Ancestry matches by parent update Aug 2023

[JanF]

Is anyone else bemused by the latest update on Ancestry DNA matches. Before this week the majority of my proven maternal and paternal matches agreed with Ancestry. I did have an issue with my dad's brother's 4 children who have all tested. 3 of them were designated as maternal matches. But as we have many relatives in the same locality, I was looking for a link with my mum.
This week, there has been a further update and many of my 'proven' relatives are now switched around. This hasn't happened on my other half's tree. I am contacting Ancestry in the hope of getting this looked at.

 

[Stuart]

I have "Updated Aug 2023" too, and beside that "1,508 labelled". That's twice as many as they managed to transfer last time (January), but the pattern is the same - new parent 1/2 assignments are roughly proportional to the number I had already. The new unassigned number is almost the same as it was after last time, so they are just doing the assignments they could have done already.

What I think happens is that they do their processing on all your matches, which labels much (but not all) of your DNA as being from parent 1 or 2. Then they apply this labelling to you matches, based on the DNA you share, and assign all the ones they can. Now, they could continue to use this labelling for each new match as it comes up, but they don't - they park them as unassigned until the next bout of monster maths.

When they do the processing again, with perhaps 5-10% more matches, it should give a very similar labelling. In that case almost all of your parent 1/2 assignments - the ones you had before and the new ones - should be the same as with the old DNA labelling. But there is always the possibility that, for one user's matches, this processing will produce a result that's significantly different. If it's much better, then fine - you've been lucky - but it could be much worse. You might hope Ancestry could check for this: i.e. measure if the new solution is "better" or "worse" than the previous one. However, I don't think they have any measure of "better" so they can't do that.

So it's probably just bad luck if your assignments have got worse (i.e. more of them wrong) - at least for now; maybe Ancestry will get better at this. Telling them that's what you have seen can only help, especially if they really do have no way to work this out for themselves.

 

[Winksetter]

Is anyone else bemused by the latest update on Ancestry DNA matches. Before this week the majority of my proven maternal and paternal matches agreed with Ancestry. I did have an issue with my dad's brother's 4 children who have all tested. 3 of them were designated as maternal matches. But as we have many relatives in the same locality, I was looking for a link with my mum.
This week, there has been a further update and many of my 'proven' relatives are now switched around. This hasn't happened on my other half's tree. I am contacting Ancestry in the hope of getting this looked at.

This is is all a bit like their ethnicity estimates to me - worth ignoring. Some of mine are OK, some are not, most are not assigned anyway. Most clearly identified in Thrulines are not identified or wrongly identified at all in this. It looks like a bit of window dressing to me.

 

[Stuart]

This is is all a bit like their ethnicity estimates to me - worth ignoring. Some of mine are OK, some are not, most are not assigned anyway. Most clearly identified in Thrulines are not identified or wrongly identified at all in this. It looks like a bit of window dressing to me.

Remember that the processing that sorts DNA by parent was developed to provide better phasing, so it is now part of Ancestry's standard matching process. What comes out of that is what it is - so they have tried to make that look useful to us, as a marketing exercise. With limited success, as you say.

My parents came from opposite ends of Great Britain, thus from widely separated ancestral populations, which should make SideView more reliable. It also means I find its results helpful, within limits.

What might be more useful is reliable assignment of "both sides" matches. Now I have just acquired some more of these: before this update I had one - my niece - now I have 21. Those new ones are not big (all under 20 cM), and none of them has a common ancestor flag, nor is one I've worked out and confirmed, and none has any shared matches.

That separation of my parents' backgrounds makes true "both sides" matches unlikely, particularly so in earlier generations, and I've found none so far. I suspect most if not all of these new assignments are "noise" false matches, and the change is due to Ancestry lowering the threshold for accepting a match as "both sides".

 

[IanL]

I agree with Stuart. I also have parents with widely separated geographies. Previously I had no matches allocated to both sides. Now I have 19, all under 20cms. Some time ago I set up what I called my "possible IDS match" group. All low cm amounts, who have shared matches with each other but who, so far, have no matches with anyone in my extensive tree. I recognise that I may just not have found the link yet. Quite a few of the "both sides" list appear in my "possible IDS match" group.

 

[peter]

I linked to this blog entry last year which explained how matches are assigned.

Note that if you only share one segment with a match it will be designated as 'Both sides' if you inherited that segment from both parents. You can check whether your parents shared DNA at GEDmatch (mine didn't)..

The other circumstance is: "A match will be assigned to both sides if two or more segments are labeled from one parent and two or more segments are labeled from the other parent." I think this may well have been relaxed as Stuart suggested above - all but 2 of my 18 both sides matches share precisely 2 segments with me.

