To further analyse the nature of such type-I error inflation, we describe markers for whom their imputed association statistics were concordant or discordant in relation to their empiric association statistics. On Table 2 the results of such comparison highlight that there is a great concordance between association statistics determined by the allelic frequencies derived from imputed and genotyped information. A total of 317.255 markers were concordant, 317.217 were correctly considered not associated by both empiric and imputed frequencies and, a group composed by 38 markers was considered associated by both measures. Nevertheless, a group of 35 markers showed significant discordance between association statistics. The term discordance in this manuscript refers to differences regarding a pre-defined cut-off value observed in association statistics derived from imputed and directly genotyped allelic frequencies. Interestingly, in the diabetes dataset we found no situation in which an association would be claimed by empirically genotyped markers, but not by inferences relying upon imputed genotypes. The complete list of markers, association P-values and an indication of their concordance in terms of association statistics determined by imputed allelic frequencies can be obtained in Additional file 2, Table S1.
