But when I check the same Ancestry member against my own test, it comes back as 15cM/2 segments and the maternal/paternal attribution is given as "Unassigned".
This is something I've noticed. Biggles may be able to ad a bit more on the science, because I am not that well read on it.
That is, you have two DNA matches (child and parent) and the child has more shared DNA than the parent.
The first aspect to understand is on the inheritance of DNA, which is random, within certain limits and aligning with certain norms. We inherit a random mix of DNA from each parent, with a segment from one and a segment from the other. I believe that the average size of an inherited segment is about 20cM, but can theoretically be an entire chromosome. Where I've noted these matches where a child has a larger share of cM to you than the parent is with matches under 30cM. So it's in this area where there is quite a high chance of inheriting a full chunk from a parent that represents a full chunk inheriting from a grandparent.
As you probably know Ancestry has their Timber system to strip out shared cM that it thinks are coincidental (rather than inherited by descent). It seems the most likely cause of these cases is that for some reason Timber doesn't recognise the so much of the shared cM as coincidental. Maybe the adjoining portion of DNA, inherited from both or the other parent happens to be identical by chance. Ancestry tries to strip those out, but maybe can't all the time.
The other possible answer is that you have extra shared DNA from the child's other parent. In my experience I'd say Ancestry's maternal, paternal, both guides for DNA matches in the range of about 30cm+ are close to 100% accurate. But below that it's more likely to get into "unknowns" and assigning people to the wrong side. When this is the case, I find that it's liable to assign many of the matches of a line on the wrong side. Below 30cM the unknowns mount up and it's probably less than 90% accurate with smaller matches of around 8-12cM.
***
On the 9cM match, from what I have seen these have the possibility of being quite deep. I have absolutely certain clusters going back to around the 1650s-1660s on a few lines. In a couple of them, there are matches from child, parent, grandparent, with low cM in the range of 10cM. And the child, parent and grandparent all have roughly the same shared cM.
This relates to the above, where the average inherited size of a segment is around 20cM. The smaller the segment the more likely it is to be inherited over multiple generations at exactly or roughly the same size. There's also regions of DNA that are more or less likely to be split. So when you have matches that show common ancestry going back to the 1600s, they are probably from those regions.
So, what is the maximum number of generations one can inherit these small segments? Certainly to 8th cousins, but presumably more.
But there's also the question of whether a 9cM match on Ancestry may be a false positive. I think there are no to few false positives >= 40cM, 30-30cM - maybe a small amount, 20-29cM - maybe more, 15-19cM - maybe more, 10-14cM - maybe more; and so on. Hard to say. The largest match I have that may be a false positive is 43cM. I have a few clusters of DNA matches where the largest match is in the 20-30cM range. These are always huge clusters with 100s of matches, and the common ancestry always goes back to the Carolinas in the US. I don't know what to make of that. The genealogical record of the Carolinas seems to be poor, with very mixed, jumbled, unsourced and abominable public trees. With clusters where the common ancestry is in the US, it's possible that there was an immigrant ancestor and people have just botched in any old US line, as there's probably no record of place of birth. So maybe those clusters are just muddled, but the size of them suggests that they may be false positives. Since Timber is meant to strip out commonly occurring parts of DNA; and these are very large clusters at lower cM.
When I have matches that are isolated, without other shared matches who share the same ancestry, I note that and don't consider it proof like I would if they were in a cluster with 10 other people who have shared ancestry. In such cases more tests might help. Much more tests in the case of 9cM matches.
I have a line of ancestry from Fife, starting in 1794, with no other ancestry in the area. In my aunts/uncle's DNA tests I found two clusters of matches going back to that area, but with 23cM being the largest shared, few matches and not many with determinable shared ancestry. Recently I got hold of a half 4th cousin's DNA test (who was not a match of my relatives) and he had a few of these shared matches in common, and then other matches. So this had led me to put it down as genetic proof. I was a little cautions to seal it as there are two generations where there are no matches and then a faint cluster. Whereas if you have a lot of matches, you can often link one generation into another into another.
