Neil (![[info]](http://l-stat.livejournal.com/img/userinfo.gif){kind=small} eldar) wrote in mooism, |
I once attended a seminar on Benford's law. Basically it's great for looking at sets of figures (typically accounting...) to spot ones that have been made up.
Generally speaking, in a genuine sample of sufficient size, there will be more values starting with a 1 than any other digit. (It's not significantly more than the 11.1...% you'd perhaps expect, but enough for it to be spotted in a good sample size.)
So applying it to election results when there are enough votes cast is perfectly valid.
[Edit] Correcting myself - it was over 10 years ago I attended this seminar - 1's occur far more frequently than 11.1...%!! So yeah if there are too many vote counts starting with a 7, I'd say they're bogus.
Generally speaking, in a genuine sample of sufficient size, there will be more values starting with a 1 than any other digit. (It's not significantly more than the 11.1...% you'd perhaps expect, but enough for it to be spotted in a good sample size.)
So applying it to election results when there are enough votes cast is perfectly valid.
[Edit] Correcting myself - it was over 10 years ago I attended this seminar - 1's occur far more frequently than 11.1...%!! So yeah if there are too many vote counts starting with a 7, I'd say they're bogus.
