I was thinking about haplotype association analysis recently.

I was thinking that a lot of haplotype association methods use an EM algorithm independent of case control status under the null of no association.

I was then wondering if there are cases in which there are rare disease associated haplotypes that are absent or nearly absent from reference studies used for imputation purposes.

I reasoned, if this is true, one might see less extreme association summary statistics for imputed variants than for genotyped variants on such a haplotype, if say the imputation algorithm is selecting the best fit but the true fit is not represented.

So, last I wondered if anyone has ever tested the average test statistic for genotyped verses imputed variants across a large number of well-powered GWAS studies.

I conducted an initial search, but did not find a manuscript like this. Is any one aware of this?

To be clear - it need not address haplotypes. I am just interested in studies that looked at genotyped vs imputed variants in one or more GWAS.