I had a look at this a few years ago and came to the conclusion that there were too many random variables in the calculations to derive anything meaningful. This is what I found -
Thanks for your input, I was looking at a point at which you can presume there will be a differentiation. From the comparisons I have, it's somewhere after 90-110cM, and although the range can be quite large, multiplying the MyHeritage figure by 0.6 is a decent average. So a seemingly pertinent 60cM, becomes 36cM.
I also feel anything below this 90-110cM range on MyHritage might be a false positive. And I tend to think that about 50% or more of matches in the 20-60cM range are false positives.
Of my Ancestry matches 20cM-self, of the ones that I can parse out the tree for enough generations I'd say the hit rate of finding a relation or what looks to be an NPE is around 80-90%. While the same on MyHeritage is closer to 10-20%. There are useful matches on MyHeritage, but it's a lot more work to flush them out.
depending on whether it hits an area Ancestry arbitrarily deem 'too matchy'.
On this, I do wonder how often Ancestry might be stripping out chunks of DNA they'd be better off leaving in. On 23AndMe I have one match who is a 4th cousin to my aunt's test and 98cM, yet on Ancestry her mother is just 8cM.
Having gone though over 1,000 matches in multiple tests, I was for quite a while under the impression that either a 2-3X great-grandfather (Breese) might have been born via infidelity. Because there appeared to be zero DNA matches who correlated to any line of his ancestry and overlapped with other relevant matches. It was only after I began going through matches systematically figuring out how all are related that I discovered that my largest cluster of matches, from a 3X great-grandmother (Bennett), had multiple matches who almost all shared two or more lines of ancestry. So they all just got lost in a huge cluster, when in reality there were at least three clusters in there, with 80-90% of them sharing multiple lines. So once I found one connection, they were put aside and I didn't go back and look for more.
Further, on this Breese line I thought may be an NPE, there are some lines that come from it that I'd generally expect to have yielded DNA matches on Ancestry, but as it stands I have zero. Yet I have matches on those lines from MyHeritage, which overlap on segments with closer Breese matches there (i.e. 3rd cousin, 4th cousin).
So in this case I lean to believing that Ancestry has stripped out too much DNA. So, it would be good if they had a feature to also show your DNA matches before Timber has stripped out segments. I imagine the majority of them would be false positives, but it seems likely that some of them would not. You could pull up the results of the ones that have a common ancestor found in the tree, which would probably throw up some more likely connections.
The Timber feature is great, as it does seem strip out all or close to all of the false positives, at least down to 20cM.
I haven't looked into the clusters of matches you can now see with ProTools, below 20cM. So can't say if they are also low/free of false positives. But from what I can see - it seems the smaller the cM of the primary (or largest cM) match in a cluster - the more distant the relationship is likely to be.
I find, for example, you can have a cluster of matches where the largest cM is 70 (a 3rd cousin) and then there are other matches down to 8cM (with relationships as far back as 6th cousin). While if the largest cM match in a cluster is 20-21cM, then the relationships seems to be pushed back a bit further of the conceivable range. i.e. a 21cM match could be a 2C1R, but in these, from what I've seen the relationship is always further back than you'd expect.