In article ,
msnyder{at}redhat.com says...
>
> Mark Borgerson wrote:
> > In article ,
> > msnyder{at}redhat.com says...
> > >
> > > Mark Borgerson wrote:
> > > > In article ,
> > >
> > > > > They picked them because they COULD obtain a very large
> > > > > and very representative sample from them.  It wasn't a few
> > > > > dozen -- it was a significant fraction of *all* of the large
> urban
> > > > > counties in the US, and "large" doesn't
mean "bigger than a
> > > million".
> > > > >
> > > >
> > > > A large sample---yes.  Representative of the whole US?   I doubt
> it,
> > > > or they would have said so.
> > >
> > > They did say so, Mark:
> > >
> > >     The sample was drawn from State prosecutor files in 33 of the
> > >     75 [most populous counties in the U.S.].  The counties were
> > >     widely scattered, from Los Angeles and San Diego, Denver and
> > >     Dallas, to Philadelphia and Dade County (Miami).  [...] The 75
> > >     counties are where a little over half of all murders in the
> > >     nation occur.
> > >
> > > The sample, therefore is representative of a little over half
> > > of all murders in the U.S.
> > >
> >
> > Yes,  but they represent the half that occurs in a specific
> > demographic data set---large counties with populations
> > greater than about 500,000.   The DOJ acknowledges that
> > the patterns of homicide are different between large cities
> > and smaller cities, suburban and rural areas:
> >
> >
> > This report from the DOJ says :
> > " Intimate homicides (spouses, ex-spouses, boyfriends, and
> girlfriends)
> > made up a larger percentage of murders in rural areas than in
> suburban
> > or urban areas"
> >
> >
> > http://www.ojp.usdoj.gov/bjs/homicide/city.htm
> >
> > Since murders in rural and suburban areas are more likely to involve
> > intimates and family members,  I think that my hypothesis that
> > the spousal murder statistics for the 75 largest counties may not
> > be representative of the statistics for rural and suburban areas
> > is supported.
> >
> >
> > You might also find it interesting that murder by poison (more common
> > among women offenders) represents a disproportionately large
> percentage
> > of murders in suburban and rural areas.  (See the DOJ report cited
> > above).   That would indicate to me that murder of men by women
> > might actually be more common in suburban and rural areas  than is
> > the case in the large cities.
>
> Then why all the quibbling?  A study that represents more than half
> of the *entire possible sample* is  certainly large enough to be
> significant.  It's always possible for those outside the sample
> to have changed the results, but you yourself argue that it isn't
> all that likely to have changed them in the direction that would
> make them less interesting.

You engaged in a study of subset of homicides--intimate homicides---but
the sample population that you picked accounts for no more than 40.9% of
those homicides (probably less, since the DOJ study cited says that
40.9% occur in cities of more than 100,000, while the bottom end of the
75 largest counties is about 600,000 people.)
>
> Mark, this is what you do on a regular basis -- find some trivial
> quibble and then beat it into the ground.  Why don't you try
> commenting on the content instead?

1.  I dont' feel its a "trivial quibble"  to point out that the
demographics of your sample population may not be representative
of the entire population of the USA.  Pointing that out isn't
a criticism of your analytical results.  It is a question
about whether the result should be applied outside the
sample population.   It may be a minor quibble to you,
but it is a valid question to a statistician.

2.  I did comment on the content.  I provided a cite to a
DOJ study which explicitly says that the characteristics
of spousal murder  differ between urban and suburban/rural
areas.   That is  a definite  indication that studies
based on that sample may not be applicable outside the
sample population.   In any discussion based on statistical
analysis, discussion of the characteristics of the sample
population is pertinent.

Mark Borgerson







--- UseNet To RIME Gateway {at} 4/1/05 8:48:01 PM ---