Weighted analysis of METs

Dear all,
I am trying to analyze the data from METs set up using different trial designs on different sites. The replicate data for all trials were not available, therefore it could be carried out as the analysis of G x E means (estimates) only. However, there is one option for improvement, using the attached standard errors (which are available) as weights.

Basically, starting model could be simple one:
yield ~ !r gen*env

When the weight variable (reciprocals of squared standard errors) is added to the model, it improves the fit, keeps all of the variance components at similar level, except it dramatically inflates the residual variance. Chapter 6.7 from the manual addresses this problem (I believe), giving a suggestion to use !S2==1 qualifier to fix the unit variance? All my attempts to do so have failed so far, therefore I would be grateful for any help.

Thanks,
Jerko

-------------------- m2f --------------------

Sent using Mail2Forum (http://www.mail2forum.com).

Read this topic online here:
http://www.vsni.co.uk/forum/viewtopic.php?p=655#655

-------------------- m2f --------------------
Received on Wed Jul 21 2009 - 13:55:35 EST