increasing nAGQ causes error

test2<-glmer(USED~1+IMPS+(1+IMPS|CITY),family=binomial,data=subset2)
test2A <- update(test2,nAGQ=2)

Dr. Bolker,

Thanks for your prompt reply.  I created a subset of my dataset (attached). 
This subset contains 14 of the 22 cities in the full dataset, and one 
covariate (% impervious surfaces [IMP]).  I had convergence issues before 
standardizing covariates, so I used:

subset2$IMPS<-(subset2$IMP-mean(subset2$IMP))/(sqrt(var(subset2$IMP)))

Fitting all 3 covariates takes quite a long time, and I get the same error 
when using just one of them:

test2<-glmer(USED~1+IMPS+(1+IMPS|CITY),family=binomial,data=subset2,nAGQ=8)
test2

Error in asMethod(object) : matrix is not symmetric [1,2]

Perhaps it's noteworthy that I did not receive the error when using a 
smaller subset with only the first 8 cities (in alphabetical order).

I look forward to your response.  Thanks again...

-Grant


----- Original Message ----- 
From: "Ben Bolker" <bolker at ufl.edu>
To: "Grant T. Stokke" <gts127 at psu.edu>
Cc: <r-sig-mixed-models at r-project.org>
Sent: Friday, May 29, 2009 12:11 PM
Subject: Re: [R-sig-ME] increasing nAGQ causes error

 This sounds worth digging into, but it's hard to dig
into without a reproducible example.  I don't get the
problem with the GLMM example in the lme4 package:

example(lmer)
update(gm1,nAGQ=8)
update(gm1,nAGQ=10)

etc.

 Can you post your data set, or a subset or simulated
data set that gives the same problem, somewhere?

 Ben Bolker

Grant T. Stokke wrote:

Hello All,

I'm new to R and new to this mailing list, so I hope I've presented
the proper info in this post.  I'm using GLMMs to model the selection
of urban roosting locations by crows.  My dataset consists of 22
cities, with each city containing 1000 unused locations and from 83
to 2000 used locations.  I have three covariates for each used or
unused location which I've standardized across all observations: %
canopy (CANS), % impervious surfaces (IMPS), and nighttime light
level (LTS).  Using the default setting nAGQ=1, my full model is
fitted without error:

CIL_CIL<-glmer(USED~1+CANS+IMPS+LTS+(1+CANS+IMPS+LTS|CITY),family=binomial,data=sitescale)
 CIL_CIL

Generalized linear mixed model fit by the Laplace approximation
Formula: USED ~ 1 + CANS + IMPS + LTS + (1 + CANS + IMPS + LTS |
CITY) Data: sitescale AIC   BIC logLik deviance 15122 15237  -7547
15094 Random effects: Groups Name        Variance   Std.Dev. Corr
 CITY   (Intercept) 321.698184 17.93595 CANS          0.073271
0.27069  0.026 IMPS          0.661947  0.81360  0.080 -0.698 LTS
455.829122 21.35016 -0.988  0.031 -0.132 Number of obs: 27128,
groups: CITY, 22

Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept)
-16.21878    4.01227  -4.042 5.29e-05 *** CANS          0.07881
0.07162   1.100 0.271147 IMPS          0.63137    0.17750   3.557
0.000375 *** LTS          19.50827    4.78822   4.074 4.62e-05 ***
--- Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects: (Intr) CANS   IMPS CANS  0.020
 IMPS  0.074 -0.500 LTS  -0.989  0.025 -0.124

When I increase nAGQ to 8, however, I get the following error:

CIL_CIL.nAGQ8<-glmer(USED~1+CANS+IMPS+LTS+(1+CANS+IMPS+LTS|CITY),family=binomial,data=sitescale,nAGQ=8)
 CIL_CIL.nAGQ8

Error in asMethod(object) : matrix is not symmetric [1,2]

I get the same error message with other values for nAGQ (I tried nAGQ
= 2, 3, 5, and 50).  Is there anything I can do to fit the model
using nAGQ > 1 without error?  Thanks in advance for your help!

-Grant Stokke

sessionInfo()

R version 2.9.0 (2009-04-17) i386-pc-mingw32

locale: LC_COLLATE=English_United States.1252;LC_CTYPE=English_United
States.1252;LC_MONETARY=English_United
States.1252;LC_NUMERIC=C;LC_TIME=English_United States.1252

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

other attached packages: [1] mgcv_1.5-5         lme4_0.999375-30
Matrix_0.999375-26 lattice_0.17-22

loaded via a namespace (and not attached): [1] grid_2.9.0
nlme_3.1-90 tools_2.9.0

CIL_CIL.nAGQ8

Error in asMethod(object) : matrix is not symmetric [1,2]

traceback()

18: .Call(dense_to_symmetric, from, "U", TRUE) 17: asMethod(object)
16: as(from, "symmetricMatrix") 15: .class1(object) 14: as(as(from,
"symmetricMatrix"), "dMatrix") 13: .class1(object) 12: as(as(as(from,
"symmetricMatrix"), "dMatrix"), "denseMatrix") 11: .class1(object)
10: as(as(as(as(from, "symmetricMatrix"), "dMatrix"), "denseMatrix"),
 "dpoMatrix") 9: asMethod(object) 8: as(sigma(object)^2 *
chol2inv(object at RX, size = object at dims["p"]), "dpoMatrix") 7:
vcov(object) 6: vcov(object) 5: summary(x) 4: summary(x) 3:
printMer(object) 2: function (object) standardGeneric("show")(<S4
object of class "mer">) 1: function (object)
standardGeneric("show")(<S4 object of class "mer">)



[[alternative HTML version deleted]]

-- 
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