Ancestry's New ThruLines Needs Us To Check Its Results by Alice Kalush 3/25/2019 Ancestry's new ThruLines DNA functionality is really cool and is likely to help a lot of people break their brick walls. But while all of the people it for which it identifies ThruLines do share DNA with you, it may not always be correct in identifying the common ancestor you share with your match. How to Use ThruLines: There are two things we need to be doing with our DNA results from ThruLines. The first thing is that when you find a new ThruLines ancestor, check your tree and your match's tree and decide if you believe before you add the new ancestor to your tree. The second thing is that we need to get the word out that ThruLines connections aren't always accurate and encourage others to check before adding new people to their trees. Wrong John Gable! I'm descended from Wilhelm Gebel's son John Gable b. 1761 and Elizabeth Benz. Ancestry has substituted my John Gable's brother John, who was born in 1774 for my John Gable in ThruLines and in Hints on the Beta version. It is also showing John b. 1774 as being the husband of Elizabeth Benz (wrong!). This is probably incorrect for all descendants of Jacob Gable and Elizabeth Hambright, Catherine Gable and Christian Prowant, Benjamin Gable and Anna Koppes and Henry Gable and Lydia Baer on your accounts too. There are two examples later in this document which have the wrong John as the common ancestor (the Shaum example and the Prowant example). 1
Here's what ThruLines has found so far about my Gable matches. In addition to the ones shown below, I do have DNA matches to descendants of Wilhelm Gebel's children Valentine, John b. 1774, Esther and Barbara that ThruLines didn't find. The branches that ThruLines did find were the ones with that actually have Wilhelm Gebel in the tree of at least one person from their branch. I have about 160 Gable matches in my DNA Comparison spreadsheet (pre ThruLines) versus 50 total ThruLines found. Even with my in depth Gable DNA work I found a bunch of new DNA matches that I believe are valid from multiple different branches of the Gable family. Most of them were from the Prowant branch (descendants of Wilhelm/John b. 1761/Catherine Gable and Christian Prowant) but a few were from other branches (and not all descendants of my John either). What ThruLines is Actually Doing As a software developer I worked on a project matching up medical data between different databases where the information being matched wasn't entered in the same way. This resulted in Congress changing the way that funds were allocated for treating occupational illnesses and injuries. We had the same problem that Ancestry is having matching up our trees with other people's trees in order to give us Hints and ThruLines - people not entering their data the same and not all having the same information. I used a two stage method to find matches. 2
The current Ancestry DNA Hints, which I'm calling a stage one algorithm, requires that the surname be spelled the same and that both you and your match have the same common ancestor in your tree in order to show a hint. Using this methodology hints are about 95% correct (possibly lower for common surnames). This is accurate enough that most of what we find can be trusted. In my medical research I added a second stage to locate more potential matches and then used humans to view them and decide if they matched or not. A human can use a wider knowledge base to determine if two records match. Programming a computer to do what a human can do at the same level is difficult. I found this in my health data matching project and I believe that Ancestry's task is much more difficult. Ancestry ThruLines (and Beta version Hints)'s new algorithm is operating in a similar way to how my second stage algorithm works. It is much more liberal about what it considers to be a match than the previous version. It identifies more potential matches than the current version does, but some of them aren't correct. This is why we need to check what ThruLines is finding. I'll be giving some examples of what ThruLines is doing and how it is doing it later in this paper. When I compare what I found with the current Ancestry version with ThruLines, I'm seeing that Ancestry is actually doing a better job of finding new accurate ThruLines hints for beginning Ancestry users with smaller trees than for more advanced users with larger, well researched trees. In fact, they weren't doing well at all for finding believable new ancestors for me, but were doing a good job of locating new DNA matches with people that share a common ancestor I already knew about. ThruLines combines information from your tree, your match's tree and even other people's trees to connect you. In the process of checking out my new ThruLines matches I looked at why Ancestry connected