
Family Finder Advanced Topics Advanced discussion about Family Tree DNA's Family Finder Product. 

Thread Tools  Display Modes 
#11




Can you point to a specific instance matching every detail in that scenario from your personal experience? Or is that just a generalized 'gut feeling'?

#12




Quote:
I don't think you are going to prove NPE with what you are trying to do based off math as 4th cousins are too far removed to make a determination based off a non match between 2 people. If you can't find a YDNA participant can't you find other descendants to test with Family Finder? 
#13




Thanks.
I'm assuming that the degrees of cousinship that you mention are through their direct paternal lines. I'm stipulating that specifically because I have a lot of cousinmarriages in my own family tree where the most recent relationship between agnatic relations is NOT in their direct male lines. In the specific case of 5th cousins, all through the direct male line, my model predicts at a ~96% confidence level that the true odds that any two of them will register as matches to one another on Family Finder are ~81%. That implies that the chance that only two of five randomly selected cousins from this pool match is ~5%. In the specific case of 4th cousins once removed, both through the direct male line, my model predicts at a ~94% confidence level that the true odds that any two of them will register as matches on Family Finder is 100%. That leaves a ~6% chance that the true odds that any two randomly selected cousins meeting these conditions match in Family Finder is less than 100%. So pretty similar cases. A small but nonnegligible chance of them not matching. I never said that I would conclusively prove an NPE for cousins this distantly related through autosomal DNA alone. Only that these scenarios could indicate a heightened probability. I was soliciting actual case histories for this very reason. Just last night I asked for specific feedback on my calculation of the confidence levels. These cases are very interesting, but probably not completely decisive in and of themselves, either. It's a sample size of 7 individuals, and they're certainly not randomly drawn. If these men are like the typical Y DNA project participant, several of them are likely immediate relations to one another. And in any case, their selection is biased because I specifically asked for cases which ran counter to my model. Most likely I need to revisit my confidence calculation. But the significance of the differential in male and female recombination rates, applied exponentially, suggests that there's something here worth defining more clearly. Last edited by Frederator; 15th July 2017 at 01:40 PM. 
#14




Another thought is that my minimum sharing threshold of 7cM is too small. I think that strictly speaking, the absolute minimum is 9cM, although people with largest segments of 7.5cM may match provided their total shared cM is >=20 cM.
My formula is indifferent as to distinctions of the size of individual segments inherited from the MRCAs. I didn't really regard either of these as a particular hindrance because I was able to replicate FTDNA's published generic matching predictions. Of course, that's not saying that FTDNA's figures are necessarily correct, either. Similar products return significantly different predictions according to ISOGG. https://isogg.org/wiki/Cousin_statistics They're not immediately comparable because they do test different arrays of SNPs and have different matching thresholds, but it's something to consider. 
#15




Yet another thought crossed my mind from the very beginning of this project. Namely, that what FTDNA published was NOT the odds that any two randomly selected cousins would match, but rather the odds that enough of the MRCAs' DNA would survive intact among ALL their descendants that at least two of them would match. That's very different.
It's a depressing thought because it would imply that the odds of any two randomly selected cousins register as matches is so small beyond 3rd cousins as to be not worth considering. But I have a hard time believing that. I wouldn't like to think the companies would have published the statistics that they did if that were so. It would be grossly misleading. 
#16




No, it's almost certainly in my calculation of the confidence level.
I'm pretty sure that my calculation doesn't reflect the impact of an unusually low paternal grandfather's contribution early in the chain of descent from the MRCAs. I think my model may be an adequate reflection of the aggregate average contribution %, but there may simply be no coming back from a sharp drop early in the process, with respect to the target MRCAs. I don't have a very good idea what the best answer would be. But at least I can provide a lower bound. Using the singletrial distribution, I can say that in the case I originally posited, the odds of an NPE would be only ~67%, as compared to ~96% using the binomial distribution with the total # of grandfather relationships in the line of descent as the # of trials. I don't think the chance of an NPE could be less than ~67%, if the fact that I replicated FTDNA's generic results means anything. But ~67% would still leave a 1 chance in 3 that the true probability of matching was less than 100%. There's a lot of room in there. Last edited by Frederator; 15th July 2017 at 02:54 PM. 
Bookmarks 
Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)  
Thread Tools  
Display Modes  

