Saturday, October 13, 2012

Can This Possibly Be True?

So, night before last I was listening to an hour-long interview on NPR's Fresh Air with Nate Silver, the polling guru behind the website, which looks at the polls in presidential (and senatorial) elections, weighs their bias, averages the results, and offers his estimation of where the race really stands (right now he gives Mr. Obama about a two-thirds chance of winning, largely because of his strong lead in states like Ohio and Wisconsin and Pennsylvania).*

The whole interview is of interest, if you like this sort of thing (which I do).

The most interesting points to me were, first, his explanation (in part of a discussion of availability of data) of just how expensive and time consuming medieval manuscript production was: he said a single manuscript cost the equivalent of $25,000 in today's money. If true, that explains very well why books were so rare, and precious: why the Merton College library kept all its books in a single huge chest with three great locks, the key to each of which was carried by a different person.  Or why, later on when they got more books (I assume in the first century or so of printing), they chained them to the shelves to keep them going walkabout.  He said that cost fell by about 500%**  with the advent of printing -- still expensive, but no longer prohibitively so.  I'm sure there were more and less expensive Mss (depending on the length of the text and the amount of decoration incorporated into it), but if Silver's figure is anywhere near right, it's illuminating.

Second, there was his explanation of why, despite so much information being available, polls often fail to predict what really will happen.*** The two key factors turn out to be (1) taking into account what you don't know and (2) judging the trustworthiness of new information you come across.

The first is where most polls and predictions fall down. The weather service is careful not to give predictions too far out, since weather systems are too complex to predict with confidence in the long term. However, they're pretty reliable in the short term: if they say it's going to cool off over the next two days, they're probably right.  So here knowing yr limits produces better results.

 In Silver's case, he gives both a "Now Cast" (the odds of either Obama or Romney winning if the race were held today) vs. the predicted election day figure, which he cautions is based on what we know now, and is always subject to the unexpected -- the levy may break, terrorists attack, a smoking gun or compromising footage come to light ("47%"), or someone well and truly put foot in mouth. 

In the case of Tolkien (you knew this was going to get around to Tolkien in the end, didn't you?), we know a lot -- I've recently taken to playing the "What Was Tolkien Doing One Hundred Years Ago Today? game with Wayne & Christina's CHRONOLOGY -- but there are things we don't know. For example, we know relatively little about Tolkien's life at Leeds, the five years that laid the basis of his professional career, and therefore tend to act as if those years were less important than the ones that are better documented.

We also quite often come across conflicting stories -- did THE HOBBIT originate as an oral story he told to his kids during the 1920s, as Tolkien's two eldest sons insisted, or as a written text in the early 1930s, as Tolkien himself repeatedly stated (and the bulk of the documentary evidence supports)? Did C. S. Lewis write THE DARK TOWER in 1938 or 1946?  Are MR. BLISS and FARMER GILES written before THE HOBBIT, or after THE HOBBIT, or both (i.e., drafted before but with final versions after)?  When we get a new piece of evidence, we weigh how probable it is based on what we already know.

For example, with a 1999 account by a Swedish woman who was an au-pair girl in the Tolkien household in 1930-31 saying Tolkien began THE HOBBIT while she was there, the probability of errors in the latter account is high. There was a long gap between the event and its being recorded, the person involved was v. old at the time it was recorded, and her memories being taken down by a reporter rather than written up by herself. And yet what she says fits in remarkably well with the preponderance of other evidence, of which she herself wd have been unaware; this veracity in what we can check lends credence to elements in her account not found elsewhere.

By contrast, the evidence of John and Michael Tolkien contradicts a well-established pattern. Tolkien is often wrong about dates, but in such cases he almost always errs by predating the event -- as when he claimed he'd begun THE LORD OF THE RINGS before THE HOBBIT was published (rather than three months afterwards), or that he'd delivered the OFS lecture in 1938 (rather than 1939) and written LEAF BY NIGGLE in 1938-39 (rather than 1942-43). The pattern is clear, and it lends weight to a later dating being more probable than an earlier dating.

In general, a good reminder about handling evidence and drawing reasonable conclusions from it.

--John R.
current reading: THE UNOFFICIAL HOBBIT HANDBOOK (2012), ORESAMA TEACHER (2007-present)

*this is down from an 85% chance just a few weeks ago, when Mr. Obama was leading in ten out of twelve 'swing' states after a long string of Romney gaffs.

**for a transcript of Silver's exact words, see the last paragraph under the following link:

--of course, a handmade and carefully illuminated Ms. today on vellum or parchment might well run that much as a luxury item -- even high-quality limited-edition machine-produced reproductions run in the hundreds or thousands.

***according to Silver,  pundits are almost always wrong, but this seems to be by their statements being based on advocacy, not probability.

--Silver does note that local weathermen are far less accurate, deliberately overpredicting bad weather as a way of creating a little drama in hopes of attracting viewers.

UPDATE: As two readers have pointed out in Comments (thanks Ardamir; thanks Troels), the women in the example I used was Icelandic, not Swedish. My mistake. Troels also points out that by a factor of 500 is not the same as 500%. I'd thought my meaning was clear enough; apparently not.
   Thanks, as always, for the corrections.


Ardamir said...

The woman was Icelandic, not Swedish.

Troels said...

Right, Ardamir, I was just about to say the same ;-)

And there's quite a difference between a reduction of price of 500 times (price is divided by 500 - new price $50|2012) and by 500% (new price - you get $100,000|2012 for taking the book . . .).

One of the benefits of Bayesian statistics is that it can model the effect of our prior beliefs given new evidence. Thus the interview with Arndís Þorbjarnardóttir is not likely to change positions by very much with respect to what people believe regarding the genesis of Tolkien's Hobbit story (she has a clear interest in somehow amplifying the importance of the time she spent with the Tolkien's by attributing the genesis of the story to this time, just as she may not have known that Tolkien had been telling bits and pieces of the story orally before her arrival).

And just to be clear — I don't think the evidence is sufficient for either proposition to be much more probable than the other, so I'll remain seated on top of the fence for the time being :-)

John D. Rateliff said...

Thanks to both Ardamir and Troels for the correctives. I'll add an Addendum to the main post.

Perhaps a better example would have been the story about Tolkien having gotten behind enemy lines during WWI and running into German calvary, whose helmet inspired the Black Riders. It's a poorly-sourced story, coming from someone who had no known ties to Tolkien; it was recorded many decades after the event, and it's pretty much impossible given what we know about conditions at the front (trench warfare not allowing for calvary charges). The Bayesian principle as described by Silver would lead us to reject this new evidence because it's too much of an outlier.

No need to agree on the 1920s vs 1930s date; just using it as an example of new evidence coming in that adds to, without greatly changing, an already complicated cluster of evidence. Here I adopt Bernal's principle of going with whichever synthesis offers greater plausibility.