have one factor (as SStotal = SS + SSerror). I'm also pretty sure that the formula we use to calculate omega squared is the correct one (https://en.wikipedia.org/wiki/Effect_size#Omega-squared.2C_.CF.892). The code to calculate each effect size is as follows:

  eta <- SS / SStotal
  etaPartial <- SS / (SS + SSerror)
  omega <- (SS - (df * MSerror)) / (SStotal + MSerror)
  
• MarieFM

  April 2016 edited April 2016

  Hi,

  I think I agree with JRlakey, unless i'm mistaking something in the formulas... I also use the previous version of JASP 7.1.12, so maybe it has been corrected.

  Here an example of my data in a repeated measures ANOVA:

  • Factor 1 (within): SS= 1212112. Error = 4953057.
  • Between factor: SS= 92859.
  • Interaction: SS = 103498.

  Hand calculation of eta² should be 1212112/ (1212112+92859+103498) = 0,86.

  And partial eta²: 1212112/(1212112+4953057) = 0,197.

  Yet, JASP gives me 0,193 for eta² and 0,197 for partial eta²

  Am I doing the calculation in a wrong way ?

  Thanks for your help.

  Cheers,

  Marie
• RaviSelker

  April 2016 edited 4:40AM

  Hi Marie,

  Thanks for you response. It's definitely not possible for eta squared to be bigger than partial eta squared as SStotal will always at least include SSerror and SS. In your hand calculations you forgot to include the SSerror in SStotal, and you included the SS of the between-subject factor that shouldn't have been included. See for example http://wilderdom.com/research/ExampleCalculationOfEta-SquaredFromSPSSOutput_Mixed.pdf to see how partial eta squared is calculated.

  Cheers,
  Ravi
• MarieFM

  April 2016 edited 4:40AM

  Hi Ravi,

  Thanks for your help.
  Indeed it works better this way, and the hand calculation matches perfectly with JASP.

  Cheers,
  Marie