[R-sig-ME] bootstrapping random effects using confint() in lme4
Ben Bolker
bbolker at gmail.com
Fri Oct 14 21:13:13 CEST 2016
I answered this offline but wanted to encourage everyone, when in
doubt, to post to r-sig-mixed-models at r-project.org rather than e-mailing
me directly ...
confint.merMod has a FUN argument that allows bootstrap CIs on an
arbitrary function (as long as it returns a numeric vector) to be
computed, e.g.
confint(fitted_model,method="boot",FUN=function(m) unlist(ranef(m)))
(We should probably add this as an example to the documentation ...)
One can also get an estimate of the uncertainty in the conditional
modes by extracting the "postVar" attributes from the elements of
ranef(fitted_model,condVar=TRUE) ... if fm1 is the fitted model,
> cvar <- lapply(ranef(fm1,condVar=TRUE),function(x) attr(x,"postVar"))
> apply(cvar[[1]],3,diag)
On 16-10-14 02:08 PM, Jonathan Miller wrote:
> Dr. Bolker,
>
> I am sorry for what I imagine is a pretty straightforward question. My
> name is Jonathan Miller and I am a Phd student at NCSU in Civil
> Engineering. I have tried to look through online resources for a post on
> this, but have not been successful in determining what I need to do.
>
> I have been using lme4 to run a logisitic mixed model and it has worked
> very well. Recently, though, I was interested in determining the
> uncertainty in the random effects in my model. It contains three levels of
> random effects, estuaries(32), states(5) and programs(7). I think the
> function confint() is what I need to do, but I am having trouble getting
> the outputs for the individual random effects(i.e. estuaries, states,
> programs).
>
> My code is below:
>
> m3=glmer(Pres~ Temp_mean + Sal_mean + salsq + NEAR_DIST
> + (1|STATE) + (1 | UNIQUEID) + (1|program),
> data=bighead, na.action = na.exclude, nAGQ=0,
> family=binomial(link="logit"))
>
> confint.merMod(m3,method="boot",nsim=100)
>
> My output just gives me a summary of each fixed effect and each random
> effect group. I think there must be a relatively simple way to get the
> output of the bootstrapping for each individual random effect. It seems
> the answer lies in FUN, but I haven't gotten it to work for me yet.
>
>
> 2.5 % 97.5 %
> sd_(Intercept)|UNIQUEID 4.720458e-01 0.98029041
> sd_(Intercept)|program 3.168555e-01 1.26836628
> sd_(Intercept)|STATE 3.019507e-07 1.30279935
> (Intercept) -6.088106e+00 -3.81230673
> Temp_mean -6.882682e-02 -0.05817473
> Sal_mean 1.696714e-01 0.20005360
> salsq -2.253061e+00 -1.89913153
> NEAR_DIST 3.895820e-01 0.50612062
>
>
> Any help would be greatly appreciated.
>
> Thank you,
>
> Jonathan Miller
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
More information about the R-sig-mixed-models
mailing list