About the 2.5% IBD cutoff?

I am just curious about why 2.5% was chosen, not any other value?

This is a rule of thumb to exclude individuals who show excessive relatedness with others in your sample. The reason why you want to do this is twofold:

  1. (Closely) related individuals also share common environments. If you include closely related individuals in the analysis you could introduce confounding into your analysis from the common environment
  2. Closely related individuals act as outliers in the analysis and carry disproportional weight. This has the effect that the variance components which normally represent tagged genetic variation only (c.f. SNP heritability), will represent all genetic variation (ie more akin to traditional heritability)
    Hope this makes sense.