The structure of the dataset is more complex in this scenario than in the previous one, but we follow the same basic principles of imputation: phase the observed data, then impute alleles in each haplotype separately, conditioning on as much observed data as possible. Here, the goal of the phasing step is to end up with three sets of haplotypes: , the known haploid reference panel haplotypes at SNPs in T, U1, and U2; , the unobserved diploid reference panel haplotypes at SNPs in T and U2; and , the set of unobserved study haplotypes at SNPs in T. If there are NDR individuals in the diploid reference panel, their haplotypes can be represented as , where is the haplotype pair for diploid reference individual i.
