Dear List,
As I understand, GLMM (in experimental lme4) and glmmPQL (MASS) do
similar things using somewhat different methods. Trying both,
I get the same coefficients, but markedly different std. errors and
p-values.
Any help in understanding the models tested by both procedures?
Dieter Menne
UseMASS<-T # must restart R after changing because of nlme/lme4 clash
if (UseMASS){
library(MASS)
summary(glmmPQL(y ~ trt + I(week > 2), random = ~ 1 | ID,
family = binomial, data = bacteria))
} else
{
library(lme4)
summary(GLMM(y ~ trt + I(week > 2), random = ~ 1 | ID,
family = binomial, data = bacteria,method="PQL"))
}
(MASS output)
Fixed effects: y ~ trt + I(week > 2)
Value Std.Error DF t-value p-value
(Intercept) 3.412012 0.5185028 169 6.580509 0.0000
trtdrug -1.247355 0.6440627 47 -1.936698 0.0588
trtdrug+ -0.754327 0.6453971 47 -1.168780 0.2484
I(week > 2)TRUE -1.607256 0.3583378 169 -4.485310 0.0000
(lme4 output)
Fixed effects: y ~ trt + I(week > 2)
Estimate Std. Error DF z value Pr(>|z|)
(Intercept) 3.41202 3.93293 169 0.8676 0.3856
trtdrug -1.24736 1.52156 47 -0.8198 0.4123
trtdrug+ -0.75433 1.21963 47 -0.6185 0.5363
I(week > 2)TRUE -1.60726 2.19660 169 -0.7317 0.4644
I believe the distinction is explained in the lme4 documentation but, in any case, the standard errors and the approximate log-likelihood for glmmPQL are from the lme model that is the last step in the optimization. The corresponding quantities from GLMM are from another approximation that should be more reliable. "Dieter Menne" <dieter.menne at menne-biomed.de> writes:> Dear List, > > As I understand, GLMM (in experimental lme4) and glmmPQL (MASS) do > similar things using somewhat different methods. Trying both, > I get the same coefficients, but markedly different std. errors and > p-values. > Any help in understanding the models tested by both procedures? > > Dieter Menne > > > UseMASS<-T # must restart R after changing because of nlme/lme4 clash > if (UseMASS){ > library(MASS) > summary(glmmPQL(y ~ trt + I(week > 2), random = ~ 1 | ID, > family = binomial, data = bacteria)) > } else > { > library(lme4) > summary(GLMM(y ~ trt + I(week > 2), random = ~ 1 | ID, > family = binomial, data = bacteria,method="PQL")) > } > > (MASS output) > Fixed effects: y ~ trt + I(week > 2) > Value Std.Error DF t-value p-value > (Intercept) 3.412012 0.5185028 169 6.580509 0.0000 > trtdrug -1.247355 0.6440627 47 -1.936698 0.0588 > trtdrug+ -0.754327 0.6453971 47 -1.168780 0.2484 > I(week > 2)TRUE -1.607256 0.3583378 169 -4.485310 0.0000 > > (lme4 output) > Fixed effects: y ~ trt + I(week > 2) > Estimate Std. Error DF z value Pr(>|z|) > (Intercept) 3.41202 3.93293 169 0.8676 0.3856 > trtdrug -1.24736 1.52156 47 -0.8198 0.4123 > trtdrug+ -0.75433 1.21963 47 -0.6185 0.5363 > I(week > 2)TRUE -1.60726 2.19660 169 -0.7317 0.4644 > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html-- Douglas Bates bates at stat.wisc.edu Statistics Department 608/262-2598 University of Wisconsin - Madison http://www.stat.wisc.edu/~bates/
On Fri, Jan 09, 2004 at 12:26:21PM -0600, Douglas Bates wrote:> I believe the distinction is explained in the lme4 documentation but, > in any case, the standard errors and the approximate log-likelihood > for glmmPQL are from the lme model that is the last step in the > optimization. The corresponding quantities from GLMM are from another > approximation that should be more reliable.It would be interesting to see what glmmML, which uses yet another approximation, gives on this particular data set. Could you (Dieter) try it, and perhaps also share your data with us?