Paul, I'd like to see the distribution curve of [percent of n instances of a shared segment] on the y-axis, vs cMs on the x-axis for each cousinship (3rd to as far out as is reasonable, maybe to 10th cousins. Actually the ultimate graphic would be for these 8 curves to be on one chart. So if 4th cousins would show an average of 14cM, we would expect the curve to peak at 14cM (the most percentage of say n=1,000 or 10,000 instances); and to be less at 13cM and 15cM and have tails in both directions - perhaps show the 2 S.D. spread) It might be a pretty wide chart for 1-3 cousins; and I think the interest is in what the curves, 2 S.D.s and tails would look like for the more distant cousins. I don't know how to scale the y-axis to reflect the fact that 4th cousins only match (above a threshold) 50% of the time, and 6th cousins only 2% of the time.... Maybe just the curve for when they do match and then we'd need to multiply by the probability of matching at all. This set of curves would give us something concrete to go on. Over time we might be able to gather some real data to validate or tweak the curves. These curves would also show the effect of using a 5cM or 7cM or 10cM or 15cM threshold in our spreadsheets.... Thanks for taking this on. Jim Bartlett www.segmentology.org On 09/15/15, Paul Rakow via<genealogy-dna@rootsweb.com> wrote: Yes, we can calculate the spread too. I'm thinking of putting up some tables and graphs from this simulation. I would like some discussion and input on what would be most useful. What is the best way to measure the spread in a quantity? What would be the easiest to understand? Standard deviation? Or upper and lower quartiles (half of your cousins will lie between X and Y, one quarter below X, one quarter above Y)? I plan to plot some histograms too - some of the distributions are a long way from the normal distribution bell curve, and it can be useful to know what the tail looks like. ---- Which are the chief quantities of interest? I thought of: Total shared cM Largest matching segment Average number of matching segments Probability a relative will fail to match you. Any others? ------ In the cases where a threshold matters, what would be a good number to use? I was thinking of using the gedmatch default of 7 cM, are there other interesting values? Thanks, Paul Rakow On Tue, September 15, 2015 18:20, Wjhonson wrote: > > But I would also be interested in what values are obtained at the edges > of 1 standard deviation > > If we have one hundred fourth cousins and ten of them test... will we > find one that matches us at 70cms? > > That's the sort of thing that could be useful information so we aren't > chasing ghosts > > > > > > -----Original Message----- > From: Paul Rakow via <[1]genealogy-dna@rootsweb.com> > To: genealogy-dna <[2]genealogy-dna@rootsweb.com> > Sent: Tue, Sep 8, 2015 12:56 pm > Subject: Re: [DNA] 9th generation common ancestors > > > > > Anne - thanks again for the link about longest segments. > > > I've built a > measurement of the average length of the longest segment into the > simulation, and tried it out for a few cases. > > For cousins: > > > Simulation: > > > 1st > cousins, 91 cM , average length of longest segment. > > 2nd cousins 49 cM > > > > 3rd cousins 24 cM > > > > Observed numbers (Shared cM project, Blain > Bettinger) > > > 1st cousins, 83 cM > > > 2nd cousins 49 cM > > > 3rd cousins 27 cM > > --- > > > I would consider that a success, everything within about 10%. > > > > For ancestors/descendents > > > Simulation > > > Grandparent/grandchild > 193 cM average longest segment. > > > Great-grandparent/great-grandchild 131 cM > > (This includes lots of cases where the "longest matching segment" is > an entire chromosome). > > Observed numbers (Shared cM project, Blain > Bettinger) > > > Grandparent/grandchild 170 cM average longest > segment. > > Great-grandparent/great-grandchild 102 cM > > > > So, for > descendents the simulation is looking a bit high, but not absurdly so. > > > Paul > References 1. mailto:genealogy-dna@rootsweb.com 2. mailto:genealogy-dna@rootsweb.com