FINDING THE VALUE OF N

THE NUMBER OF SEGMENT-MATCHES WHICH PREDICTS A FAMILY CONNECTION

To achieve 219 false segment-matches for both randoms and relatives, a value for N of 8 or 9 works best. Perhaps the conservative approach would be to keep all those white and green pairs with 9 or more segment-matches. That relatives and randoms have the same number of false segment-matches is a corollary of the obvious assumption that the 187 extra relative segment-matches are all real.

The method here is to delete from the relative match-count all segment-matches from white and green relative pairs having less than N segment-matches, and then totalling the deleted segment-matches to compare with the target. This method depends upon 2 elements - (1) the false segment-matches included in the remaining areas and (2) the real segment-matches included in the deleted areas - balancing each other.

This is something unproven but which we judged reasonable - though perhaps the chance of false matches arising in the selected areas (not just the white or green pairs with N or more segment-matches, but also including a few in yellow - for example where our triangulation discussion shows a false segment-match), is somewhat greater than real segment-matches being included in the discarded area, in which case the green arrows should be a little higher. The error would be no worse than the error (60) for the randoms.