> > "Dieter Menne" <dieter.menne at menne-biomed.de> writes: > > > Dear List, > > > > As I understand, GLMM (in experimental lme4) and glmmPQL (MASS) do > > similar things using somewhat different methods. Trying both, > > I get the same coefficients, but markedly different std. errors and > > p-values. > > Any help in understanding the models tested by both procedures? > > > > Dieter Menne > > > > > > UseMASS<-T # must restart R after changing because of nlme/lme4 clash > > if (UseMASS){ > > library(MASS) > > summary(glmmPQL(y ~ trt + I(week > 2), random = ~ 1 | ID, > > family = binomial, data = bacteria)) > > } else > > { > > library(lme4) > > summary(GLMM(y ~ trt + I(week > 2), random = ~ 1 | ID, > > family = binomial, data = bacteria,method="PQL")) > > } > > > > (MASS output) > > Fixed effects: y ~ trt + I(week > 2) > > Value Std.Error DF t-value p-value > > (Intercept) 3.412012 0.5185028 169 6.580509 0.0000 > > trtdrug -1.247355 0.6440627 47 -1.936698 0.0588 > > trtdrug+ -0.754327 0.6453971 47 -1.168780 0.2484 > > I(week > 2)TRUE -1.607256 0.3583378 169 -4.485310 0.0000 > > > > (lme4 output) > > Fixed effects: y ~ trt + I(week > 2) > > Estimate Std. Error DF z value Pr(>|z|) > > (Intercept) 3.41202 3.93293 169 0.8676 0.3856 > > trtdrug -1.24736 1.52156 47 -0.8198 0.4123 > > trtdrug+ -0.75433 1.21963 47 -0.6185 0.5363 > > I(week > 2)TRUE -1.60726 2.19660 169 -0.7317 0.4644 > > > > ______________________________________________ > > R-help at stat.math.ethz.ch mailing list > > https://www.stat.math.ethz.ch/mailman/listinfo/r-help > > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html > > -- > Douglas Bates bates at stat.wisc.edu > Statistics Department 608/262-2598 > University of Wisconsin - Madison http://www.stat.wisc.edu/~bates/ > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html-- G?ran Brostr?m tel: +46 90 786 5223 Department of Statistics fax: +46 90 786 6614 Ume? University http://www.stat.umu.se/egna/gb/ SE-90187 Ume?, Sweden e-mail: gb at stat.umu.se
Goran, from my reply to a message from Douglas Bates; ">" is quoted from a mail by DG.> I believe the distinction is explained in the lme4 documentation but, > in any case, the standard errors and the approximate log-likelihood > for glmmPQL are from the lme model that is the last step in the > optimization. The corresponding quantities from GLMM are from another > approximation that should be more reliable.I have compared glmmPQL, glmmML, geese and GLMM, results and code see below. I am aware that glmmPQL uses another method to handle the problem, and geese (geepack) has considerable different assumptions, but the results are very similar. On the other hand, I had expected that glmmML results, if reasonable at all, should be close to GLMM. Yet they are not, but rather come close to the other three. See for example the last factor (I(week>2): significant effects in all cases except GLMM. ----------------------------- (glmmPQL) Fixed effects: y ~ trt + I(week > 2) Value Std.Error DF t-value p-value (Intercept) 3.41 0.519 169 6.58 0.0000 trtdrug -1.25 0.644 47 -1.94 0.0588 trtdrug+ -0.75 0.645 47 -1.17 0.2484 I(week > 2)TRUE -1.61 0.358 169 -4.49 0.0000 (glmmML) coef se(coef) z Pr(>|z|) (Intercept) 3.579 0.701 5.10 3.3e-07 trtdrug -1.369 0.694 -1.97 4.8e-02 trtdrug+ -0.789 0.700 -1.13 2.6e-01 I(week > 2)TRUE -1.627 0.482 -3.38 7.3e-04 (geese from geepack) estimate san.se wald p (Intercept) 2.844 0.529 28.92 7.56e-08 trtdrug -1.113 0.586 3.61 5.76e-02 trtdrug+ -0.634 0.544 1.36 2.44e-01 I(week > 2)TRUE -1.325 0.368 12.96 3.18e-04 (GLMM) Estimate Std. Error DF z value Pr(>|z|) (Intercept) 3.41202 3.93293 169 0.8676 0.3856 trtdrug -1.24736 1.52156 47 -0.8198 0.4123 trtdrug+ -0.75433 1.21963 47 -0.6185 0.5363 I(week > 2)TRUE -1.60726 2.19660 169 -0.7317 0.4644 --------------- data(bacteria,package="MASS") UseMASS<-F# must restart R after changing because of nlme/lme4 clash if (UseMASS){ library(MASS) # required for bacteria options(digits=3) print(summary(glmmPQL(y ~ trt + I(week > 2), random = ~ 1 | ID, family = binomial, data = bacteria))) library(glmmML) print(glmmML(y=="y"~trt+I(week>2), data=bacteria,cluster=bacteria$ID)) library(geepack) summary(geese(y == "y" ~ trt + I(week > 2), family = binomial, id = ID, corstr = "exchangeable",data=bacteria)) } else { library(lme4) # try a well documented "old" lme to compare lme4/lme3 for normal things # data(Orthodont,package=nlme) # fm <- lme(distance ~ Sex*I(age-11), data Orthodont,random=~I(age-11)|Subject) # summary(fm) summary(GLMM(y ~ trt + I(week > 2), random = ~ 1 | ID, family = binomial, data = bacteria,method="PQL")) }