>Leninology wrote, regarding the improbability of finding 300 violent deaths in
>a sample of 12,000 if the true number in Iraq was 60,000
>
>"Could you put a figure on it? Seriously, I'd be interested to know if there's
>a way of
>calculating the odds."
>
>There is a way of calculating the odds, but I can't really do it. You'd need
>the raw data and/or some assumptions. Basically, you'd either assume that the
>data was normally distributed or take a fit to the empirical distribution and
>look at the tails. But we are talking fractions of a per cent here; if there
>were 60,000 violent deaths in Iraq since the war you'd expect to find 27 deaths
>in a sample of 12,000, not 300. The design effect (the fact that it's a
>cluster survey and finding one death in a cluster means you're likely to find a
>few more) is going to mean that it's biased upward somewhat, but even so, for
>the true value to be a tenth of the sample mean is really very unlikely indeed.
>
>
>
Daniel's right on this: the chance is so astronomically small it's
meaningless to quibble about the number of zeros to the right of the
decimal point. Given these data, arguing in favor of the official
60,000 number is about as plausible as assuming that the lottery ticket
my wife bought yesterday is the $23 mil powerball winner.
Miles