Warning messages: 1: In logLik.reStruct(object, conLin)

Question

I am running multivariate linear mixed effects model on lme function on simulated data as:

lme(value ~-1 + cs_d0 + va_d0 + as.numeric(times):cs_d0 + as.numeric(times):va_d0 ,
 random =  ~ -1 + (cs_d0 + va_d0) + as.numeric(times):va_d0 + as.numeric(times):cs_d0  |id, 
 weights=varIdent(form=~1| cs_d0),corr=corCompSymm(form=~1 | id),
 control=lmeControl(returnObject=TRUE,singular.ok=TRUE, opt="optim"), data =D0Train)

just to clear some the variables: cs_d0 and va_d0 are dummy variables . and I used a melt function to combine the dataset.
values have the values of cs_d0 and va_d0.

the problem that I face it is:

Warning messages:
1: In logLik.reStruct(object, conLin) :
  Singular precision matrix in level -1, block 1
2: In logLik.reStruct(object, conLin) :
  Singular precision matrix in level -1, block 1

I don't have any idea what this warning messages does mean!!! also, I think the results is not sensible

Linear mixed-effects model fit by REML
  Data: D0Train 
  Log-restricted-likelihood: -7255.602
  Fixed: value ~ -1 + cs_d0 + va_d0 + as.numeric(times):cs_d0 + as.numeric(times):va_d0 
                  cs_d0                   va_d0 cs_d0:as.numeric(times) 
             29.7567029              29.7567029               0.6767829 
va_d0:as.numeric(times) 
              0.6767829 

Random effects:
 Formula: ~-1 + (cs_d0 + va_d0) + as.numeric(times):va_d0 + as.numeric(times):cs_d0 | id
 Structure: General positive-definite, Log-Cholesky parametrization
                        StdDev       Corr                
cs_d0                   1.623495e-02 cs_d0  va_d0  v_0:.(
va_d0                   2.790607e-03  0.006              
va_d0:as.numeric(times) 1.298373e-03  0.001 -0.007       
cs_d0:as.numeric(times) 1.679235e-04 -0.001  0.000  0.000
Residual                1.093950e+01                     

Correlation Structure: Compound symmetry
 Formula: ~1 | id 
 Parameter estimate(s):
        Rho 
-0.05660665 
Variance function:
 Structure: Different standard deviations per stratum
 Formula: ~1 | cs_d0 
 Parameter estimates:
       1        0 
1.000000 1.000006 
Number of Observations: 1920
Number of Groups: 120

so, in the fixed effects: the intercepts of cs_d0 and va_do are same also the slops have equal output!!! Just I ask it here if any one might help me, please?

Seems like you could have an overfitted model, although it's difficult to tell without access to your dataset `D0Train` or some other dataset the reproduces the problem. See [this link](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example) for some ideas on creating reproducible examples. — aosmith, Oct 05 '15 at 21:30

Warning messages: 1: In logLik.reStruct(object, conLin)

0 Answers0