Differential error induced by imputation may yield SNPs that appear to differ substantially between cases and controls purely as a result of the imputation. Past studies have shown that differential genotyping error between cases and controls can inflate Type I error rates (e.g. Moskvina et al. 2006). A recent study by Sebastiani et al. (2010) which built a model using 150 SNPs to predict longevity has been criticized for not controlling for different chips used with different frequencies between cases and controls. Critics suspect that many of the significant SNPs it identified are artifact of differential genotyping errors between these different chips (Alberts 2010; Carmichael 2010).
