[R-sig-ME] glmmPQL: std.errors in summary and vcov differ

[Ben Pelzer]

Dear list,

With glmmPQL from package MASS,  I ran a logistic regression model with 
an intercept only, which is random across 33 countries:

m1 <- glmmPQL(MATH_Top1 ~ 1,
               random = list(country = ~ 1),
               family=binomial, data=pisas)
summary(m1)
sqrt(vcov(m1))

There is a total of N=22997 students in data.frame pisas.

The summary(m1) functions shows:

Fixed effects: MATH_Top1 ~ 1
                 Value Std.Error    DF  t-value p-value
(Intercept) -2.176564 0.1140811 22964 -19.0791       0


whereas the sqrt(vcov(m1)) function shows:

             (Intercept)
(Intercept)   0.1140786


My question is why the two std. error estimates 0.1140811 and 0.1140786 
of the fixed part of the intercept differ slightly. The same holds for 
std. errors of the fixed effects of predictor variables, which I did not 
add above for reasons of simplicity. Thanks for any help!

Ben.



	[[alternative HTML version deleted]]