The RMA paper by Irizarry et al. (2002) specifies how probe sets are summarised (background-correction, quantile normalisation, fit a linear model). I would like to know how metaprobesets are summarised.
The Aroma Project provided vignettes for Affymetrix Expression analysis with RMA and Firma. The summarization step for these two methods use RmaPlm and ExonRnaPlm respectively. The RmaPlm methods use an approach called 'median polish' for summarization. Wikipedia has a basic explanation with a few references. A step-by-step description of the algorithm can be found here and a more formal definition here.
The median polish is an exploratory data analysis procedure proposed by the statistician John Tukey. It finds an additively-fit model for data in a two-way layout table (usually, results from a factorial experiment) of the form row effect + column effect + overall median.