Sunday, March 03, 2013

Not all priors are equally defensible.

There appears to be a lot of Bayesian thumb sucking going on, maybe the first was Eli's duo with Socrates, and, of course the bunnies know that James has been going on about uniform priors, and there is always Andrew Gelman.  Now some, not Eli to be sure, might think that the recent election also gave a strong push to priors and p values and such.  Nate Silver of the NYTimes blog five thirty eight has a book out which is reviewed in Science by Sam Wang and Ben Campbell, who also are in the election prognostication business.  Silver, of course, is another guy with a Bayesian hammer looking for statistical nails and finding them all about.  Eli thought a  couple of paragraphs towards the end capture what the Rabett has been trying to beat into bunnies heads.

Our biggest criticism of the book is that although statistics and Bayesian inference are powerful ideas, they are not a cure all.  In his enthusiasm for the good Reverend, Silver has stuffed a fair bit into the same Procrustean bed.  Silver uses the old fox-hedgehog analogy, saying that foxes (including himself) use many ideas, whereas hedgehogs focus on one subject only.  But here he is a hedgehog with one big idea:statistics.
However, Bayesian reasoning works only if the prior is adapted for the task.  According to Silver, many of today's "half baked policy ideas" could be rectified by Bayesian thinking, but that is only part of the story.  The more difficult task is determining good priors.  Silver rejects bad priors effectively in his own field of electoral forecasting by dismissing much of the noise of political punditry.  In other fields he does not always bring the same critical attitude.

Scientific research is often confronted by political and economic forces that are not always appreciated by nontechnical outsiders.  Silver, somewhat perversely takes climate scientists to task for bringing politics into their work.  If anything climate scientists have been dragged unwillingly into a dispute with political interest groups such as the Heartland Institute.  At this point in history, human-induced global warming is a fact an no longer a matter of disputing probabilities.  The book's extended treatment of scientific fringe figures has the inadvertent effect of giving credence to antiscientific views that fly in the face of experimentation and hypothesis testing on the greenhouse effect dating back to Arrhenius over a century ago.  When Silver, now himself a prominent pundit, depicts a "controversy" he highlights the challenge scientists face in convincing people that carbon dioxide is a pollutant.  Not all priors are equally defensible.

15 comments:

Anonymous said...

The book's extended treatment of scientific fringe figures has the inadvertent effect of giving credence to antiscientific views..."

There is nothing "inadvertent" about it.

Silver might be a lot of things, but he ain't dumb.

Just like Steve Levitt, Steven Mosher and Tom Fuller, Silver understands that "controversy" sells books.


~@:>

jyyh said...

Sorry for off topic, but what do you have against our brothers and sisters??
http://www.newscientist.com/article/dn23228-bunnies-implicated-in-the-demise-of-neanderthals.html

Steve Bloom said...

That was just a warning, jyyh. The bunnies play hardball.

Anonymous said...

There is irony in that a few years ago I noticed a spate of denialist attacks on the use of Bayesian methods and specifically on the use of any priors.

Now they are all for Bayesian Analysis a la James Annan in the mistaken belief it will give them the answer they want.

Man with no Name

Anonymous said...

Hi
As a bunny who was introduced into the topic of climate change in the early 90's by "HotHouse Earth: The Greenhouse Effect and Gaia" and was SCARED, based on what I had read (I get that J.G. is somewhat discredited.) But as a layperson, I have been scared since the EARLY 90's. Now? After being ridiculed by my friends for years (because I am a layperson who cannot articulate the danger properly) I feel pretty much conditioned into helplessness. I want to say, "We are locked in, people, to at least 2 degrees!" And nobody seems to care, aside from y'all, who have spent less time (IMHO, but I understand) freaking out. Yes, get arrested, yes, start speaking out, yes, tell it like it is. BUT, really, I feel like the debate is over, the time for freaking out is done. And the debate over what the effects are is, essentially, useless. We are headed into disaster no matter what, right now. Two degrees? Are you kidding me? Everyone on this forum knows what 0.8 has done, and it is going to get worse. Much worse. Just wait for 1.5, let alone 2!
Yes, I saw the recent commentary on Tamino, and Eli, and RC on this topic. Yes, I read on the Guardian about the guy who is doing a show in GB who talked to a climate change researcher who is teaching his kid to handle a gun. Doesn't help, because I feel that despite your passion and mine, we are FD. We are past that. We were probably past that 5 years ago, based on the "lag" in the system, etc. Please talk me out of being wrong.

a_ray_in_dilbert_space said...

"If the prior distribution, at which I am frankly guessing, has little or no effect on the result, then why bother; and if it has a large effect, then since I do not know what I am doing how would I dare act on the conclusions drawn?"--Richard Hamming

When using a minimally informative prior, isn't Bayesian methodology really just a trick for turning likelihood into a probability?

a_ray_in_dilbert_space said...

