I'm a beginner on the method estimating heritability through SNP array data. I found a strange thing: I estimated heritability of thousands of SNPs, but I got larger heritability after I pruned out LD relationship (r2=0.2) and estimated heritibility of independent SNPs. Is that normal? Why we get larger heritability using a subset of SNPs than all SNPs?
Yes, you can estimate the heritability explained by SNPs after doing the LD pruning. But I usually fit all the SNPs in the model. From the Figure 2 in Yang et al. 2010 Nature Genetics, actually I don't think the estimate will change too much whether fitting all the SNPs or a subset of SNPs, unless a very small proportion of SNPs. Have you check the SD? If the SD is large, probably, it is hard to conclude that the estimate after LD pruning is larger.
PS: I usually fit the first 10 or 20 PCs as covariates to correct for genetic structure and population structure when doing the analysis.