This work is about the design and statistical analysis of gene expression experiments with variable pool sizes. Conditions for unbiased contrasts in such experiments were derived. They were analyzed with linear mixed models, introducing a new type of variance component that accounts for unequal shares of individual samples in common pools. The relevance of this blending error variance was then investigated for four experimental data sets from different species. Finally a biomarker search for hygienic behavior in Apis mellifera workers was conducted, using nerve tissue gene expression data.<eng>