Anon@11:45, I first found out about climate change back in the late '80s when I was a grad student. At that point, the consequences seemed remote. I learned more about it in the '90s, when the problem seemed more imminent, but there was still plenty of time to avert the worst consequences. Now the question is whether human civilization will survive in any form we would recognize. If we keep fighting, it just might.

J Bowers said...

"Eli's duo with Socrates"

Try Homer instead.

willard said...

Nice find, J Bowers!

Socrates' character was Eli's idea.

Twas an ironic one, considering that probability had to wait a bit to emerge.

Only Socrates' can match Eli's irony.

Anonymous said...

jyyh.

I have one word by way of reply: Caerbannog.

Personally, I'm a little dubious about the rabbit idea. Game of various types would not have been all that scarce in the Neanderthal period, and bunnies aren't a huge constituent of the proteinaceous biomass: they'd be less likely to be so in a more forested past.

Perhaps it was death by mal de caribou - maybe (contrary to the article) Neanderthals were actually too good at catching bunnies! Just because there's no midden evidence don't mean it ain't so!


Bernard J.

David B. Benson said...

a_ray_in_dilbert_space --- minimally informative prior: definition?

Bernard J. said...

(Being the first part...)

3/3/13 11:45 PM Anonymous.

My first exposure to global warming was as a school boy in the late 1970s. In 1989 (+/- a year or so) I heard David Suzuki speak, and warn that the 1990s were the last decade for effective preventative action. As a fresh-faced graduate myself I was scared then, just as you were.

I started commenting on blogs about 7 years ago, when desperation drove me to decloak from my usual social reticence and add my bit to counter the fast-growing movement of internet denialism, which was one of the greatest millstones to action at the time. By then I was extremely concerned, but noises from Australian non-conservative politicians lead me to believe that we would manage an eleventh-hour start to blunt the worst of the impacts not already laid-in.

About three years ago I was even more extremely concerned, because it was patently evident that organised denialist interests had the upper hand in the message that was reaching the lay public, and hence feeding back to politicians. By this time I was of the opinion that it was too late to stop serious damage to the biosphere after 2020-2025, and that humans would not likely engage in any serious reduction of emissions before then.

These days, with the current willful ignorance of the Australian Coalition and of the majority of the public that seems determined to vote them in, I am certain that humans as a species do not have the capacities of understanding and of will to act until after we've committed the planet to profoundly serious warming.

Bernard J. said...

(Being the second part...)

As an example, there was a report on our local ABC radio today in which an agricultural strategist was speaking of ignoring whether humans were causing the warming, and persuading farmers that they needed to adapt to the coming warming whatever the cause. There was also the comment made that adaptation research would require decades, so farmers needed to support it now.

There are two profound flaws with this 'enlightened' thinking. First, if the human cause of warming is not explicitly acknowledged and confronted then that cause cannot be addressed and thereby stopped, and so the warming will simply continue whilst humans are attempting to adapt to the early stages. It's like trying to put out bushfires whilst giving the arsonists free reign to blithely drive down the road to the next forest, and the next, and the next... One may as well say "It doesn't matter what's causing the fires, we've got to learn how to improve our fire-fighting strategies"...

And secondly, whilst the commenter was correct about the time-scale required in order to adapt to the currently committed-to (and realised) amount of warming, he fails to understand that much research into adapting to warming of between 0.5-2.0° C over baseline will be made redundant if and when temperature anomalies continue to increase and reach 2.5° C and more over baseline. It's a futile exercise to learn to adapt to a temperature regime that is only going to exist for a matter of a few decades, especially if much of that research is effectively realised only after those decades are over.

And people who think that human society in its current guise can adapt to temperature anomalies over 3° C have rocks in their heads. And yet this seems to be the implicit policy of much of the Western world's government - it's certainly the attitude displayed by the current Australian conservative Coalition.

I now despair for the security and integrity of the planet that I will leave to my children. However, having said all of this, I will also note that any action we take now, as soon as we can, will make less awful the future for billions of humans and for countless other species. We're now into "do you want dozen of kicks in the nuts, or hundreds of kicks in the nuts" territory. Every year, every month, every week, and every day that we delay adds another few kicks in the nuts - and the more we procrastinate, the more kicks will be delivered.

It might already be too late to avoid serious impacts, but it's not too late to avoid the worst of them.

a_ray_in_dilbert_space said...

David Benson
A minimally informative prior is one that has negligible effect on the Posterior probability. Maximum entropy would be another way to say it. The less information in the prior, the more the posterior looks like the likelihood--e.g. the data.

David B. Benson said...

a_ray_in_dilbert_space --- Thank you.