Emmanuel Tillard
2005-Jun-16 08:31 UTC
[R] identical results with PQL and Laplace options in lmer function (package lme4)
Dear R users I encounter a problem when i perform a generalized linear mixed model (binary data) with the lmer function (package lme4) with R 2.1.0 on windows XP and the latest version of package "lme4" (0.96-1) and "matrix" (0.96-2) both options "PQL" and "Laplace" for the method argument in lmer function gave me the same results (random and fixed effects estimates, standard error and p.values). However, Loglikelihood and deviance are different. here is an example reproduced with the bacteria data set available in the MASS package: library(lme4) library(MASS) data(bacteria) bacteria$week2 <- as.factor(ifelse(bacteria$week <=2, 0, 1)) model.PQL <- lmer(y ~ trt + week2 + (1 | ID), family = binomial, data = bacteria, method ="PQL") model.Laplace <- lmer(y ~ trt + week2 + (1 | ID), family = binomial, data = bacteria, method ="Laplace") model.PQL model.Laplace> model.PQLGeneralized linear mixed model fit using PQL Formula: y ~ trt + week2 + (1 | ID) Data: bacteria Family: binomial(logit link) AIC BIC logLik deviance 152.443 138.8685 -80.2215 160.443 Random effects: Groups Name Variance Std.Dev. ID (Intercept) 3.2721 1.8089 # of obs: 220, groups: ID, 50 Estimated scale (compare to 1) 0.7800484 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) 3.41227 0.65884 5.1792 2.228e-07 *** trtdrug -1.24743 0.81841 -1.5242 0.1274555 trtdrug+ -0.75440 0.82009 -0.9199 0.3576229 week21 -1.60737 0.45527 -3.5306 0.0004146 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1> model.LaplaceGeneralized linear mixed model fit using Laplace Formula: y ~ trt + week2 + (1 | ID) Data: bacteria Family: binomial(logit link) AIC BIC logLik deviance 304.9488 291.3743 -156.4744 312.9488 Random effects: Groups Name Variance Std.Dev. ID (Intercept) 3.2721 1.8089 # of obs: 220, groups: ID, 50 Estimated scale (compare to 1) 0.7800484 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) 3.41227 0.65884 5.1792 2.228e-07 *** trtdrug -1.24743 0.81841 -1.5242 0.1274555 trtdrug+ -0.75440 0.82009 -0.9199 0.3576229 week21 -1.60737 0.45527 -3.5306 0.0004146 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 is anybody else aware of this? or did I forget something important ? Many thanks for your help. -- Emmanuel Tillard Veterinaire CIRAD-EMVT Unite de recherche 18 UMR868 Elevage des Ruminants en Regions Chaudes (ERRC) Campus ENSA-INRA 2 place Viala 34060 Montpellier cedex 1 tel: 0499612265 (fixe) 0633850598 (gsm) fax: 0467545694 e-mail: tillard at cirad.fr
ronggui
2005-Jun-16 10:48 UTC
[R] identical results with PQL and Laplace options in lmer function (package lme4)
i am aware of this but i don't know why. On Thu, 16 Jun 2005 10:31:25 +0200 Emmanuel Tillard <tillard at cirad.fr> wrote:> Dear R users > > I encounter a problem when i perform a generalized linear mixed model (binary data) with the lmer function (package lme4) > with R 2.1.0 on windows XP and the latest version of package "lme4" (0.96-1) and "matrix" (0.96-2) > both options "PQL" and "Laplace" for the method argument in lmer function gave me the same results (random and fixed effects estimates, standard error and p.values). However, Loglikelihood and deviance are different. > > here is an example reproduced with the bacteria data set available in the MASS package: > > library(lme4) > library(MASS) > > data(bacteria) > bacteria$week2 <- as.factor(ifelse(bacteria$week <=2, 0, 1)) > model.PQL <- lmer(y ~ trt + week2 + (1 | ID), family = binomial, data = bacteria, method ="PQL") > model.Laplace <- lmer(y ~ trt + week2 + (1 | ID), family = binomial, data = bacteria, method ="Laplace") > > model.PQL > model.Laplace > > > model.PQL > Generalized linear mixed model fit using PQL > Formula: y ~ trt + week2 + (1 | ID) > Data: bacteria > Family: binomial(logit link) > AIC BIC logLik deviance > 152.443 138.8685 -80.2215 160.443 > Random effects: > Groups Name Variance Std.Dev. > ID (Intercept) 3.2721 1.8089 > # of obs: 220, groups: ID, 50 > > Estimated scale (compare to 1) 0.7800484 > > Fixed effects: > Estimate Std. Error z value Pr(>|z|) > (Intercept) 3.41227 0.65884 5.1792 2.228e-07 *** > trtdrug -1.24743 0.81841 -1.5242 0.1274555 > trtdrug+ -0.75440 0.82009 -0.9199 0.3576229 > week21 -1.60737 0.45527 -3.5306 0.0004146 *** > --- > Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 > > > model.Laplace > Generalized linear mixed model fit using Laplace > Formula: y ~ trt + week2 + (1 | ID) > Data: bacteria > Family: binomial(logit link) > AIC BIC logLik deviance > 304.9488 291.3743 -156.4744 312.9488 > Random effects: > Groups Name Variance Std.Dev. > ID (Intercept) 3.2721 1.8089 > # of obs: 220, groups: ID, 50 > > Estimated scale (compare to 1) 0.7800484 > > Fixed effects: > Estimate Std. Error z value Pr(>|z|) > (Intercept) 3.41227 0.65884 5.1792 2.228e-07 *** > trtdrug -1.24743 0.81841 -1.5242 0.1274555 > trtdrug+ -0.75440 0.82009 -0.9199 0.3576229 > week21 -1.60737 0.45527 -3.5306 0.0004146 *** > --- > Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 > > is anybody else aware of this? or did I forget something important ? > > Many thanks for your help. > > -- > Emmanuel Tillard > Veterinaire > CIRAD-EMVT > Unite de recherche 18 > > UMR868 Elevage des Ruminants en Regions Chaudes (ERRC) > Campus ENSA-INRA > 2 place Viala > 34060 Montpellier cedex 1 > > tel: 0499612265 (fixe) > 0633850598 (gsm) > fax: 0467545694 > e-mail: tillard at cirad.fr > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html