[R-sig-ME] ordinal model and lsmeans mean class computing
espesser
robert.espesser at lpl-aix.fr
Wed Jun 15 23:33:06 CEST 2016
Dear All,
I tried to understand how the predicted mean class of an ordinal mixed
model was computed in lsmeans().
(I want to compute mean classes from the fixed effects only; lsmeans
included all the random terms)
# from the data wine, I estimated the model
library(ordinal)
data(wine)
fm22.clmm= clmm(rating~temp+contact +(1|judge),data=wine, Hess=T)
# get mean classes with lsmeans
lsmeans(fm22.clmm, ~ temp, mode="mean.class")
temp mean.class SE df asymp.LCL asymp.UCL
cold 2.339608 0.166799 NA 2.012688 2.666528
warm 3.479936 0.201921 NA 3.084178 3.875694
Results are averaged over the levels of: contact
Confidence level used: 0.95
Warning message:
In is.na(x) : is.na() applied to non-(list or vector) of type 'NULL'
I supposed that the predicted mean class is the mean class number (from
1 to 5) weighted by the estimated probabilities. Therefore I tried:
library(plyr)
wine$fitted = fm22.clmm$fitted
ddply(wine, ~temp, summarise, sum(fitted*as.numeric(rating)) /sum(fitted))
temp ..1
1 cold 2.269931
2 warm 3.487144
There are some discrepancies with the results of lsmeans.
I got similar (small) discrepancies for other data and clmm models.
I certainly missed something, and I would really appreciate your help!
Best regards,
Robert
here is the sessionInfo:
sessionInfo()
R version 3.2.2 (2015-08-14)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1
attached base packages:
[1] stats graphics grDevices utils datasets methods
[7] base
other attached packages:
[1] plyr_1.8.3 ordinal_2015.6-28 lsmeans_2.23
[4] estimability_1.1-1
loaded via a namespace (and not attached):
[1] Rcpp_0.12.2 lattice_0.20-33 codetools_0.2-14
[4] mvtnorm_1.0-5 zoo_1.7-13 ucminf_1.1-3
[7] MASS_7.3-45 grid_3.2.2 xtable_1.8-2
[10] nlme_3.1-121 coda_0.18-1 multcomp_1.4-5
[13] Matrix_1.2-2 sandwich_2.3-4 splines_3.2.2
[16] TH.data_1.0-7 tools_3.2.2 survival_2.38-3
--
Robert Espesser
CNRS UMR 7309 - Université Aix-Marseille
5 Avenue Pasteur
13100 AIX-EN-PROVENCE
Tel: +33 (0)413 55 36 26
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list