Hi Sisay, 

What do you mean by data structure? How should your table structure be to be loaded into GenStat? If so, it should follow the same table structure as it would be used for ANOVA. I am not sure if this is your question, but if you clarify, I might be able to help you.    

Best, 

Geraldo

Thanks Geraldo for your reply and interest to help.

by 'data structure' I was referring to the way to tell the model components to Genstat ; My design like I mentioned in my question is alpha lattice which includes incomplete blocks in each replication. for ANOVA , I think the blocks are complete in a sense that all the individuals to be tested are included in each block. But in my case only 30 accessions are included per block. So my question is how to introduce Block as a factor in the mixed model filed in Genstat?

thanks before hand for your help!

sisay

I believe your model would then be:

response = location + replicate within location  + incomplete block within replicate + genotype + error

You can assume the incomplete block, genotype and the error to be normally distributed (chek this assumption you might need do some transformation) with location dependent variances, i. e., the variances would be different among each environment. For the genotype term, you can assume a covariance between treatments, i. e., between each environment. Use these variances to calculate your repeatabilities if you want to. 

Then re-run this model by assuming genotype as fixed to generate your genotypic BLUEs. 

A simpler approach would be using the model below:

response = location + replicate within location  + incomplete block within replicate + genotype + genotype x location + error 

Decide which term you judge to be considered random besides your genotype (to estimate overall BLUE across env.) and/or GxE (to estimate BLUEs by env.) that should be fixed to estimate your BLUEs.

I am not 100% sure about this, but I think this model should be appropriate in your case. Does anybody agree? 

Best, 

Geraldo

Dear Geraldo,

Many thanks indeed!

I will try these options you suggested me and hope it will help.

Sisay

Almost there.

1) Your incomplete block has to be random, and make sure you get a different variance component for each of your sites, otherwise you combine sites with different scales.

2) You need to make your error term (residual variance) different for each site, this means you allow for heterogeneity across sites. This can be done with GenStat using the outside=TRUE command from REML.

Good lcuk

Salvador

Thanks for your input Salvador. 

Cheers, 

Geraldo