lmer vs. lme [R for BE/BA]

posted by ElMaestro  – Denmark, 2011-01-28 09:42 (3918 d 14:12 ago) – Posting: # 6505
Views: 62,639

Ahoy d_labes,

» But it lacks the separate estimability of the intra-subject error variance for Test and Reference, I think. As far as I know (from the R help lists) there are no plans to introduce somefink like weights in lmer().

Here are two partial outputs from lme and lmer:

With lme:
Log-restricted-likelihood: 98.12818
(...)
Random effects:
  Formula: ~trt - 1 | subj
  Structure: General positive-definite, Log-Cholesky parametrization
          StdDev      Corr
trt1     0.149013060 trt1
trt2     0.122163414 0.15
Residual 0.007643612

With lmer:
AIC    BIC logLik deviance REMLdev
-170.2 -150.0  95.12   -229.3  -190.2
(...)
Random effects:
Groups   Name Variance   Std.Dev.  Corr
subj     trt1 2.2218e-02 0.1490555     
         trt2 1.4913e-02 0.1221188 0.150
Residual      3.3117e-05 0.0057548


Note the slightly better likelihood found by lmer. Changes quite a bit (%-wise, I mean) on the residual sigma.

Pass or fail!
ElMaestro

Complete thread:

Activity
 Admin contact
21,749 posts in 4,547 threads, 1,545 registered users;
online 3 (0 registered, 3 guests [including 3 identified bots]).
Forum time: Thursday 00:54 CEST (Europe/Vienna)

Imagine if every Thursday your shoes exploded
if you tied them the usual way.
This happens to us all the time with computers,
and nobody thinks of complaining.    Jef Raskin

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5