[R-sig-ME] confidence intervals for random slopes

Ben Bolker bbolker at gmail.com
Wed Oct 15 17:20:07 CEST 2014

Veronika Bokony <bokony.veronika at ...> writes:

> Dear all,
> thank you very much for the responses. With Ben Bolker's example I can now 
> calculate the SD for each of my random slopes. I have one remaining 
> question though: how do I turn these into SE? Do I use the sample size per 
> each level of random factor? (I have different sample size for 
> each "subject".) Or is it more tricky with these mixed models?
> Thanks and bests,
> Veronika

  Sorry this fell through the cracks.

  These standard deviations are standard deviations *of the estimates*,
or in other words they're more or less equivalent to SE already.  So you 
should just use +/- 1.96 SD (or whatever width you want).  One property
(that you could consider as a bug or a feature) of the mixed model 
estimation is that, unlike if you were to fit a linear regression to
the slopes individually (in which case you would get different SDs for
each slope, and they would indeed be inversely proportional to the
sample size for the subject), a pooled estimate of the uncertainty is
used, so all the slopes have the same SD.

More information about the R-sig-mixed-models mailing list