Once the DNA samples are hybridized to the arrays and the arrays are read and quantified using some detection method [61], genotypes are called using algorithms that are specific to the different arrays. Some decisions must be made during this phase to trade off sensitivity for specificity and may lead to missing data if the algorithm cannot distinguish among the three genotypes with certainty. Interestingly, it has been noted that excessively stringent conditions on acceptable genotype calls may introduce bias and inflate the false positive rate because the pattern of missing data may be informative rather than random. For example, rare genotypes may be more likely to be missing than common genotypes [46]. On the other hand, excessively relaxed conditions on acceptable genotype calls may introduce erroneous genotypes in the data to be analyzed and cause spurious associations due to technical errors. A good heuristic seems to opt for more relaxed thresholds on genotype calling to maximize the power of the study followed by careful assessment of the quality of the genotype calls of those SNPs identified in the analysis. Inspection of the plots showing the clusters of genotypes can help to detect these technical errors.
