[BioC] spatial correlation and design question

[Gordon K Smyth]

> Hi,
> I am trying to estimate the spatial correlation for a series of arrays.
> My design encompasses 9 separate groups (4-5 technical reps per group),
> and each array has each gene spotted 4 times. When I estimate the
> spatial correlation as per the guide, I only get one number.  Given
> that I have 9 distinct groups as defined by my design, shouldnt I get 9
> separate correlations for use in the gls.series function?

The technical report on within-array replicates which accompanies the
Limma User's Guide explains why there is only one correlation.

This and your previous question really relate to the difference between
least squares regression and one-sample statistics.  You probably need to
consult a local statistician or a stats textbook re multiple regression. 
The question of whether you can consider variability to be constant across
groups obviously applies to the variance as well as the spatial
correlation.

Gordon

> thanks
>
> Simon.