[R-sig-ME] Difference in variance estimates by nlme and lme4

[YANG, Richard ChunHSi]

Dear all;

	I'm using both nlme and lme4 to estimate random effects of a
data set with three nested levels Site/Block/Spacing. Estimates of both
fixed and random components are almost identical with one exception of
Block:Site. The estimated std. dev. was 0.0000000 from lme4 and
8.819877e-07 from nlme (see attached output); both are small and close
to 0, in fact. But refitting a reduced model by dropping Block, the
estimated  intercept std. dev. for Site level was 1.4298e-7, and
1.1499e-06, respectively for lme4 and nlme; two estimates differ by 8
times. 

	The anova method reports 

> anova(ltvlmer, ltvlmer.R)
Data: temp2
Models:
ltvlmer.R: log(TotVolha + 10) ~ SDI * Dq * SpH + Standage + (1 | Site) +

ltvlmer.R:     (1 | Site:Spacing)
ltvlmer: log(TotVolha + 10) ~ SDI * Dq * SpH + Standage + (1 |
Site/Block/Spacing)
              Df     AIC     BIC logLik Chisq Chi Df Pr(>Chisq)
ltvlmer.R 12    -747.21 -705.45 385.61                        
ltvlmer    13    -741.47 -696.23 383.74     0      1          1

> anova(ltvlme, ltvlme.R)
         Model df       AIC       BIC          logLik     Test
L.Ratio   p-value
ltvlme       1  13  -545.5575 -500.8061 285.7788                        
ltvlme.R    2  12  -552.3547 -511.0457 288.1774 1 vs 2 4.797232  0.0285

The anova from lme4 indicates no significant difference between two
models, but significant at the 5% level by nlme.

	The discrepancy maybe result from different optimization
implementation. Did I miss anything?

> sessionInfo()
R version 2.11.0 (2010-04-22) 
i386-pc-mingw32 

locale:
[1] LC_COLLATE=English_Canada.1252  LC_CTYPE=English_Canada.1252
LC_MONETARY=English_Canada.1252 LC_NUMERIC=C                   
[5] LC_TIME=English_Canada.1252    

attached base packages:
[1] grDevices datasets  splines   graphics  stats     tcltk     utils
methods   base     

other attached packages:
[1] lme4_0.999375-33   Matrix_0.999375-44 lattice_0.18-5     nlme_3.1-96
svSocket_0.9-48    TinnR_1.0.3        R2HTML_2.1         Hmisc_3.8-1

[9] survival_2.35-8   

loaded via a namespace (and not attached):
[1] cluster_1.12.3 grid_2.11.0    svMisc_0.9-57  tools_2.11.0  

Richard

 <<lmerVs lme.Output.txt>> 	


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: lmerVs lme.Output.txt
URL: <https://stat.ethz.ch/pipermail/r-sig-mixed-models/attachments/20110630/067cab24/attachment-0003.txt>