By Graeme Archer
What are we learning from the row over the statistical methods used by rock-star Leftish economist Thomas Piketty in his new "Tax everyone, everywhere" book?
Well: nothing we've not learned before.
The first and important point to make is that the FT has done us a service,
by subjecting the book to the sort of statistical quality review that
its hagiographers, publishers and author appear to have overlooked. I
say "appear" because I've not read his book, and never would. My life is
too short (yours may not be) to read books by economists which are
aimed at political audiences; the primary function of such objects
appears to be the provision of topic matter for columnists who don't
like novels. The book is important not for its content, as such, but for
the fuel it provided for those columnists whose opinions never stray
far from the "Tax everyone, everywhere" theme to begin with. (Just my
opinion. I'm not against such books, just the idea that I ought to read
them.)
We've been here before. Remember The Spirit Level?
Another attempt by Leftists to use statistical methods (ones which
would shame an undergraduate) into proving that inequality causes, inter alia, suicide, depression, general bad things. Rainy bank holidays? Not quite. But getting there.
That work was taken apart by the amazing Christopher Snowdon, who wrote The Spirit Level Delusion: Fact-checking the Left's new theory of everything.
Nothing new under the sun. Remember the Hockey Stick? Climategate? Even I wrote about them, back in 2009
and, unlike with Piketty's book, I did read all the source material in
the Hockey Stick case. Those authors did have a statistical case to
answer, in my opinion: but their political objective was secured, while
the statisticians argued away in the background.
Thomas Piketty, 'rock-star economist' (Photo: Reuters)
This phenomenon – of using data analytic methods to "prove" a
contentious political issue one way or another – is more often found on
the Left. But let's not forget badger culling, where "the science", if
anything, was found to detract from the "Culling badgers will alleviate
bovine TB" hypothesis. I've written about that (endlessly), here and here, yet the cull continues, under a Right-wing administration. Sometimes the politics trumps the science.
And that is entirely the way it should be. Bayesian
statisticians can explain what's happening. The commonplace view of
scientific method is that hypotheses ("Inequality will always rise",
"Inequality causes depression", "Temperatures have just shot up",
"Culling badgers cures bovine TB") can only be rejected, or falsified.
Experimental data is used in an attempt to "disprove" the null version
of these hypotheses (that is, their converses: "Inequality isn't
rising", and so on), in a fairly antiquated and hugely flawed
statistical process known as "significance testing". If a significance
test is positive, people not trained in statistical reasoning –
newspaper columnists, Leftists with books to sell – have a tendency to
start claiming that "the science has been proven."
Über-litigant, Michael Mann, saw his fraudulent 'Hockey Stick' crumble along with his bogus claim of being a 'Nobel Laureate.' See his incredible shrinking résumé and accolades on his Penn State bio. Before. After.
It hasn't (even those statisticians who continue to hawk significance
testing as a valid approach to induction wouldn't make that claim), and
in any case, this isn't how our reasoning about the universe works. We
come to any particular policy topic with a set of preconceived opinions.
New data will modulate those opinions, in a process which can be
explained by Bayes theorem. If new data is hugely strong, in terms of
the signal it contains about a hypothesis, then people whose opinions
are very different a priori will, a posteriori (literally "after the data are seen"), converge in belief. If the signal in the data isn't strong, they won't.
That doesn't make either side of a political argument irrational,
merely that data of insufficient strength has yet to be generated to
overcome the (perfectly) valid world views of different human beings.
You can intuit what I mean through a thought experiment. As a Tory, I
have hugely strong belief in the importance of the integrated family
unit with regard to positive outcomes for children. It would take more
than a single study in a sociology paper which "proved" that there are
no "significant" detrimental outcomes to children raised in
"alternative" set-ups, to cause me to discard my theory. But if you are a
liberal, with a strong belief in the alternative ("Family set-up
doesn't matter"), then a single study may well have quite a strong
impact on your belief. You would end up on Radio 4, claiming "the
science has been proven".
The political mess is caused by two separate things, neither easy to
fix. First, our culture has decided that science is "objective", that
the probability of different theories being true is somehow utterly
independent of the views of the men and women who decide which theories
to investigate, or the manner by which those investigations proceed. I
reject (ironically!) this idea that science is objectively objective, which makes me a Bayesian. We remain in the minority but our views are gaining ground.
Secondly, and less philosophically, we have a problem that there
simply aren't enough journalists and media commentators properly trained
in the quantitative discipline of statistics in order to subject claims
such as Piketty's to rigorous analysis. As a statistician who also
plies his trade as a political writer, I would say that, wouldn't I?
Maybe I should read fewer novels, and pick up more economic textbooks,
after all.
Don't Miss:
No comments:
Post a Comment