[R-sig-ME] glmmPQL inquiry

[Ken Beath]

I don't know, possibly it is an error in glmmPQL, but I don't think  
compound symmetry is needed, I would expect it is the default.

You could try lmer, it may be better. My preference for binomials  
where there is a high correlation within clusters/subjects is not to  
use Laplace or PQL but adaptive Gauss-Hermite quadrature, which last  
time I looked didn't seem to be available in lmer, contrary to what  
the help says. This unfortunately means Stata or SAS.

Ken


On 07/12/2007, at 8:49 PM, kennedy otwombe wrote:

> Hi Ken,
> With my dataset that has 50 subjects each with three observations, i  
> get the error i gave if i use the compound symmetry covariance  
> structure. Would you have an idea why this is happening?
>
> Kennedy
>
> ----- Original Message ----
> From: Ken Beath <kjbeath at kagi.com>
> To: kennedy otwombe <notwombe at yahoo.com>
> Cc: R-SIG-Mixed-Models <R-sig-mixed-models at r-project.org>
> Sent: Friday, December 7, 2007 11:30:15 AM
> Subject: Re: [R-sig-ME] glmmPQL inquiry
>
> On 07/12/2007, at 7:58 PM, kennedy otwombe wrote:
>
>> Hi Sundar/Ken,
>> I would like to clarify my inquiry. The data structure i gave was
>> just a sample of my data. I actually have 50 subjects each with
>> three binary observations (hence 150 observations). So the error i
>> initially gave emanates from the analysis involving all the 50
>> subjects. It just turns out that for the sample i gave, the first
>> two subjects have only 0 as their entries but this varies as you
>> look through the data.
>> I am using the latest version of R i.e 2.6.1 and i downloaded the
>> nlme version currently available on the CRAN website. I am aslo
>> assuming my random part is the intercept.
>> Hope this clarifies my inquiry.
>>
>
> If I modify your code to include better values for y, using code
>
> library(MASS)
> navs<-data.frame(matrix(c(1,1,1,0,0,1,1,
> 2,1,0,0,1,0,1,
> 3,1,0,0,1,0,1,
> 1,2,1,0,0,1,0,
> 2,2,1,1,1,1,0,
> 3,2,0,1,1,1,0),nrow=6,byrow=T))
>
> names(navs) <- c("t","id","y","x0","x1","x2","x3")
>
> fit<-glmmPQL(y~x1+x2+x3, random=~1|id, family=binomial, data=navs)
> summary(fit)
>
> the results are fine, except for some problems due to not having
> enough subjects, as in the following output.
>
> Ken
>
>> library(MASS)
>> navs<-data.frame(matrix(c(1,1,1,0,0,1,1,
> + 2,1,0,0,1,0,1,
> + 3,1,0,0,1,0,1,
> + 1,2,1,0,0,1,0,
> + 2,2,1,1,1,1,0,
> + 3,2,0,1,1,1,0),nrow=6,byrow=T))
>>
>> names(navs) <- c("t","id","y","x0","x1","x2","x3")
>>
>> fit<-glmmPQL(y~x1+x2+x3, random=~1|id, family=binomial, data=navs)
> iteration 1
> iteration 2
> iteration 3
> iteration 4
> iteration 5
> iteration 6
> iteration 7
> iteration 8
> iteration 9
> iteration 10
>> summary(fit)
> Linear mixed-effects model fit by maximum likelihood
>  Data: navs
>  AIC BIC logLik
>    NA  NA    NA
>
> Random effects:
>  Formula: ~1 | id
>          (Intercept)  Residual
> StdDev: 8.009601e-05 0.5773503
>
> Variance function:
>  Structure: fixed weights
>  Formula: ~invwt
> Fixed effects: y ~ x1 + x2 + x3
>                  Value Std.Error DF      t-value p-value
> (Intercept)  0.000353  6171979  2  5.700000e-11      1
> x1          -30.566351  2631792  2 -1.161427e-05      1
> x2          30.565997  4161051  2  7.345739e-06      1
> x3          -0.000071  3721849  0 -1.900000e-11    NaN
>  Correlation:
>    (Intr) x1    x2
> x1 -0.853
> x2 -0.944  0.632
> x3 -0.905  0.707  0.894
>
> Standardized Within-Group Residuals:
>          Min            Q1          Med            Q3          Max
> -1.732051e+00 -5.451298e-17  8.078053e-16  1.506878e-15  1.732051e+00
>
> Number of Observations: 6
> Number of Groups: 2
> Warning message:
> In pt(q, df, lower.tail, log.p) : NaNs produced
>
>
>       
> ____________________________________________________________________________________
> Looking for last minute shopping deals?
> Find them fast with Yahoo! Search.  http://tools.search.yahoo.com/newsearch/category.php?category=shopping
>