More than 1 year ago (Feb 2013), I had used lmer to run a mixed effect model involving a binomial outcome with the following command:
nl3.lmer <- glmer( cul.bi ~ food.act + where + intlan + inter.cul + via.m + via.h + (1|Id), data=drm, family=binomial)
Everything had worked fine, without any error or warning messages and I had presented my results. When trying to look at it today I ran into two problems:
1)
summary(nl3.lmer)
Length Class Mode
1 mer S4
...instead of the usual summary.
Is it because of the change of class from a mer object to a merMod object? I read something about that change in the description of the lme4 package updates, but I want to double check this is indeed the case.... If it is, is there an easy way for summary to be able to read a mer object?
2) (I'm much more concerned about this one)
Because I couldn't use my old nl3.lmer object anymore I decided to just rerun the same model.
mod <- glmer( cul.bi ~ food.act + where + intlan + inter.cul + via.m + via.h + (1|id), data=drm, family=binomial)
Warning message:
In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, :
Model failed to converge with max|grad| = 1.09141 (tol = 0.001)
I'm a little bit alarmed. Does that mean that the model I had reported last year was actually not "good" (sorry for the lack of a better statistical qualifier)? What does it mean actually, in simple terms?
Here's the summary of that model:
Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) ['glmerMod']
Family: binomial ( logit )
Formula: cul.bi ~ food.act + where + intlan + inter.cul + via.m + via.h + (1 | id)
Data: drm
AIC BIC logLik deviance df.resid
522.0 582.3 -247.0 494.0 537
Scaled residuals:
Min 1Q Median 3Q Max
-3.6076 -0.4533 0.1179 0.4038 4.4283
Random effects:
Groups Name Variance Std.Dev.
id (Intercept) 2.626 1.62
Number of obs: 551, groups: id, 101
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.9401 1.6664 1.164 0.244317
food.actYes 0.7533 0.3250 2.318 0.020455 *
whereschool -1.5370 0.4127 -3.724 0.000196 ***
wherework -1.2763 0.7957 -1.604 0.108701
whereother -1.1453 0.3749 -3.055 0.002249 **
intlanher 3.1935 0.8812 3.624 0.000290 ***
intlanbil 0.9765 0.5805 1.682 0.092555 .
intlanno int -0.6395 0.5910 -1.082 0.279268
inter.culher 2.5482 0.5020 5.076 3.86e-07 ***
inter.culher-main 1.2225 0.4588 2.664 0.007713 **
inter.culno int 1.2966 0.6076 2.134 0.032845 *
via.m -0.8063 0.2441 -3.304 0.000954 ***
via.h 0.4862 0.2003 2.428 0.015183 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Correlation of Fixed Effects:
(Intr) fd.ctY whrsch whrwrk whrthr intlnh intlnb intlni intr.c intr.- intr.i via.m
food.actYes -0.091
whereschool -0.100 0.143
wherework -0.050 0.103 0.233
whereother -0.129 0.179 0.420 0.256
intlanher -0.065 0.078 0.019 0.052 -0.037
intlanbil -0.053 0.086 0.061 -0.068 0.046 0.072
intlannoint -0.055 -0.012 0.104 0.055 0.137 0.051 0.073
inter.culhr -0.120 0.063 -0.017 0.050 0.022 0.017 0.048 0.024
intr.clhr-m -0.118 0.079 -0.155 -0.193 -0.094 -0.011 -0.040 -0.027 0.432
intr.clnint -0.049 0.103 0.047 0.036 -0.010 0.048 0.046 -0.745 0.335 0.361
via.m -0.596 -0.103 0.083 -0.010 0.084 -0.074 0.037 0.056 -0.034 -0.024 -0.065
via.h -0.391 0.119 -0.104 0.006 -0.075 0.126 -0.036 -0.038 0.033 0.031 0.015 -0.469
I would have been happy to provide a reproducible example, but I'm afraid it wouldn't really mirror my issue. And in case it's useful:
sessionInfo()
R version 3.1.0 (2014-04-10)
Platform: i686-pc-linux-gnu (32-bit)
locale:
[1] LC_CTYPE=en_CA.UTF-8 LC_NUMERIC=C LC_TIME=en_CA.UTF-8
[4] LC_COLLATE=en_CA.UTF-8 LC_MONETARY=en_CA.UTF-8 LC_MESSAGES=en_CA.UTF-8
[7] LC_PAPER=en_CA.UTF-8 LC_NAME=C LC_ADDRESS=C
[10] LC_TELEPHONE=C LC_MEASUREMENT=en_CA.UTF-8 LC_IDENTIFICATION=C
attached base packages:
[1] parallel splines grid stats graphics grDevices utils datasets methods base
other attached packages:
[1] RColorBrewer_1.0-5 glmmML_1.0 ggplot2_0.9.3.1 pbkrtest_0.3-8 mvoutlier_2.0.4
[6] sgeostat_1.0-25 HLMdiag_0.2.5 plyr_1.8.1 prettyR_2.0-7 languageR_1.4.1
[11] msm_1.3 e1071_1.6-3 nlme_3.1-117 MASS_7.3-33 nnet_7.3-8
[16] car_2.0-19 effects_3.0-0 colorspace_1.2-4 Hmisc_3.14-4 Formula_1.1-1
[21] survival_2.37-7 lattice_0.20-29 lmerTest_2.0-6 lme4_1.1-6 Rcpp_0.11.1
[26] Matrix_1.1-3
loaded via a namespace (and not attached):
[1] bitops_1.0-6 caTools_1.17 class_7.3-10 cluster_1.15.2
[5] DEoptimR_1.0-1 dichromat_2.0-0 digest_0.6.4 expm_0.99-1.1
[9] gdata_2.13.3 gplots_2.13.0 gtable_0.1.2 gtools_3.4.0
[13] KernSmooth_2.23-12 labeling_0.2 latticeExtra_0.6-24 minqa_1.2.3
[17] munsell_0.4.2 mvtnorm_0.9-99992 numDeriv_2012.9-1 pcaPP_1.9-49
[21] pls_2.4-3 proto_0.3-10 RcppEigen_0.3.2.1.2 reshape2_1.2.2
[25] robCompositions_1.8.0 robustbase_0.91-1 rrcov_1.3-4 scales_0.2.3
[29] stats4_3.1.0 stringr_0.6.2 tools_3.1.0
Thank you in advance for any help you can offer.
(Upgraded from a comment)
- the class of objects produced by
glmerhas indeed changed (frommertomerMod). You can install a back-compatibilitylme4.0package from http://lme4.r-forge.r-project.org/repos :
install.packages("lme4.0",repos="http://lme4.r-forge.r-project.org/repos")
and see ?convert_old_lme4 for a utility function to convert the saved model fit so you can use it in lme4.0.
- Your convergence warning is almost certainly a false positive: see this mailing list thread or the tail end of this Github issue. You can either install the development version from Github (via
devtools::install_github("lme4","lme4")) or the repository listed above, or just try this:
relgrad <- with(mixmod1@optinfo$derivs, solve(Hessian, gradient))
max(abs(relgrad))
and see if the scaled gradient is small.
来源：https://stackoverflow.com/questions/23814130/glmer-model-from-early-2013-warning-message-about-convergence-when-re-running-i