[R-sig-ME] HLM post-hoc power analysis and effect sizes

[Karl Ove Hufthammer]

Julia Velten skreiv 30. jan. 2016 09:26:
> 1. Are you familiar with an R-package for calculating HLM effect sizes and/or a post-hoc power analysis?

Are you talking about *standardised* effect sizes? IMO, they don’t make 
much sense, and I would avoid reporting them, if at all possible. See, 
for example, 
http://www.graphpad.com/guides/prism/6/statistics/index.htm?stat_whats_wrong_with_standard_valu.htm

Non-standardised effect sizes *do* make sense, and are useful, and you 
get them directly from the nlme output.

For post-hoc power analysis, are you talking about power-analysis based 
on the *observed* effect? They don’t make any sense *at all*. See this 
paper:

Hoenig J. M. & Heisey D. M., 2001.
‘The Abuse of Power: The Pervasive Fallacy of Power Calculations for 
Data Analysis’
The American Statistician, American Statistical Association, vol. 55, 
pages 19–24, February.

It’s available as a PDF here: 
http://www.tc.umn.edu/~alonso/Hoenig_AmericanStat_2001.pdf

-- 
Karl Ove Hufthammer