Are data-miners made or born?


Green (1990) gave 199 students the same data but with different errors, and asked them to find an appropriate specification. All students had been taught that models should be specified in a theoretically sensible fashion, but some were also taught about how to use F tests and goodness-of-fit for this purpose. These latter students were quick to abandon common sense, perhaps because using clearly defined rules and procedures is so attractive when faced with finding a specification.

As with driving and flying fighter jets, the risk of 'accidents' when doing econometrics is at its highest amongst those with some, but not much, experience. The novice pilot, or econometrician for that matter, will take extra care to make sure everything is as it should be. Being concious of the possibility of disaster, she will not attempt flashy tricks, whether that's flying at mach-2 or drawing inferences from not-very-well-understood statistics.

As the student becomes more comfortable and builds up some confidence in her abilities, she will tend to underestimate the amount of care that needs to be applied when performing a given manoeuvre. Flight instructors are well aware of this tendency, but it is still the case that most accidents involve pilots that have had between 500-1000 hours of flying experience. A very similar risk profile holds for budding econometricians; we should be thankful that the aftermath of a 'crash' for the analyst is far less painful than for the pilot.

The quoted excerpt is from Peter Kennedy's excellent econometrics textbook, now in its fifth edition. As far as I'm concerned, the book is unique in its approach; it focuses on the intuition behind the various methods and techniques used in econometrics using simple English, with the underlying equations relegated to technical annexes.

I have long thought that the way undergraduates are being taught econometrics is far from ideal: what good is it learning the proof of why OLS is BLUE in your second week in an introductory econometrics course? Kennedy's approach is promising, and purchasing the book is a good idea for students that have had little experience with econometrics and are not yet proficient in 'translating' the maths into something that makes intuitive sense.

The issue of the Political Methodologist where Green's article comes from is here (free access), but I have to warn you that the quality of the scanning is poor.

0 comments: