[R] glmmBUGS fails to accept higher level covariates

dunner ross.dunne at tcd.ie
Thu Oct 27 19:17:16 CEST 2011


Hello 

I am using glmmBUGS to fit a multilevel model. Treatments are nested in
Course are nested in Patients. The predicted variable in total EEG duration.
The predictors are:

at the observation level : Medication dose
at the Course level: Weight in KG and Age
at the Patient level: Weight in KG and Age 

I am trying to fit a multilevel model as in lmer, but in BUGS. Here is an
example of the model I want to run:
Linear mixed model fit by REML 
Formula: totalEEG ~ workDose + (1 + WEIGHTKG + AgeYrs | MRN/COURSE) 
   Data: book 
  AIC  BIC logLik deviance REMLdev
 7041 7112  -3506     7004    7011
Random effects:
 Groups     Name        Variance   Std.Dev.   Corr          
 COURSE:MRN (Intercept) 6.5755e-06 2.5643e-03               
            WEIGHTKG    1.9015e-11 4.3606e-06 -1.000        
            AgeYrs      1.1138e-09 3.3373e-05 -1.000  1.000 
 MRN        (Intercept) 5.0897e+02 2.2560e+01               
            WEIGHTKG    2.8231e-02 1.6802e-01 -1.000        
            AgeYrs      8.1881e-04 2.8615e-02  1.000 -1.000 
 Residual               2.4965e+02 1.5800e+01               
Number of obs: 818, groups: COURSE:MRN, 114; MRN, 103

Fixed effects:
             Estimate Std. Error t value
(Intercept) 51.721608   1.669860  30.974
workDose    -0.010632   0.003246  -3.275

Correlation of Fixed Effects:
         (Intr)
workDose -0.663


bgs.toteeg<-glmmBUGS(data=book, observations="totalEEG",
covariates=list(MRN="AgeYrs", COURSE="WEIGHTKG", observations="workDose"),
effects=c("MRN", "COURSE"), family="gaussian",   modelFile="model.bug")

however, this is failing with:

Error in glmmBUGS(data = book, observations = "totalEEG", covariates =
list(MRN = "AgeYrs",  : 
  unused argument(s) (observations = "totalEEG", covariates = list(MRN =
"AgeYrs", COURSE = "WEIGHTKG"))

I have already run models with multiple predictors at the lowest level. 
glmmBUGS parameterises and runs a WINBUGS model fine. however, this full
mixed model seems not to work.

When I tried:

> bgs.toteeg<-glmmBUGS(data=book, totalEEG~workDose, reparam=c(MRN="AgeYrs",
> COURSE="WEIGHTKG"), effects=c("MRN", "COURSE"), family="poisson",  
> modelFile="model.bug")

IT compiled the WinBUGS model fine, but winBUGS stalled on an error, not
recognising a node "xobservations".

I'm learning, so it's not just a case where I can "step-up" and model it
directly in Winbugs.

R 2.13.2 on Win 7 i3Intel with lmer, nlme, R2WinBUGS, BRugs, lattice,
attached. Winbugs version 14.3. (Which I know is working fine - Brainware
problem most likely)

Thank you 

Ross

ross.dunne at tcd.ie


--
View this message in context: http://r.789695.n4.nabble.com/glmmBUGS-fails-to-accept-higher-level-covariates-tp3945251p3945251.html
Sent from the R help mailing list archive at Nabble.com.



More information about the R-help mailing list