Thursday, 8 May 2014

Confidence and genealogy

As mentioned in a previous post the presentation I most valued at the recent Ontario Genealogical Society conference was Determining how much confidence you should have in your  genealogical inferences by Paul Jones.

Early on Paul mentioned that a practical problem faced by genealogists, and everyone else, is scarce resources. Time and funds are limited. It's all well and good to say you should turn over every stone in your genealogical search but in practice you need to choose what to do and what to left undone. How do you decide?

We make decisions day in and day out relying on experience and judgement and we take increasing care as the implications of those decisions are important and soak up time and money. The probabilistic approach using Bayes Theorem proposed and illustrated in Paul's presentation is a formal way to make the allocation decision which complements the Genealogical Proof Standard.

Hopefully Paul will publish or otherwise make available his paper in the OGS syllabus updated to include an illuminating example he gave in the presentation.

In the meantime consider Paul's list of objections & misperceptions regarding the probabilistic approach and his responses.

• My great-grandfather wasn’t 60% Harry. Either he was Harry or he wasn’t. 
A probability is not a result. It’s a measure of the degree of our belief in the result.

• Statistics cannot replace research rigour. 
No one has suggested that we forgo the precepts of the GPS.

• Words will do as well. If something is unlikely, just say so. 
Does unlikely mean “1 in a million”, “1 in a thousand”, “1 in a hundred”, “1 in 10”? It makes a
huge difference in determining a future course of action.

What about the researcher who gives every hypothesis a 99% prior probability of truth? 
You can’t reject a tool because people use it badly. Otherwise we’d close down the Internet.

• No genealogist can say a finding is 68.45% certain. 
True. And no statistician would either. We’re talking ballpark estimates.

• You can’t do experiments in genealogy so the use of statistics is invalid. 
You can’t do experiments in astronomy either. But it’s empirical in that you can make
probabilistic predictions that are or are not subsequently confirmed by observation.

• Nothing in genealogy is certain and adding numbers won’t make it so.
The words of a math-phobe. In this case numbers are being used to describe uncertainty,
not prescribe certainty.

• There is absolutely no objective basis for applying numbers to hypotheses. 
We have reasonable estimates of probability of various events, e.g. naming patterns, false
paternity and so on. Bayes Theorem was proved in the 17th century.

Paul was kind enough to reference my blog posts of July 23-24 & Sept. 4-6, 2012.


No comments: