0

I am trying to compute the statistical power of my linear mixed model, which includes 2 conditions and 16 observations from 3 subjects. Fixed effect is the condition ('Cond') and random effect is subject identity (SubID).

Here is my data

         lme4 <- lmer(delta ~ Cond + (1|SubID), data = Data)

I used powerSim (simr package):

         powerSim (fit = lme4, test = fixed ("Cond" ), nsim = 1000)

And this is what I obtained: a power of 7-9%, which is very low.

         Power for predictor 'Cond', (95% confidence interval):
         8.5% ( 9.46, 15.39)
         
         Test: Likelihood ratio

         Based on 1000 simulations, (20 warnings, 0 errors)
         alpha = 0.05, nrow = 16

Given that some of the data reach significance, I am expecting a power around 60 to 80%. Could you tell me whether I did anything wrong in the computation?

Nicolas Decat
  • 1
  • welcome to SO| avoid pasting images as others cannot reproduce your analysis. see https://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example – StupidWolf Aug 27 '20 at 09:44
  • also if your question is a statistical one, cross-validated is the proper place to post it. The code seems ok. with n=16.. what kind of power are you expecting? – StupidWolf Aug 27 '20 at 09:44
  • A post-hoc power estimation is just a transformation of your *p*-value, there is little point to it. See here for example: http://daniellakens.blogspot.com/2014/12/observed-power-and-what-to-do-if-your.html – Frans Rodenburg Aug 28 '20 at 04:31

0 Answers0