Daniel Davies (permalinks bloggered) has an excellent post on data mining, using everybody’s favorite example of dodgy research, John Lott. I discussed the same issue here and “with specific reference to Lott, here.
In practical terms, Davies suggests that economists should keep log books recording their estimation strategies. I’ve urged this kind of requirement in the past. The scandalous fact is, however, that in the great majority of cases, economists don’t even keep their datasets, at least not in a publicly accessible form. To be fair, Lott did at least make his data available, though he didn’t record the data mining that must have been necessary to derive results that were so uniformly favorable to his case.
John Quiggin 
I would have thought a better example of data mining would have been the Card-Krueger study of minimum wages and unemployment, which Quiggin and the ACTU love to cite in their submissions to the Industrial Relations Commission, but which has also been found to be completely fraudulent and has been thoroughly discredited by the economics profession.
“the Card-Krueger study … has been thoroughly discredited by the economics profession”
No doubt that’s why the profession awarded Card the JB Clark medal for 1997, largely on the strength of this work.
George B, have you actually ‘Myth and Measurement’? If so I reckon you wouldn’t say that – it’s positively pedantic about methodology, and the results seemed to genuinely surprise the researchers.
Card & Krueger, though, are often misused. They’re very specific that their monopsony effect only matters where minimum wages are low (otherwise the conventional ‘priced out of the market’ effect holds). But Australia has a very high minimum wage – the highest in the OECD on some measures, and near the highest on most – so Card & Krueger shed NO light on what our Federal Minimum Wage should be. So I guess I disagree with both George and John on this.
George also mis-states the extent to which Steve Dowrick and I relied on Card and Krueger in our work on minimum wages. A large number of studies support our central argument that the relevant elasticities are small enough to ensure that the income effects of higher minimum wages outweigh the effects of reduced labour demand.
This really gets at what archpundit was talking about earlier – that people don’t understand social science. It’s wrong to claim card and krueger are definitive, and it’s wrong to claim that their results have been discredited.