The model thus predicts the intuitive result that increasing the size of a reference panel (internal or external) leads to fewer imputation errors.
A negative value indicates that the external panel will result in fewer imputation errors on average, and a value of zero indicates that the two panels are expected to produce the same number of incorrectly imputed sites.
The basic principle is described by Rubin ([2003]): A few imputations of the first part are created (say M), and then for each of these, several imputations of the second part are created (say N).
Not surprisingly, imputation j resulted in more good quality weeks, and imputation k had fewer, than imputation i. Figures 2 and 3 were similar under all 3 imputation schemes.
Longer haplotypes have a greater accuracy of imputation, but if they are too long they are shared by very few individuals and consequently contribute fewer imputed alleles.
We have interpreted ℙ(C1) as the probability that the internal reference panel is optimal and results in fewer expected imputation errors at a locus than the external panel.
Although transforming the variables prior to imputation resulted in fewer values imputed outside the range than imputing on the raw scale, this did not provide any additional benefit for the Likert and C-GHQ scoring conditions compared with carrying out imputation on the raw scale.
In such cases, quite a few imputed SNPs are in complete LD with typed SNPs, and thus have a SNP-specific imputation accuracy of 1 regardless of the MAF.
Other important inferential quantities such as null hypothesis significance tests, p values and confidence interval half-widths can suffer from substantial imprecision when just a few number of imputations are used [ 23].
Only a few pedigree-based imputation programs can impute non-genotyped individuals in the pedigree, e.g., AlphaImpute [ 4], FImpute [ 16], FindHap [ 17], and PedImpute [ 18].
