Being able to accurately and reliably link evidence from different sources to the same individual is a key skill which all family historians need to learn and master.
In part 1 we looked the five factors which underpin evidence linkage:
- Community Size
- Time Difference
- Contradictory Evidence
In this concluding part we look at how to assess linkage strength and subject it to rigorous tests.
Assessing Linkage Strength
Once you have assigned High/Medium/Low assessments to the five factors, it’s time to look at all the scores and make an overall judgement about the likelihood that these records relate to the same person. I recommend a four-point scale:
In the case of Daniel Usherwood, my table of linkage assessment scores looks like this:
Based on the favourably high uniqueness of the data, the low community size, the low geographic distance and the small time difference, this looks highly probable. However, the lack of direct evidence for the death of the wife from the 1750 marriage is troubling, so I downgraded this to Probable.
Testing a Linkage
In the software industry, where I have spent most of my career, we use the adage “Nothing works until it’s tested”. What we mean is that even if you think something is working as intended, when you subject it to testing you will always find faults; only after fixing the faults will it work as intended. The same is true of any assessment or assertion we make in genealogy.
My statement “It’s probable that the two marriage records in 1750 and 1754 relate to the same man called Daniel Usherwood”, is based on a thoughtful assessment of the five factors governing linkage confidence. The next step is to subject it to rigorous tests. They may find faults, which may or may not be resolvable. The absence of faults, or the finding of minor faults that are resolvable, will leave us with greater confidence that the two records relate to the same person.
There are four types of tests that we can apply:
Positive tests are those which search for additional evidence which might support the assertion. If new evidence is found does it support the assertion or introduce worrying additional factors?
Example: In the case of Daniel Usherwood my biggest concern was the lack of direct evidence for the death of Mary, the bride from the first marriage. I searched for burials of Mary Usherwood and found one in January 1754 at St Lawrence, Frodsham, which helpfully stated “Mary, wife of Daniel Usherwood”. This supported Daniel’s, admittedly swift, remarriage six months later. With the additional evidence I was able to downgrade the contradictory evidence to Low and upgrade my linkage confidence to Highly Probable:
Negative tests are those which search for evidence which might contradict the assertion. If we find contradictory evidence, does it fatally undermine the assertion or is it resolvable? If we are unable to find contradictory evidence, despite trying hard to do so, it enhances our confidence that the assertion is correct.
Example: For Daniel Usherwood, I chose to do the following additional searches around the same time period:
- burials of Daniel Usherwood
- other marriages for Daniel Usherwood
I found no relevant burials but, surprisingly, found two more marriages:
- September 1756: Marriage bond and licence, but no actual marriage ceremony, between Daniel Usherwood of Preston-on-the-Hill in Runcorn parish, yeoman and widower, and Mary Oaks of Budworth parish.
- April 1757: Marriage bond, licence and ceremony between Daniel Usherwood of Preston-on-the-Hill in Runcorn parish, yeoman and widower, and Ann Wild of Antrobus.
Could it really be the case that the same man married (or nearly did so) FOUR times in seven years?
Once again, the uniqueness of the data was High, the community sizes Low, the distances were Low, and time differences were also Low. Contradictory data was again Medium, because of a lack of direct evidence for the swift demise of his second wife, Jane.
So, I searched for a burial of the second wife between July 1754 and September 1756. In June 1756 I found a burial at Daresbury, Cheshire for “Jane, wife of Daniel Usherwood of Preston”. So he really had been unlucky enough to lose a second wife so shortly after the first.
But what about the third “near-marriage” to Mary Oaks? Had this happened or not? I searched for burials for both Mary Oaks and Mary Usherwood but found none. I searched baptisms of children to Daniel & Mary Usherwood but found none. I searched for alternate marriages for Mary Oaks, again finding none. Ultimately, I concluded that, for reasons now lost to history, they simply broke off the engagement before marrying.
All other elements of data lined up nicely, so I was once again able to reduce the contradictory data to Low and set the linkage confidence to Highly Probable. Furthermore, my chain of evidence had now expanded from two marriage records to three marriages, two burials and an aborted engagement.
Sometimes the potential ways in which evidence can be interpreted leave the researcher with several possibilities. In scenario testing we identify the different options and look for evidence in support of each of them. If we find evidence for only one of the scenarios and an absence of evidence for the alternatives, then we can say with confidence which of the scenarios is correct.
Example: For Daniel Usherwood I considered that there were a few possibilities:
- There were two or more men called Daniel Usherwood of marriageable age in the 1750s.
- There was just one man called Daniel Usherwood who was very unlucky with the survival of his young wives.
- I had found a serial killer in the family! (Okay, this one is a joke.)
To establish if scenario 1 could be true, I searched for baptisms of anyone called Daniel Usherwood (and suitable name variants) in the years before 1750, finding just:
- Daniel Usherwood, baptised March 1699 at St Lawrence, Frodsham
- Daniel Usherwood, baptised November 1730 at St Lawrence, Frodsham
I investigated the 1699 baptism in further detail. This turned out to be a first cousin once removed who, according to a burial in 1700, didn’t survive childhood. The 1730 baptism was for our Daniel, adding another link to our chain of evidence. I concluded that the evidence only supported scenario 2.
The White Queen Test
Suggested by American genealogist Robert Charles Anderson, this test is named after the White Queen from Alice Through the Looking Glass who would believe six impossible things before breakfast. (I recommend Mr Anderson’s excellent book Elements of Genealogical Analysis, New England Historic Genealogical Society, 2014.) The White Queen test literally turns your assertion upside down, asking you to imagine what you would have to believe for the assertion to be untrue. The more ridiculous and unlikely the scenario required to believe it is untrue, the more likely the original assertion is correct.
Example: For my Daniel Usherwood assertion to be untrue I would have to believe that there were two (or more!) men of the same uncommon name, living in the same geographic area, at the same time, with the same occupation, but that baptism and burial records had survived for only one of them despite the excellent survival and online representation of other nearby Cheshire parishes.
Which is more likely: the White Queen scenario or that one man lost two wives to childbirth or illness and married three times in seven years?
Given what I know of the 18th century medicine and mortality rates, and from my knowledge of the survival of parish records in north Cheshire during the mid-18th century, I am satisfied that the only reasonable scenario is that all the marriages relate to just one man, my 6x great-grandfather Daniel Usherwood, who lost his first two wives after tragically short marriages.
Linkage of two pieces of evidence to a single, unique individual is a fundamental skill for all family historians. You should always assess uniqueness, community size, geographic distance, time difference and contradictory information when deciding the confidence level to apply to a linkage. All linkages should be subjected to rigorous testing to be sure that you’ve considered all the options and eliminated the incorrect ones – use positive, negative, scenario and White Queen tests for this. Remember, a linkage is only safe if it has been tested!
How many of your conclusions depend on linkages between two of more pieces of evidence? How many of them have been carefully evaluated and thoroughly tested?
Leave a comment