Is it necessary that both the genotyed and imputed SNPs explain more variance than only genotyped SNPs for the same samples? In my dataset, I found the genotyped and imputed SNPs (5M SNPs) explain less variance than genotyped SNPs(76K SNPs). Is it reasonable?
It is possible. First of all, how different are the estimates? You may ask whether the difference is just due to sampling. Secondly, if there is a real difference, it's also possible because the properties (e.g. LD and MAF distribution) of the genotyped SNPs are different from that of the imputed SNPs. If the properties of the causal variants are more similar as that of the genotyped SNPs than that of the imputed SNPs, you will probably see a larger estimate using the genotyped SNPs than the imputed SNPs.