Doubt Versus a Bayesian Outlook

There’s lots of primary causes of the recent assault on truth that are non-technological. In fact, most causes have very little to do with technology. I’d point people to the excellent book The Merchants of Doubt which details the well-funded and and well-planned corporate assault on science that began as early as the 1950s around the issue of whether cigarettes cause cancer. There was a simple but profound realization Big Tobacco had 50 years ago — they didn’t have to refute the conclusion of the science that clearly, even back then, pointed to tobacco as a primary cause of lung cancer. They just had to introduce doubt.

The neat thing about doubt is it makes you look and feel like a pretty deep thinker. America loves doubt. Every four years we run an election for 18 months and then treat the people who haven’t decided until the last week of the election as if they were some sort of free-thinkers rather than the most politically ignorant population in the country. The mythology of doubt is strong.

Reporter:  “So what do you think about the election, Bob?”

Independent: “Well, I’m not sure. Clinton has some good points, but Trump seems like a strong leader. I like to take my time thinking about these things.”

Reporter: “Well, it’s quite the important decision. Back to you, Maria!”

The mythology of doubt is that we have things which need to be “proven”, and until they get proven we we are in a state of doubt: we really don’t know what to believe. Who can say?

But doubt is not actually what you want. Doubt is just certainty from another direction, and these two orientations — doubt and certainty — form a binary worldview that promotes polarization, narrow thinking, and poor policy outcomes.

What you really want is not doubt. What you want, for lack of a better word, to be Bayesian in your outlook. The famous statistician and epidemiologist Jerome Cornfield, responsible for much of the revival of Bayesian approaches in epidemiology in the 1960s and beyond, used to talk about the “Bayesian Outlook”.

The Bayesian Outlook is at its heart simple, but it’s also profound. Here’s Cornfield:

The Bayesian outlook can be summarized in a single sentence: any inferential or decision process that does not follow from some likelihood function and some set of priors has objectively verifiable deficiencies. The application of this outlook is a largely extra-mathematical task, requiring the selection of likelihoods and priors that are appropriate to given problem situations, with the determination of what is appropriate requiring, in Fisher’s words (in another context), ‘responsible and independent thinkers applying their minds and imaginations to the detailed interpretation of verifiable observations. (Cornfield, 1969)

There’s a field of Bayesian statistics that is fairly developed and beyond the scope of this post. But as Cornfield notes, Bayesian approaches are not really about the math — they are about a way of looking at the world. And given that I think it’s possible to talk about having a “Bayesian outlook” when it comes to fact-checking.

What does this mean in practice? As an example, I use this tweet occasionally in my presentations:

Is the part about the Nazis true? It’s either true or not, of course. But we can only view that truth through an array of probability.

When I first see something like this, my immediate reaction is it has a good chance of being true. Why?

Well, there are priors. I know Schumer is Jewish, of European descent. And I know that the Nazis and their collaborators killed a substantial portion of of that population, maybe about 40%. I also know you have, by definition, eight great-grandparents. The chances that at least one of the eight great-grandparents might have died in WWII at the hands of Nazis or Nazi collaborators is something that had a reasonable chance of being true before this tweet.

We call these the priors: they exist before this tweet makes its way to me. One key component of Bayesian analysis is that we begin with a set of priors, and pay careful attention to the selection of those priors before assimilating new information.

Now as to the new information: the fact that someone tweeted this claim makes the claim more probable, to some extent. This is a specific claim. It came to me through a feed where I weed out the worst misinformation offenders pretty regularly. The second statement, about Trump’s father, is true.

It seems plausible. But I follow my prime habit with social media: check your emotions. Never reflexively tweet something that factual that feels “perfect”.

A quick search shows there’s a 1998 article from the New York Times that says that “aides say” seven of nine of his great-grandmother’s children were killed by Nazis. That’s good, and raises the likelihood it’s true. The old priors plus this new information become our new priors. We’ve moved from plausible to probable.

But I want to hear it from Chuck Schumer’s mouth, not some unnamed aides responding to a campaign attack in 1998.

And when you start to try to find Schumer saying it it gets less clear. There is Holocaust after Holocaust event that Schumer has attended — and yet this fact never makes the papers or his speeches:


Absence of evidence is not strong evidence of absence. But it is evidence, especially as it starts to pile up. With each failed attempt to find support for this, my disposition towards this fact inches down, moving from likely and sinking back towards plausible.

Then, at some point, I change my search terms. One of the unreliable sites on this question — a forum post —  mentions a “porch” where his great grandmother was killed. That’s a specific detail that is likely to get me closer to the event. So I throw it in and look what comes up:


And when we go to that top result we find testimony from Schumer at a congressional hearing on reparations for Holocaust survivors:

Senator Schumer. Now I am going to give my opening  statement, and first I want to start by thanking our Chairman,  Chairman Leahy, for letting me have the gavel today in order to  explore this exceptionally important topic: how to resolve what  I hope, what we all hope are among the last remaining  reparation claims stemming from the murder of 6 million Jews during the Holocaust. We all know the horror of the Holocaust.  My great-grandmother, who was the matriarch of her family, was told to leave her home. She and her family had gathered on the front porch. They refused to leave, and they just machine-gunned all of them down in 1941. So, obviously, I have personal experience with the horrors of the Holocaust, but the horrors are just awful.

Sometimes we refer to the horror as “unspeakable.” But unspeakable is exactly what the Holocaust must never become.  Those who perpetrated it, those who benefited from it want us not to speak. But we are here to speak and to have this hearing.

Now that’s a good source — official testimony from Schumer himself. From a written statement. The weight of this evidence outweighs everything prior, but is still added to it. It’s not just that Schumer is telling a story here, but that he is telling a story about an event that was plausible to begin with.

Is it bulletproof? No. Schumer could, of course, be lying, or exaggerating. He might have heard or remembered the story told him wrong. But right now, the best information we have is this testimony plus the remarks of others (such as aides) over a 20 year period. We have enough here, in absence of other evidence, to call this claim true.

But unlike “doubt” or “certainty”– the demand that something anything less perfect knowledge one way or another must leave us in a useless middle ground, we end up, with each step, getting better, more informed priors even as our decisions on what is true vacillate. By the end we call this true, because to overcome what we know here would require strong evidence that currently doesn’t seem to exist. But we’d be excited to get new information, even if it contradicted this, because it would build a better set of priors, both for this and other related claims.

This post is pretty nascent stuff — and maybe I’ve bit off a bit more than I can chew here. But I suppose what I’m saying is that fact-checking on a complex claim looks a bit like this:


We’ll get this together in a better presented post at some later time. But I do think one of the primary goals of fact-checking is to get students to think about truth in more nuanced ways, and this is the sort of direction I see that going, instead of the cynical skepticism we often peddle.




3 thoughts on “Doubt Versus a Bayesian Outlook

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s