 

[Pauline]

There's no evidence in my DNA that my parents were related, and mostly I have found 'by parent' feature useful. I do now have a few matches showing as both sides, but apart from my sister (who is related to me on both sides), these are all 16 cM or less. My sister has a few more matches identified as both sides, including two at just over 30 cM, one across 1 segment, the other across 4. I have identified common ancestors with the latter but not the former. However, these 'both sides' matches are a very small proportion of the whole - less than 0.1% - so not really worth worrying about.

Where I do find the assignment particularly useful is when a match has shared matches on both sides of my tree, and while being aware of the possibility of error in the assignment, it helps to know which side to start on.

 

[peter]

Considering SideView generally, I'm really only interested in matches that can help me knock down 'brick walls', and if there's someone who has one of my ancestral surnames in their tree it's helpful to know whether Ancestry thinks we are connected on the relevant side of my tree. Even if Ancestry are right only 90% of the time, it still saves me a lot of wasted effort.

Going back to 'both sides' matches, I have one 'both sides' match, with no tree, who shares 87cM (unweighted) across 5 segments. We have 14 shared matches, one of whom is my brother, but all of the other 13 have been identified by me as on my father's side, or have been assigned to that side by Ancestry.

Assuming there is consistency across different aspects of SideView this suggests that there is at least one segment I share with this 87cM match which isn't common to any of the 13 shared matches. Since I have already identified some distant documented cousins who are related to me on both sides of my tree (or vice versa) it wouldn't be at all surprising if Ancestry found some through DNA. The GEDmatch check I referred to above can only look at the DNA we inherited from our parents, so only positive results are meaningful - remember that even 3rd cousins may not share any DNA.

 

[Helen7]

I think this may well have been relaxed as Stuart suggested above - all but 2 of my 18 both sides matches share precisely 2 segments with me.

I agree the condition for 'both sides' must have been relaxed. I now have 39 of these, all except 4 of them across just 2 segments. The exceptions are my son (obviously correctly labelled as both sides), 2 others between 25 and 35 cM across 3 segments (both of whom I've identified as connected on my maternal side) and one of only 8 cM (but 20 cM unweighted) across 3 segments. Basically I'm ignoring the 'both sides' allocation as it's not very helpful.

 

[peter]

Basically I'm ignoring the 'both sides' allocation as it's not very helpful.

It's not very helpful, but nor would I ignore it. It's flagging up that Ancestry have allocated a segment as being from the side you weren't expecting, and whilst they might have allocated it incorrectly, if you go back far enough all of our cousins are related to us on both sides of our tree, so it would be strange if none of them showed up amongst our matches.

 

[Helen7]

if you go back far enough all of our cousins are related to us on both sides of our tree,

Well, yes, I can see that, but the chances of being able to go that far back are very small, which is why I say I'm ignoring it. I have more recent leads to follow - and it would help if more Ancestry matches would actually reply to my messages!

 

[peter]

Well, yes, I can see that, but the chances of being able to go that far back are very small, which is why I say I'm ignoring it. I have more recent leads to follow - and it would help if more Ancestry matches would actually reply to my messages!

If you mean ignore it until it becomes relevant, then that's what I do with almost all of my matches - however Ancestry assigns them. But if there is a shared match, or the other person has direct ancestors whose surname is a close match for one of my ancestral surnames, then it's useful to know what Ancestry thinks.

 

[Andrew Lloyd]

Slightly off topic, but may well be related to the Aug 23 update:

Where have all the Unassigned matches gone??

Over the past 3-4 weeks, all my new matches are coming in as 'Paternal side' or 'Maternal side'. In that period I have had just one 'Unassigned' new match. This is out of proportion with my existing matches where the Unassigned category makes up almost 50% of all matches.

Not just me - the same is happening on all other DNA results that I have visibility.

So am I to look forward to most of my 11,000 existing unassigned matches coming off the fence in the near future??

 

[peter]

So am I to look forward to most of my 11,000 existing unassigned matches coming off the fence in the near future??

I would think so, I have seen occasional on-screen messages about there being an update in progress. Phasing requires a lot of processing, so I'm not surprised they're focusing on recent matches - which are more visible - than past matches.

I'll be interested to see what happens to one of my matches which is currently unassigned. If I look at shared matches there are multiple cousins from each side of my tree, although the only connection I can divine is on my father's side. The shared matches on my mother's side are 2nd cousins, a brother and sister - but since we only share one-quarter of our tree (their mother was my mother's 1st cousin) the odds are that the match is in the part of their tree that they don't share with me.

That's not the only interesting thing about this match - as I am a collaborator on my 2nd cousin's test I can see that Ancestry have made it a both sides match for her.....

 

[Stuart]

I think this is something different, based on what I am seeing. In the past, all new matches were unassigned at first, and only assigned (if possible) at the next SideView update (more than six months apart, recently). The text Ancestry provide to explain "unassigned" still says as much:

If a match appears in this category, it could be because their test was processed after our last update or because we don't have enough information to assign them to one parent or the other.

