pol and fixing variance components

Hi,

I want to fit a random meta analysis in asreml-R according to the model

y_i = mu + m_i + e_i

where m_i is a measurement error with known standard error (sd_i) and
e_i is an iid residual with a variance to be estimated.

I think that setting up a random regression with

pol(sd, -1, raw=T):units

generates a design matrix Z with the standard errors long the
diagonal. By editing the G.param file I can fix the associated
variance to 1 so that the (co)variance matrix for the data is:

ZZ'+Ve*I

where ZZ' is a diagonal matrix with the measurement error variances
along the diagonal.

If I do this I get (slightly) different answers for the intercept than
I get using other techniques (MCMCglmm, GLS), so I am thinking that
perhaps the pol function is not doing what I think it is. Is it
possible to output the design matrices and/or get some clarification?

Thanks for your time,

Jarrod

PS I've fitted us(dummy):units for the rcov formula, where dummy is
just a vector of ones. This (I think) forces the variance component in
the G.param to be one, rather than being fixed at the residual variance.

-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.