In article , "John M. Wildenthal"
 wrote:


> If the NCVS doesn't survey incarcerated individuals it certainly *under*
estimates
> rapes, though there shouldn't be any DGUs by prisoners ;)  IIRC, prison
rapes are
> a substantial proportion of total rapes.  Back to non-prisoner rapes,
NCVS, and
> Kleck.

You are correct, prisoner rapes are not reasonably part of the discussion.
 
> If the rapes are of the spousal variety and successfully fended off,
again, the
> NCVS respondent may decide no crime was committed and therefore not be
asked about
> DGU.  In some sub-cultures some domestic violence appears to be
considered normal
> or at least tolerated.
> 
> Second, if the respondent felt reporting a rape that was successfully defended
> might lead to prosecution over the means of defense, they have a
disincentive to
> accurately respond to the NCVS survey.  This is true of all of the NCVS
covered
> crimes.  I know if I had a DGU in Chicago or New York, I wouldn't want
to report
> it to someone working for the government.  I *might* report it to an
independent
> researcher.

This problem, if it is one, arises with Kleck as well as the NCVS. The
NCVS is conducted by the Bureau of the Census, not law enforcement.  I
personally would be more suspicious of the intentions, and affiliations,
of someone claiming to be doing a survey of gun use as some sort of
allegedly private poll, than I would be of someone straightforwardly
claiming to be working for the census.  Obviously some of us are more
suspicious of phone pollsters than we are of census takers, and some are
more suspicious of census takers than phone pollsters.  You may be able to
control for this differential suspicion, but I don't see any easy way to
do it. 

> > What these anomalies tell us is not that the NCVS significantly
> > underreports rapes, which is probably untrue, but in any case irrelevant
> > (since all of the surveys under consideration derive their results from
> > people who *are* willing to talk about the crime), but that studies using
> > Kleck's methods are susceptible to extreme inflation in DGUs produced by a
> > relatively small number of false positives.
> 
> I think it should also be noted that since Kleck & Gertz do not
necessarily define
> crimes the same way the NCVS does, that we should expect some
discrepancies as the
> definitions vary.
> 
> The NCVS was not primarily designed for estimating DGUs.  It was designed to
> estimate crimes.  The design choices made (specifically the lack of anonymity
> encouraging false negatives) suggest that the NCVS count is probably biased
> downward in their attempt to avoid false positives.  The Kleck & Gertz
methodology
> has the possibility of either direction - as Jim notes in other messages, some
> might not answer Kleck & Gertz for fear that the survey is really an FBI
or ATF
> sting.
 
> Jim, you seem to be greatly worried about the possibility of false
positives.  

The reason we know the false positives are there is that we get impossible
anomalies, like the rape problem and the missing dead and wounded problem
(which Kleck himself admits exists, and IMO has no good explanation for).
We easily derive the absurd figure, which seems to have started this
thread, that there were 14 crimes prevented by defensive *gun* uses in the
first 12 hours of the new year in the Austin metropolitan area.  This is
more than half of the crimes normally reported here in this time period. 

There is another reason we can't trust Kleck's figures, which may or may
not be due to the false positive problem, i.e., the wildly different
results in different polls using Kleck's methods, some of them conducted
by Kleck and some by others. This may have more to do with the small
sample of real DGUs in a survey of the size done by Kleck, but intuitively
(and I admit I can't prove this) I would expect there would be a greater
variability in the number of liars in a small sample than in the number of
real DGUs. 

>How
> do you suggest changing the Kleck & Gertz methodology, reducing the
likelihood of
> false positives without encouraging (additional?) false negatives?  Or
perhaps, if
> you believe that false positives are such a problem and lead to
significant bias,
> you could devise a way to estimate the amount of bias.  Or you could
work on a way
> to estimate the false negatives for the NCVS data.  Bias estimation has
been done
> in Economics for revealed valuation, but this is a much more difficult
problem.
> My only suggestion would be to have the two surveys' overlap the same
respondents
> and see what information could be gleaned from any resulting discrepancies.

Frankly, I don't know how to weed liars out of a sample.  It would, I
think, be unethical to breach an assurance of anonymity given to a
respondent by checking out his story with the police, in the cases where
he said he reported it to the police.  (The NCVS, of course, is prohibited
from revealing personal data to other agencies). Even if you did, the
percentage of liars derived from an independent check of checkable facts
would not necessarily get you a percentage of liars when you have a story
where the facts are *not* checkable.  It's a difficult problem. 
Overlapping surveys would bias the second survey in the direction of the
results of the first one, because the respondents in the second would
remember what they said in the first, and suspecting that the results
would be compared, would reproduce the false positive or false negative
(if any) of the first survey.

I think honest people will have to say simply that the NCVS figures are
too low, and Kleck's too high.  How much too low or too high cannot, as
far as I know, be accurately ascertained.  

Another real problem with Kleck's survey (or any other) can be overcome by
simply changing a word. Instead of calling the results "Defensive Gun
Uses" we can call them "Confrontational Gun Uses" (CGUs).  Kleck has no
way of knowing that the reported CGUs were actually defensive in nature,
as claimed.  It is unreasonable to think that in 100% of the CGUs, the guy
with the gun was acting defensively.  In the absence of any hard evidence,
50% is a better estimate.  But neither you, nor I, nor Kleck, has any way
of finding out, unless we add investigative journalism to the tricks of
the pollsters trade.

Best regards,

--Jim McCulloch