The expression values should be compared at log2 scales.
Quantile normalization is optional and is recommended for microarray data.
If you are not sure, click View Data to see if the data needs normalization.
EdgeR uses negative binomial model is to estimate the dispersion parameter for each tag, a measure of the degree of inter-library variation for that tag.
Estimating the common dispersion gives an idea of overall variability across the genome for this dataset.
Estimating the tagwise dispersions estimates a distinct, individual dispersion for each gene.