of the r2 statistic. We plotted separately the four statistics against the observed bias between association statistics of the complete set of markers (Figure 4). The analysis of the plotting organized in Figure 4 outlined that the use of the max value r2 or the variance within a haplotypic block cannot be considered good predictors since the bias is randomly distributed throughout the variable values range. Otherwise, when the same procedure was applied to the mean and median values, significant insights could be obtained from the analysis. More prominent biases are concentrated in SNPs with relatively lower values, especially in the case of the median. Polymorphisms with lower mean or medians for r2 measures have increased odds of showing higher deviation when the empiric and imputed association P-values are compared. Imputed markers showing these specific haplotypic conditions should be analyzed carefully before their use.
