[R-lang] Re: confidence interval for condition means from lmer

[James Kirby]

> On 3 Mar 2017, at 20:55, Dan Mirman <d...@danmirman.org 
> <mailto:d...@danmirman.org>> wrote:
> 
> I think "lsmeans" package can also do this, 

And will give virtually identical results (here, with adjusted covariance 
calculations disabled):
library(lmeans)
library(effects)
library(languageR)

> m <- lmer(RTlexdec ~ WordCategory*AgeSubject + (1|Word), english)

> as.data.frame(effect("WordCategory:AgeSubject", m))
  WordCategory AgeSubject      fit          se    lower    upper
1            N        old 6.664176 0.002820814 6.658646 6.669706
2            V        old 6.656649 0.003619112 6.649554 6.663744
3            N      young 6.443781 0.002820814 6.438251 6.449311
4            V      young 6.432612 0.003619112 6.425517 6.439708

> eng.lsm <- lsmeans(m, "WordCategory", by="AgeSubject")
> eng.lsm
AgeSubject = old:
 WordCategory   lsmean          SE df asymp.LCL asymp.UCL
 N            6.664176 0.002820814 NA  6.658647  6.669704
 V            6.656649 0.003619112 NA  6.649556  6.663742

AgeSubject = young:
 WordCategory   lsmean          SE df asymp.LCL asymp.UCL
 N            6.443781 0.002820814 NA  6.438252  6.449310
 V            6.432612 0.003619112 NA  6.425519  6.439706

Confidence level used: 0.95 

Behind the scenes, lsmeans relies the PBmodcomp function from the pbkrtest 
package, which uses parametric bootstrap. cf. Jan’s roll-your-own method:
> cells
  AgeSubject WordCategory Prediction     LoCI     HiCI
1        old            N   6.664176 6.659008 6.669882
2      young            N   6.443781 6.438246 6.449022
3        old            V   6.656649 6.649747 6.663679
4      young            V   6.432612 6.425705 6.439766

James
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.