[EMAIL PROTECTED] wrote: > Dear Paige, > > Would you mind explaining how widely differing cell sizes in an anova reduce > power?
Actually someone did the work for me ... he sent me an e-mail with simulated examples showing that the power does indeed decrease when the sample sizes differ; and the wider the difference the greater the decrease in power. Alas, I deleted it. I hope that person will post his results here instead of in a private e-mail. I don't have a proof that the power decreases (although there probably is one somewhere). This is just the "standard wisdom" that I have come to understand with respect to ANOVA. The common sense layman's explanation is that your best chance of detecting differences is if you have highly precise estimates of all of your cell means ... if one of those cells has very few data points compared to the others, your estimate of that particular cell mean is much less precise, your ability to detect differences is less and therefore you have lower power. -- Paige Miller [EMAIL PROTECTED] http://www.kodak.com "It's nothing until I call it!" -- Bill Klem, NL Umpire "When you get the choice to sit it out or dance, I hope you dance" -- Lee Ann Womack > "Paige Miller" <[EMAIL PROTECTED]> wrote in message > [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > >>[EMAIL PROTECTED] wrote: >> >>>Should the cell sizes of the causal factor in an ANOVA be normally >>>distributed? >> >>There is no requirement that the cell sizes be normally distributed. The >>only requirement in the general linear model is that the errors are >>normally distributed, and that is only necessary if you want to use >>F-tests or t-tests. >> >>It is however true that ANOVA loses power when the samples sizes are >>radically different, but even so, widely different sample sizes do not >>violate any distributional requirement. >> >>-- >>Paige Miller >>[EMAIL PROTECTED] . . . . . . . . . . . . . . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
