[R] lmer, CHOLMOD warning: matrix not positive definite

[Douglas Bates]

On 4/2/07, Seyed Reza Jafarzadeh <srjafarzadeh at gmail.com> wrote:
> Hi,
>
> I am getting a warning message when I am fitting a generalized linear
> mixed model (m1.2 below).
>
>
> CHOLMOD warning: matrix not positive definite
>  Error in objective(.par, ...) : Cholmod error `matrix not positive
> definite' at file:../Supernodal/t_cholmod_super_numeric.c, line 614
>
>
> Any idea?
>
> Thanks for your help,
> Reza
>
>
>
>  > sessionInfo()
> R version 2.4.1 (2006-12-18)
>  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:
>        lme4      Matrix     lattice
> "0.9975-13" "0.9975-11"   "0.14-17"
>
>
>  > m1.1 <- lmer(o ~ pv1o + pv2o + pv1toa + pv2toa + sesblf + (1 |
> prov) + (1 | pm), data = mydata[1:1392,], family = quasipoisson)
>  > m1.1
> Generalized linear mixed model fit using Laplace
>  Formula: o ~ pv1o + pv2o + pv1toa + pv2toa + sesblf + (1 | prov) + (1 | pm)
>    Data: mydata[1:1392, ]
>  Family: quasipoisson(log link)
>   AIC  BIC logLik deviance
>  2622 2675  -1301     2602
> Random effects:
>  Groups   Name        Variance Std.Dev.
>  prov     (Intercept) 0.495021 0.70358
>  pm       (Intercept) 0.047755 0.21853
>  Residual             1.935867 1.39135
> number of obs: 1392, groups: prov, 24; pm, 12
>
> Fixed effects:
>              Estimate Std. Error t value
> (Intercept) -0.207562   0.202715  -1.024
>  pv1o         0.056575   0.006861   8.246
> pv2o         0.023042   0.007216   3.193
>  pv1toa       0.026497   0.003309   8.008
> pv2toa      -0.001075   0.003467   -0.310
> sesblf2      0.307805   0.195291   1.576
>  sesblf3      0.067867   0.194097   0.350
> sesblf4      0.232652   0.194075   1.199
>
> Correlation of Fixed Effects:
>          (Intr) pv1o   pv2o   pv1toa pv2toa ssblf2 ssblf3
> pv1o     0.035
> pv2o    -0.014 -0.568
>  pv1toa   0.017 -0.239  0.050
> pv2toa  -0.104  0.035 -0.267 -0.772
> sesblf2 -0.498 -0.038  0.037 -0.067  0.102
>  sesblf3 -0.483 -0.042  0.021 -0.096  0.085  0.508
> sesblf4 -0.483 -0.052  0.029 -0.099  0.083  0.508  0.518
>
>
> > m1.2 <- lmer(o ~ pv1o + pv2o + pv1toa + pv2toa + sesblf + (pv1o | prov) + (1 | pm), data = mydata[1:1392,], family = quasipoisson)
> CHOLMOD warning: matrix not positive definite
>  Error in objective(.par, ...) : Cholmod error `matrix not positive
> definite' at file:../Supernodal/t_cholmod_super_numeric.c, line 614
>
>
> > m1.3 <- lmer(o ~ pv1o + pv2o + pv1toa + pv2toa + sesblf + (pv1toa | prov) + (1 | pm), data = mydata[1:1392,], family = quasipoisson)
> > m1.3
> Generalized linear mixed model fit using Laplace
> Formula: o ~ pv1o + pv2o + pv1toa + pv2toa + sesblf + (pv1toa | prov)
> +      (1 | pm)
>    Data: mydata[1:1392, ]
>  Family: quasipoisson(log link)
>   AIC  BIC logLik deviance
>  2459 2522  -1218     2435
> Random effects:
>  Groups   Name        Variance   Std.Dev. Corr
>  prov     (Intercept) 0.88071235 0.938463
>           pv1toa      0.00065327 0.025559 -0.712
>  pm       (Intercept) 0.04588791 0.214215
>  Residual             1.76873748 1.329939
> number of obs: 1392, groups: prov, 24; pm, 12
>
> Fixed effects:
>               Estimate Std. Error t value
> (Intercept) -0.3854583  0.2384690  -1.616
> pv1o         0.0545945  0.0069507   7.855
> pv2o         0.0266912  0.0072733   3.670
> pv1toa       0.0369315  0.0063083   5.854
> pv2toa      -0.0008906  0.0034099  -0.261
> sesblf2      0.3326815  0.1908401   1.743
> sesblf3      0.1012759  0.1896883   0.534
> sesblf4      0.1968002  0.1899092   1.036
>
> Correlation of Fixed Effects:
>         (Intr) pv1o   pv2o   pv1toa pv2toa ssblf2 ssblf3
> pv1o     0.016
> pv2o    -0.023 -0.306
> pv1toa  -0.485 -0.143 -0.014
> pv2toa  -0.080 -0.055 -0.306 -0.367
> sesblf2 -0.414 -0.037  0.033 -0.028  0.099
> sesblf3 -0.402 -0.034  0.027 -0.046  0.080  0.509
> sesblf4 -0.399 -0.039  0.031 -0.056  0.079  0.507  0.517

Thanks for including the information that you did.  Would you be kind
enough to rerun the model that failed with the additional argument
control = list(msVerbose = 1).  That will help to show exactly what is
happening to the parameter estimates in the model during the course of
the iterations.