I am checking genotyping concordance by using the same sample genotyped multiple times on the same illumina SNP array. Someone recommended that I only look at non-reference homozygous sites for concordance. I have seen some mentions of this online, but to no real explanation for why this would be used.

My intuition is that it is more likely that a non-variant call would be made, thus making use of only non-reference homozygous calls only for the concordance calculation a more stringent measure of concordance. For some reason I can not wrap my head around why this would be the case.

The only thing that I can think of is that the cluster definitions are based more on observations of reference calls, so there is a bias towards those types of calls by default. Is that intuition correct?