Times, November 2007
Why Thinking-By-Numbers Is the New Way to Be Smart
Hb, 272, notes, index, ISBN 978-0-553-80540-6, $25
commercial success of Steven Levitt's 'Freakonomics' (reviewed in
FT202), here's another folksy introduction to the exciting world of
heavy-duty statistics. 'Super Crunching' is Ayres' cutesy name for the
statistical analysis of big data sets, with the aim of making better
decisions - and, to hear him tell it, it's not just super and exciting
but 'very cool'.
Ayres, a law professor at Yale with a background in econometrics, looks at examples of such analysis in areas from fine wine investment to evidence-based medicine. The tension here is between the evidence of the data and expert opinion - and, as Ayres repeatedly describes, analysis usually beats the expert hands-down.
The demonstrable supremacy of evidence over expert opinion is surely something that any fortean could applaud. However, such analysis is only as good as the assumptions underpinning it and the rigour with which it's carried out. Often, much like the aphoristic expert, every statistical study has an equal and opposite study. For example, Ayres devotes a couple of enthusiastic pages to studies which show a link between male circumcision and reduced rates of HIV infection, a link first proposed by US urologist and circumcision enthusiast Aaron Fink. However, perhaps due to deadline pressures, he doesn't mention more recent studies which show that the link all but disappears when other cultural and social factors are taken into account - exactly the kind of flaw that Ayres rightly criticises in other studies.
To my mind, Ayres doesn't put quite enough emphasis on the need for caution when dealing the fruits of such studies. Flawed statistics have been used to bolster controversial policy decisions, notably John Lott's discredited work on concealed gun laws, which Ayres gives a pretty balanced account of here.
The private sector is more advanced than any government in the use of these analytical methods, however, with the pure economic motive of extracting the most profit from customers based on exhaustive records of previous behaviour. This inevitably brings up concerns about individual privacy, which are very lightly touched on here.
More generally, there's also the problem that nearly all of the statistical methods rely on the assumption of a normal distribution to whatever phenomenon is underpinning the data. As more critical writers, notably Benoit Mandelbrot and Nassim Nicholas Taleb, have noted, too much faith in the normal distribution is more the cause of a lot of current complaints.
Charmingly, Ayres closes this readable but shallow account by arguing that a better public appreciation of normal distributions and standard deviations would make the world a generally better place. In the set of utopian missions, it's a bit of an outlier.