me with these new matches. In the Prowant family, which shares my ancestors John Gable (Wilhelm's son) and Elizabeth Benz, I found several examples of connections that I believe are probably accurate (but I'd want to check more if these were my ancestors). These connections were largely based on my large tree and that I had made an attempt to enter information in my tree about the Prowant family. But most of these people's trees earliest Gable descended ancestor was a great grandchild of Wilhelm Gebel. Ancestry connected us in ThruLines using my tree which contained this great grandchild. Based on that I tried to pick well researched trees to use for the information in my tree, this may be reliable, but I don't know what these other people did to check their research. This is what ThruLines can do for us - give us good leads! 3
But this approach doesn't always work Here is an example where four different people's trees are used to make the connection between my DNA results and Esther Maze's DNA results in our trees. You can tell that four trees were used because in the boxes of the chart below it starts with my tree for Wilhelm Gebel and his son Daniel, then it says From CKlauer's tree (for Margaret Gable), From thubbard1967's tree (for Daniel Todd) and From patriciapearce2001's tree (for Esther Mae Todd Maze). In the screenshots on the next few pages I clicked on these boxes and Ancestry shows the profile of each person in the chart in the tree that was used for that person. 4
The connection between my tree and Chris Klauer's tree looks good. Chris Klauer's tree (CKlauer) has the top portion of the list through Margaret Gable. From CKlauer tree: Daniel Jefferson Todd 1847 1910 BIRTH 7 JAN 1847 Ohioville, Beaver, Pennsylvania, USA DEATH 18 FEB 1910 Athens Township, Ringgold, Iowa, USA Lucinda Miller 1851 1938 BIRTH 5 FEB 1851 Warsaw, Wells, Indiana, USA DEATH 30 MAR 1938 Iowa, USA This was connected to thubbard1967's tree which doesn't have parents for Daniel Todd and Lucinda Miller and doesn't have an Esther Mae Todd born in 1914 listed in their children. Connecting the Daniel in CKlauer's tree to the Daniel in thubbard's tree is believeable to me, but I'd want to find other records supporting it. 5
The final step connected to PatriciaPearce's tree which doesn't have parents for Esther. Frankly, I don't see a reason for this connection at all because there are no parents in Patricia's tree and Esther isn't listed in thubbard's tree. And the birth years of Daniel and Lucinda's other children makes it doubtful that Esther Mae is part of their family. Note that Esther Maze's maiden name while really Todd was listed as Maze in Patricia's tree and probably shouldn't have been connected by Ancestry for this reason (problem: equating similar text that isn't all in the surname field.) Disclaimer: I have been in contact with Chris Klauer but not the other two people whose trees are used in this example. Other matching issues: 1) ThruLines matches up people with their surnames spelled differently. This helps find the Gable/Gabel/Gebel variations but also picks up other unlikely variations. While this can find more potential matches with the different spellings, I already know from the Include Similar Surnames check box in Ancestry's Search Matches feature that this results in a significant number of additional false matches using family tree data. 6
2) ThruLines matches up people with no dates or places listed for birth, marriage, death. An example of this is shown below, were the connection between our two trees was made with Florence Shaum and Ralph Holmes. Note also the variation in spelling of Holmes versus Homes. On the next page, my tree is at the top, my matches tree on the bottom. This example may be an accurate match (27 cm, 2 segments) but the needs more work to confirm or deny. 7
8
3) ThruLines linked Susanna Stoner in my tree (no parents listed) to "father" Christian Stoner in someone else's tree which didn't have a Susanna listed as one of Christian's children. The plus side is that Manor township and Mountville are only a few miles apart. But is this the right Stoner parent or the right Christian Stoner? 9
Here is my match's tree. Note that Susanna Stoner isn't listed in it. 10
Here is my tree for Susanna Stoner. Note that she has no parents in my tree. 11
4) Here's one that's actually probably a good ThruLines tree connection. Notice that two generations separate Nelson Daniel Prowant from our common ancestor Johann "John" Gebel and that the information Kelly Rittenhouse and I have for Nelson Daniel Prowant (on the next page) matches up but I don't have a wife or children. Some of the other Prowant descendants found by ThruLines have a gap of three or four generations and very small trees (making their Prowant/Gable connection more doubtful). 12
Kelly's tree: My tree: 13