[BioC] residuals.MArrayLM

[Gordon Smyth]

Dear Tiandao,

I don't have a neat answer to that question. A model always tries to 
be good enough for the purpose, and you have to very precise about 
what the purpose is before you can even properly address the 
question. For example, the p-values may still behave well for ranking 
genes even if the data are far from normal. It isn't possible to 
fully put a model to the test on the basis of an individual 
microarray data set. There are many research questions here.

I personally think that with microarray data it is most fruitful to 
examine data quality at the array level, rather than attempting an 
microarray equivalent of the sort of residual analysis you might do 
for a univariate regression problem.

Best wishes
Gordon

At 12:43 PM 16/10/2007, Tiandao Li wrote:
>Dear Dr. Smyth,
>
>Double check. Last time it was my typo. :-)
>
>My question is: from a statistical point, how do we know the linear model
>generated by lmFit is a good analogy of the real data?
>
>Tiandao
>
>
>On Tue, 16 Oct 2007, Gordon Smyth wrote:
>
>At 05:36 PM 15/10/2007, Tiandao Li wrote:
> >Dear Dr. Symth,
>
>Smyth!
>
> >Thank you so much for your valuable help. This is what I was thinking of.
> >May I ask one more question since I searched for days, how do we
> >objectively assess lmFit result(s) using least squares or robust methods?
>
>I don't understand your question, or rather, it could mean a lot of
>things. Can you please try to ask more specifically what it is that
>you want to know.
>
>Best wishes
>Gordon
>
> >Regards,
> >Tiandao