How to be Human (Though an Economist) by Deirdre McCloskey, kindly recommended by Jason Soon. As an anecdote, it’s hard to top the story of the reaction of the dean of the Iowa business school when Donald (as he then was) announced his impending change of gender.

His response, after sitting for a moment in slack-jawed amazement, was a stand-up comic routine. “Oh, thank God! I thought you were going to confess to converting to socialism. (Relieved laughter- he was going to react as a friend.) “This is great for our affirmative action program: one fewer* man, one more woman” (more laughter) ” And wait! I can cut your salary to two-thirds of the male level (not so funny). And then seriously “That’s a strange thing to do. How can I help?” And he did

* I wonder if business school deans, even civilised ones, really use “fewer” rather than “less” in circumstances like this, or if McCloskey has done some editing here? Not that it matters to the story.

In the pursuit of the goal of humanising economics and economists, McCloskey recommends a variety of reading. In a couple of places she notes, as an indicator of a civilised economist, acquaintance with the companion volumes of Samuel Johnson’s A Journey to the Western Isles of Scotland and Boswell’s Journal of a Tour to the Hebrides with Samuel Johnson. I agree, particularly regarding Boswell’s book, which is a kind of pilot episode of his Life of Johnson, the first genuinely modern biography, and still one of the best in existence.

I’ll turn now to the bits that are interest mainly to economists, and other social scientists (or, as McCloskey might prefer to put it, scholars of society).

The book consists of reprints of short articles, so there is some repetition, particularly of McCloskey in Cassandra mode in her denunciation of what she sees as the three emblematic evils of modern economics – existence theorems, significance testing and social engineering. I’ll defend social engineering another time, but for the moment I’ll consider what defences can be made for existence theorems and significance testing.

Existence theorems, for McCloskey are the archetypal example of ‘blackboard economics’, mathematical games yielding purely qualitative results that can be overturned with modest changes in assumptions. They were the high point of mathematical economics in the 50s and 60s, and I went into economics in part because I thought my (long-decayed now, sad to say) expertise in fixed-point theorems would assure me of a comparative advantage on this topic. Members of Generations X and Y can substitute game theory to get the same associations.

I’ll offer a partial, and somewhat backhanded, defence. One of my professional interests has been the development of new and improved index numbers. I hope that McCloskey would approve of this as a useful form of economics, since my interest in the topic (commonly regarded as deathly dull) was stimulated by the treatment in The Applied Theory of Price by D.N McCloskey. Anyone who enters this field rapidly becomes aware that while various properties might be desirable in an index number, they can’t all be had. There are a wide variety of ‘impossibility theorems’ demonstrating the non-existence of index numbers with various properties. Familiarity with such theorems can save a lot of pointless effort, and they are therefore worth looking for. But an impossibility theorem is just the negative form of an existence theorem (or, if you prefer, an existence theorem proves the impossibility of the corresponding impossibility theorem).

This is a rather prosaic defence, that certainly does not justify the high status accorded to the kind of theory exemplified by existence theorems. But the argument can be pushed a bit further by considering the most famous impossibility theorem, that of Arrow who showed (roughly speaking) that no voting system having a set of seemingly desirable properties could work for all possible sets of voter preferences. This impossibility theorem precluded a lot of potential effort in designing ideal voting systems. It also stimulated some more limited, but still, I think, important positive results. In particular, a variety of voting systems can be made to work as long as voter preferences are arranged on a spectrum (say from left to right) with a peak in the centre

It is a surprising fact that, in smoothly functioning political systems, this tends to happen, even when a given system deals with issues that are seemingly unrelated. For example, there is no obvious link between the view that the Tampa refugees were treated unfairly and the view that Australia should become a republic, but I’d guess that most people who held the first view also hold the second (not true in reverse, since more people support the republic than supported the refugees, but I’d guess that the great majority of monarchists backed the government on the refugee issue). 100 years ago, the correlation would probably have been reversed- the Bulletin favored republicanism and “Australia for the White Man”.

Having offered a backhanded defence of existence theorems, I’ll be even more backhanded in defence of tests of statistical significance. McCloskey is right in pointing out that there is no relationship between statistical significance and economic significance and that, even among adherents to the logically dubious tenets of classical statistics, very few people interpret significance tests and confidence intervals correctly. She is also right in saying that no-one nowadays takes a finding of a statistically significant relationship very seriously, since most such findings can be reversed with a change of specification, and since many “statistically significant’ relationships may be quantitatively unimportant.

Still, going the other way, I wouldn’t want to place much weight on an estimate that was quantitatively important, but statistically insignificant. The rigmarole of t-tests, Type 1 and Type 2 errors, and so on may be nonsensical, but at does at least weed out some potential claims that are unsupported by the evidence . In the absence of a generally-agreed loss function, requirements for “statistical significance” at least provide some sort of benchmark for paying attention.