The actual number is 34%, as I stated above. The survey didn't ask how
many times it had happened. An average of two for those that had been
thwarted is my guess based on assuming that the distribution of the
number of thwartings was Poisson (ie, each criminal was equally likely to
encounter an armed victim).
jmaraldo writes:
I have not had the pleasure of making Mr. Poisson's acquaintance - ie, I
know nothing of statistical methods :=) But why does he suppose that the
likelihood of encounter with an armed victim is equal for all criminals
though criminals choose the type of crime in which they will engage?
Because, it makes the calculations simpler :-). Of course, it's only
an approximation. A quick simulation reveals that even if the risk
differs greatly for different criminals, so that 80% of the DGUs are
against 20% of the criminals, the average number of encounters for
those criminals that had at least one would be three. This amounts to
a 50% increase in the estimate of DGUs to about 100k.
And on a slightly different subject, aren't we comparing apples and
oranges when we compare the response of convicts to the response of
citizens at large? Why assume that convicts' experience of meeting armed
resistance is representative of the experience of criminals generally,
when the convicts comprise a (very small) subset of the nation's criminal
population? Wouldn't a random sampling of civilians be more representative
of the general population's experience of armed defense than a random
sampling of convicts?
Other things being equal, yes. However, it seems unlikely that errors
introduced by using a somewhat biased samples will not be that great.
(If DGU makes it more likely for a criminal to be caught, then the
bias will cause an OVERestimate.)
I don't see how we can explain the enormous disparity in the estimates
by sample bias. Some people must not be telling the truth.
Generally we would expect non-criminals to more honest, but the choice
is between most of the criminals lying, and 3% of the general
population lying.
In any case, Kleck is clearly wrong when he cites the 34% statistic as
being supportive of his result. Without exception, every single
crosscheck I have been able to carry out indicates that his estimate
is way too high.