You may see updates in the future, in which case some unassigned matches may be assigned to a side.

Obviously that has changed. Since my last update, a month ago, all three numbers have gone up - by 33 and 37 for parents 1 and 2, and 29 for unassigned. Going down the match list by newness, the totals for the last month match that exactly. But they do not appear evenly spread in the lists, which is highly suspicious! (I have not counted parents 1 and 2 separately, but lumped then together below.)

For Last 7 days, I have 23 assigned
For 1 week, I have 1 unassigned at the top, then 33 assigned
For 2 weeks, I have 14 assigned
For 3 weeks, I have 28 unassigned.

So I think that:
1. They are now assigning new matches using the labelling by side from the last full update (I've said in that past I didn't see why that was not being done)
2. Positions in the new-old listing are not being determined just by when the match was processed, but in batches based on their Sideview results. I have no idea why - I'd guess thus is new processing, and they have not got it quite right yet.
3. The explanation of "unassigned" hasn't been updated.

 

[peter]

All of my matches in the last 3 weeks have been assigned to one parent or another, with a single exception which is unassigned. All matches older than 3 weeks are unassigned until about 2 1/2 months ago - which presumably was the cut-off for the last update.

 

[Stuart]

All of my matches in the last 3 weeks have been assigned to one parent or another, with a single exception which is unassigned. All matches older than 3 weeks are unassigned until about 2 1/2 months ago - which presumably was the cut-off for the last update.

The first part of that is true of what I described above. The second part is also of what I see. But that interpretation depends on what you think determines the position of a match in the list. I'd always thought it was the date of match processing, noting that if that's not done every day for every user the spread of matches by date will be a bit uneven.

But if that is true, it means that from a date about two months ago until a date about three weeks ago I got a lot of new matches, of which not one was assigned. I do not believe that. The ordering must be based, in part at least, on something else.

 

[peter]

But if that is true, it means that from a date about two months ago until a date about three weeks ago I got a lot of new matches, of which not one was assigned. I do not believe that. The ordering must be based, in part at least, on something else.

I don't see why. We know that Ancestry have changed the way they assign matches, and it seems pretty clear that they are now processing new matches as they happen. We also know from Ancestry's now out of date explanation (which you quoted above) that previously matches would have to wait for the next update.

What we don't know is when Ancestry are going to process the unassigned matches since the last update and reprocess the older unassigned matches using their new algorithm. We also don't know how reliable their new algorithm is. It may be there is a White Paper on their website that we haven't discovered yet.

 

[Stuart]

I've repeated the counting process I did a week ago, (but automated much it). Here's last week's breakdown by recency for comparison:
For Last 7 days, I have 23 assigned
For 1 week, I have 1 unassigned at the top, then 33 assigned
For 2 weeks, I have 14 assigned
For 3 weeks, I have 28 unassigned.
The total of those matches the change in the overall counts since a month ago.

This time:
For Last 7 days, I have 32 assigned - that matches the change in the overall counts since a week ago
For 1 week, I have 1 unassigned at the bottom, then 23 assigned
For 2 weeks, I have 39 assigned
For 3 weeks, I have 8 assigned.
For 1 month I have only unassigned (108)

So that's still consistent with matches being listed by time of match processing, which now includes SideView assignment, but with an implausible complete lack of unassigned new matches.

My last unassigned match has just moved down the list into the "two weeks ago" list, at around midnight. There's a few tipping over other boundaries down the list too - as you'd expect if the age is worked out from a timestamp. That I saw no changes since this morning until midnight looks like the age is taken from the datestamp alone, except my midnight is in BST, and it seems very unlikely Ancestry use that.

There must be an anomaly in these lists due to the switch from dividing into weeks to by months. Probably the fourth week back ("Three weeks ago") is nine or ten days long (or, in March, seven or eight).

 

[peter]

an implausible complete lack of unassigned new matches

I don't think it's implausible - we know that Ancestry use a phasing algorithm, so in theory they should be able to assign every match. The fact that they haven't done so in the past suggests a cautious approach; that they are doing it now suggests that they are more confident than previously.

After all they're not reliant only on the phasing algorithm - they can make use of the relationships that members have manually entered, and the Common Ancestors hints. Whilst Common Ancestors might not always be 100% correct, I suspect that it rarely assigns a match to the wrong side of the tree. They may also make use of Shared Matches, although this a less reliable guide.

Personally I don't need 100% accuracy - I only look at distant cousins where there is already a hint (a surname, a birthplace, a shared match) as to how we might be connected. If Ancestry assign the match to the other half of my tree it's unlikely to be worth investigating further, but if they show it as unassigned that doesn't help me at all.