Re: Crime Data

[Kromkowski@aol.com]

In a message dated 01/31/00 9:02:54 PM Pacific Standard Time, Mowery writes:

<< I have found in practice there are two good ways to address this.
 First is to use a P-chart of "murders/population."  This can be problematic,
 however, as population data is not always available.  >>

Very curious.  While I have used U-charts to plot crime data, the P-chart is 
not one that comes to mind.  

I have generally stayed out of this thread (not sure how the titled got 
chosen and where it actually began), but offer some thoughts for moving 
forward.

1.  That we haven't we haven't come up with a consensus for the type of chart 
is kind of troubling since this the DEN. (My view is that when there are a 
variety of reasonable choices for type, you can't go too wrong, especially 
when data is a time series, by using the IndX-MovR.)

2.  The construction of a control chart must always be followed in some 
fashion by an identification (or call for identification by knowledgeable 
subject matter experts) of 

     (a) a proposed explanation for any identified signals of special causes 
of variation, and a way to find out if you were right.

    (b) a proposed identification of common causes of variation. For example 
in today's Baltimore Sun (Olesker), I read that the vast majority, whatever 
that means, of _victims_ have prior criminal records involving on average 
(whatever that means) 8 prior occurrences.  Apparently not a lot different 
from the profile of murders. Previous articles have also provided data about 
Age and gender and drug-usage.  So for example, hypothesized common causes 
for murder might be:

    the amount of people between 15 and 30,
    the number of males between 15 and 30,
    the number of non-incarcerated persons in population with a "large" 
number of criminal arrests,
    the availability of drugs (primarily alcohol, amphetamines and cocaine),
    the availability of hand guns,
    the availability of bullets,
    the size of the municipality
    the density of the population (both very high densities as well as very 
low densities in urban areas)
    etc.
    
and, then  

    (c) some kind of statement for future, a hypotheses or general direction 
for research or change (improvement), that builds upon (a) and (b).


3.  Many urban police departments talk about the use of mapping and 
identification of so-called "hotspots" and shifting deployments based there 
on.  Software is alleged to be involved.  

    (a) Any ideas on the method used for determining "hotspots" by these 
computer programs.

    (b)  Why does this remind me of a potential for a real life funnel 
experiment?**

** A little digression, which I may have mentioned before but which I repeat 
as a Deming Philosophy success story.  I began using voice-recognition 
software in 1994. I had followed the developments for many years prior and 
deciding to start my own office gave me reason to give it a try while I 
waited in those first few months for the work to come in. Early software, as 
a default and an ostensibly preferred method, used to continuously correct or 
adjust for the user's voice/accent.  The claim was that it was "always 
learning" your voice.  This seemed far too much like the funnel experiment.  
Having developed a friendship with someone at one of the companies, I was 
able, by data and by citations to Deming in OOTC and citations to the works 
cited by Deming and perhaps some persuasive rhetoric on my part and 
investigation on their part, to convince the company to stop using the 
continuous adjustment scheme as the default and eventually, when the software 
and processing speeds moved to continuous rather than discrete speech 
recognition, the continuous adjustment was dropped entirely, not only by the 
company I was involved with, but by the industry.

John David Kromkowski
Kromkowski@aol.com
========================================================================