It is interesting to consider how effects like this operate at different levels.
For example, if I picked up a coin, flipped it 6 times, and got heads each time, then I might reasonably be suspicious about the coin. On the other hand, when someone claims to have flipped a coin 200 times, it is the ones who never get 6 in a row that are suspect.
Similarly, a tax return that does not have numbers beginning with 1 is certainly suspicious. On the other hand, if you gave me a million tax returns, and all of them had lots of numbers beginning with 1, then I would be suspicious of the collection as a whole. Benford's Law deals only with probabilities, after all; we should expect some deviations from what it predicts.
Given the sizes of data sets that governments deal with, the ideas outlined in this article can certainly be useful. However, the virtual certainty of false positives is something that also needs to be dealt with. It is to be greatly hoped that these ideas are implemented in the context of human wisdom and manual fact checking, and not merely as a yes/no way of determining who gets treated like a criminal.
In any case, this was a thought-provoking article. Thanks for posting.
I have a copy of Nigrini's book. I've tried getting copies for my company, but other than the occasional $200+ used copy on Amazon, it appears like this book isn't just "out of print" but withdrawn from the market (the publisher sells training now, not books that anyone can read).
Comments
It is interesting to consider how effects like this operate at different levels.
For example, if I picked up a coin, flipped it 6 times, and got heads each time, then I might reasonably be suspicious about the coin. On the other hand, when someone claims to have flipped a coin 200 times, it is the ones who never get 6 in a row that are suspect.
Similarly, a tax return that does not have numbers beginning with 1 is certainly suspicious. On the other hand, if you gave me a million tax returns, and all of them had lots of numbers beginning with 1, then I would be suspicious of the collection as a whole. Benford's Law deals only with probabilities, after all; we should expect some deviations from what it predicts.
Given the sizes of data sets that governments deal with, the ideas outlined in this article can certainly be useful. However, the virtual certainty of false positives is something that also needs to be dealt with. It is to be greatly hoped that these ideas are implemented in the context of human wisdom and manual fact checking, and not merely as a yes/no way of determining who gets treated like a criminal.
In any case, this was a thought-provoking article. Thanks for posting.
I have a copy of Nigrini's book. I've tried getting copies for my company, but other than the occasional $200+ used copy on Amazon, it appears like this book isn't just "out of print" but withdrawn from the market (the publisher sells training now, not books that anyone can read).
http://www.amazon.com/Digital-analysis-using-Benfords-Law/dp...
http://www.nigrini.com/data_analysis_books.htm