--- In [email protected], "authfriend" <[EMAIL PROTECTED]> wrote:
>
> --- In [email protected], akasha_108 <[EMAIL PROTECTED]>
> wrote:
> >
> > Here are annual % change figures for DC violent crimes (and per
> > 100,000 pop figures). Its interesting to see the big swings. This
> > underscores the fact that there are probably a number of factors
> > driving crime levels. If these are not controlled in the analysis,
> > short term swings, annual, but most certainly 8 week periods, can be
> > due to factors totally unrelated to the "intervention".
>
> It's my impression that controlling for these factors
> was precisely what the researchers were attempting to
> do with their statistical methodology.
I wish their data was available -- to see what they actually tested
for and the interval of the data (annual / weekly, etc).
The problem is a lot of important variables totest for in this regard
are usually available in annual not monthly or weekly form. Maybe
weekly was available a decade + ago but its hard to find now. For
instance, even the FBI crime data is available weekly only back to
1995 (through them). And deomgraphic and economic variables are often
only available in annual form. And most available weather data is
averaged over 20-40 years to give typical days. But for this analysis,
the actual annual and monthly time series are required.
While I know you have a distaste for speculation, it would appear, per
my prior post outlining the four or more model specifications they
used -- per the summary, that the subordinate analysis ("model 4" in
my note), they may have been stuck with only annual data for this
larger set of deomgraphic, economic and LE control variables. And
thats why they did the analysis in two stages. Weekly short term data
for the ME effect and weather only. And annual data for some testing
of long term trends using the larger set of control variables.
The problem is, this is a quite weak approach. It means that in
primary analysis (and this is NOT speculation, its what they did),
there were no control variables other than weather, and ME, to explain
the variation crime. While the individual variables may have had high
significance (t values), the overall fit of the model may have been,
actually must have been marginal. As anyone can see from the annual
data, there is too much other stuff going on to be explained just by
weather.
Regardless, that is the issue I am faced with in trying to collect
data to do an independent analysis. But as I stated in prior post, if
there was an ME, it should show up in annual data. And the district
data. And using annual data, a unified model can control for all
variables in the same specification. If they were not able to do this
in the primary model, and the summary appears pretty clear that they
were not, it greatly weakens the results.
I am less and less interested in what they did, and more interested in
creating a dataset to redo the analysis, perhaps in a richer and more
defensible way than they did -- using a unified model and accounting
for economic, demographic and LE control variables that they did not
-- and over a longer time period. The core model needs to "explain" or
account for most of the variation in the crime rate in the longer time
period, before it can hope to account for the ME effect in 1993. Again
a 4% change should be detectable for a well controlled model (.15 of
1993 x 25% two month effect). Particularly if the district data is
used. It seems inherent in the "theory" that the ME effect is related
to distance from the "core" so the ME district should show a stronger
effect.
> Box-Jenkins analysis--does that ring a bell?
Yes. Box and Jenkins created the original ARIMA models, the origianl
specifications are often called Box-Jenkins models -- though ARIMA
models have expanded their range since B&J's days. And ARIMA models
can be thought of as a subset of regression models. ARIMA models can
be respecified to fit a regression format, and in doing so, much more
power is adapded to the analysis.
>
> > Also, note that in 1992, the year preceeding the study, there was
> over a 15% increase in crime. The year before was almost 0. So there
> seems to be some snapback effect, high levels may cause more police
> > crackdowns, higher funding levels etc the following year.
> >
> > The Me study "indicated" a 25% drop in crime over 8 weeks. This by
> > itself should amount to about a 4% decrease in the annual rate.
But it went up 3%. But the next year with no ME effect, crime
decreased over 8%.
>
> Could that have been the abortion effect you keep touting
> kicking in?
Partly. The abortion effect, per Levitt's work, explains the majority
of the 50% decline from 93 to 2003. But there are other factors.
What specifically caused the 92-94 variations I would suspect has to
do with police levels and LE funding/practices. And weather
variations. Abortion could account for up to about 4% reduction per year.
------------------------ Yahoo! Groups Sponsor --------------------~-->
Get fast access to your favorite Yahoo! Groups. Make Yahoo! your home page
http://us.click.yahoo.com/dpRU5A/wUILAA/yQLSAA/JjtolB/TM
--------------------------------------------------------------------~->
To subscribe, send a message to:
[EMAIL PROTECTED]
Or go to:
http://groups.yahoo.com/group/FairfieldLife/
and click 'Join This Group!'
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/FairfieldLife/
<*> To unsubscribe from this group, send an email to:
[EMAIL PROTECTED]
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
