[R-sig-ME] observation level random effects; estimated variance of variance component estimates

[Ben Bolker]

  I have a hacked version of lme4 that comments out the
error you are hitting (in the C code), and gets a plausible fit (at
least the fixed effects look pretty similar to Breslow and
Clayton 1993) -- see below.

  Don't know about your second question --

============================
> fit
Generalized linear mixed model fit by the Laplace approximation
Formula: update(formula1, . ~ . + (1 | id) + (1 | rand))
   Data: dat
   AIC   BIC logLik deviance
 499.7 527.4 -241.9    483.7
Random effects:
 Groups Name        Variance Std.Dev.
 rand   (Intercept) 0.12747  0.35702
 id     (Intercept) 0.21097  0.45932
Number of obs: 236, groups: rand, 236; id, 59

Fixed effects:
              Estimate Std. Error z value Pr(>|z|)
(Intercept)   -1.41194    1.16349  -1.214   0.2249
Base           0.88034    0.12910   6.819 9.17e-12 ***
Trt           -0.94857    0.39521  -2.400   0.0164 *
I(Trt * Base)  0.34922    0.20027   1.744   0.0812 .
Age            0.49015    0.34162   1.435   0.1514
V4TRUE        -0.10312    0.08583  -1.201   0.2296
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Correlation of Fixed Effects:
            (Intr) Base   Trt    I(T*B) Age
Base        -0.163
Trt          0.047  0.595
I(Trt*Base) -0.119 -0.653 -0.930
Age         -0.976 -0.038 -0.192  0.254
V4TRUE      -0.018 -0.003  0.002  0.000  0.001

> sessionInfo()
R version 2.8.1 (2008-12-22)
i486-pc-linux-gnu

locale:
LC_CTYPE=en_US.UTF-8;LC_NUMERIC=C;LC_TIME=en_US.UTF-8;LC_COLLATE=en_US.UTF-8;LC_MONETARY=C;LC_MESSAGES=en_US.UTF-8;LC_PAPER=en_US.UTF-8;LC_NAME=C;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_US.UTF-8;LC_IDENTIFICATION=C

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

other attached packages:
[1] glmmAK_1.2         mvtnorm_0.9-4      coda_0.13-4
smoothSurv_0.3-12
[5] survival_2.34-1    lme4_0.999375-28   Matrix_0.999375-20
lattice_0.17-20

loaded via a namespace (and not attached):
[1] grid_2.8.1    rjags_1.0.3-4 tools_2.8.1


Youyi Fong wrote:
> Dear lmers,
> 
> I have two questions regarding fitting GLMM using maximum likelihood method.
> The first one arises from trying repeat an analysis in the Breslow and
> Clayton 1993 JASA paper. Model 3 of the epileptic dataset has two random
> effects, one subject specific, and one observation specific. Thus if we
> count random effects, there are more parameters than observations. When I
> try to run the following code, I get an error saying: "Error in
> mer_finalize(ans) : q = 295 > n = 236".
> 
> require (lme4)
> require (glmmAK)
> data(epilepticBC)
> dat = epilepticBC
> dat$rand=1:nrow(dat)
> dat$V4=dat$visit==4
> formula1 = Seizure ~ Base + Trt + I(Trt*Base) + Age + V4
> fit=lmer (update (formula1, .~. + (1|id) + (1|rand)), family=poisson,
> data=dat, nAGQ=1)
> 
> Is it true that there is no way to fit such a model in an ML analysis? In
> other words, is there a way to approximate the likelihood of fixed effects
> and variance components without relying on estimates of random effects?
> 
> The second question is that when it is possible to obtain MLE of a GLMM
> model, how can I obtain an estimated variance of the variance component
> estimates using lmer or other functions?
> 
> Thank you very much for your help!
> 
> Youyi Fong
> 
> -------------------------------------------------------------------------------------
> Youyi Fong, Graduate Student, Department of Biostatistics
> University of Washington, Box 357232, Seattle, WA 98195
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models


-- 
Ben Bolker
Associate professor, Biology Dep't, Univ. of Florida
bolker at ufl.edu / www.zoology.ufl.edu/bolker
GPG key: www.zoology.ufl.edu/bolker/benbolker-publickey.asc