Comparing nominal and real quality scores on next-generation sequencing genotype calls

I seek to comprehensively evaluate the quality of the Genetic Analysis Workshop 17 (GAW17) data set by examining the accuracy of its genotype calls, which were based on the pilot3 data of the 1000 Genomes Project. Taking advantage of the 1000 Genomes Project/HapMap sample intersect, I compared GAW17 genotype calls to HapMap III, release 2, genotype calls for an individual. These genotype calls should be concordant almost everywhere. Instead I found an astonishingly low 65.4% concordance. Regarding HapMap as the gold standard, I assume that this is a GAW17 data problem and seek to explain this discordance accordingly. I found that a large proportion of this discordance occurred outside targeted regions and that concordance could be improved to at least 94.6% by simply staying within targeted regions, which were sequenced across more samples. Furthermore, I found that in certain individuals, high sample counts did little to improve concordance and concluded that quality scores for a certain sample's sequence reads were simply incorrect.