In contrast, the strategy proposed in this study, which imputes based on the intersection of SNPs genotyped on all arrays represented in the combined sample, showed no evidence of bias. We are not aware of any other method that eliminates the imputation-induced bias without some study samples being simultaneously genotyped on all the arrays being used for imputation. Estimating from more information is generally expected to provide better statistical estimates than estimating from less. For this reason, imputation using the union of SNPs available across the genotyping arrays in the studies to be combined could be expected to produce the best imputation results. However, it is known that differing haplotype information quality generates differences in imputation accuracy (Almeida et al. 2011). Extending this observation to different arrays across which there are differing amounts of genetic information (i.e., numbers of SNPs) or differing types of genetic information (i.e., differing SNP selection strategies used for Illumina and Affymetrix), one would expect differing imputation accuracy results from the different arrays. Combining imputation across arrays with differing inputs seems likely to generate systematically differential
