[BioC] Limma design : What is the

[Santy Marques-Ladeira]

Gordon K Smyth <smyth at ...> writes:

> 
> Dear Santy,
> 
> The first method is twoway anova, a generalization of a paired analysis.
> 
> The second method is a random effects approach in which the intra-donor 
> correlation is incorporated into the covariance matrix instead of the 
> linear predictor.
> 
> Both are good methods.  The twoway anova approach makes fewer assumptions 
> but the random effects approach is statistically more powerful, 
> particularly for unbalanced designs.
> 
> For a balanced design in which all donors receive all stimuli, the twoway 
> anovao approach is virtually as powerful as the random effects approach 
> and hence is preferable.
> 
> For an unbalanced design in which each donor receives only a subset of the 
> stimula, the random effects approach is more powerful.
> 
> Your experiment is almost completely balanced -- there is just one missing 
> stimuli for one donor.  Hence I would use the twoway anova approach.
> 
> Best wishes
> Gordon
> 



Dear Gordon,

First I must apologize for the delay in my answer but since I'm new around
here, I had some technical issues.

Thanks for your answer. In my service everyone use only the "two way anova
approach" to infer the "donor effect" but I'm glad to know that for
"unbalanced" experiments I should use the "random effects approach". 

Best Regards,
Santy