Comments on: Not fair, not true

By: Sinclair Davidson

Sinclair Davidson — Mon, 08 Mar 2010 00:12:42 +0000

I agree McCloskey’s overall standard is much higher than what we’re talking about here. I suspect that we’ll never really live up to her expectations – but that’s another point. The argument here, I think, fits into her ‘sign econometrics point’ – I’m very loathe to make a fuss about a coefficient that is either positive or negative if it’s not statistically significantly significant different from zero.

On the Bayesian issue, I don’t know enough about it to say one way or the other.

(On the point about Austrian economists should stop crapping on about meta issues and do economics, we are in total agreement.)

By: John Quiggin

John Quiggin — Sun, 07 Mar 2010 22:48:33 +0000

A nice point. But McCloskey is applying a much tighter standard than was violated in the present case, and I think one that is tighter than most econometricians would consider justified. Adrian Pagan, who knows more about econometrics than anyone was pretty scathing about the McCloskey paper when she presented it in Canberra a while back.

In her view simply reporting that there exists a statistically significant relationship without an analysis of the economic implications of the point estimate is an error. I don’t think this can be sustained. For (+ve) coefficients that are marginally significant (say t<3), you can't really say anything about the magnitude except that is positive, and that you have some kind of upper bound. In such cases, reporting the hypothesis test and not saying much about the estimated magnitude is perfectly reasonable. The exception would be if the effect is of a different order of magnitude than theory would predict, as for example, a tiny effect showing up as as significant in a gigantic sample.

At points, McCloskey comes close to saying that we should ignore statistical significance altogether – this is wrong in my view, though I'd prefer a Bayesian approach.

In any case, this is a far cry from translating "not statistically significant" as "insignificant" or "nonexistent" as has been done here. That really is an error so basic as to disqualify you from talking about data.

By: Sinclair Davidson

Sinclair Davidson — Sun, 07 Mar 2010 07:01:24 +0000

All the GFC proves is that economists can’t foresee the future, everyone knows that already.

By: rog

rog — Sun, 07 Mar 2010 06:57:24 +0000

Seems like a bit of an own goal Sinclair; you have provided proof that in the main economists lack statistical cogency whilst the GFC proves it.

By: Rafe

Rafe — Sat, 06 Mar 2010 21:10:42 +0000

No amount of statistical savvy is any good unless you ask the right questions. First have a hypothesis that is worth testing.

By: John H.

John H. — Sat, 06 Mar 2010 07:47:44 +0000

It is getting worse. Of the 137 relevant papers in the 1990s, 82% mistook statistically significant coefficients for economically significant coefficients (as against70%in the earlier decade). In the 1980s, 53% had relied exclusively on statistical significance as a criterion of importance at its first use; in the 1990s 64% did.

Thus I can no longer trust an economist when he\she states, the data is clear …. .