--- In [email protected], new.morning <[EMAIL PROTECTED]> wrote:
>
> --- In [email protected], "sparaig" <sparaig@> wrote:
> >
> > > And the comment, "this is precisely the type of sporadic fluctuation
> > > one must account for when total numbers are small." Exlusion is a
> > > major way of "account[ing] for" such outliers. 
> > 
> > 
> > It was NOT excluded from the data. The weekly AVERAGE was the weekly
> average with no 
> > data excluded. The SIGNIFICANCE of the outlier was dismissed because
> it was only one 
> > data point amongst many and wasn't repeated and in fact was reversed
> the next week (20 
> > one week and 4 the next).
> 
> See footnote 3.
> 
>    3. After removing the outlier of June 22, Poisson regression
> analysis indicated there was no significant difference in the level of
> homicides in June and July 1993 from the remainder of the year.
> 

Ah you're right. Interesting that something that changed the average so 
slightly over a 2-
week period would have a significant effect. It suggests that the murder-rate 
trended 
upward slightly even without the outler.


> -- 
> This implies that WITH the outlier included, there WAS a significant
> difference [higher] in the level of homicides in June and July 1993
> from the remainder of the year.
> 
> And at best, the ME had no effect on murders. At worst, it increases
> them -- assuming ME is causal. So it would appear, ME is
> "crime-stopper light". Works best, thjough modestly, on lighter-weight
> crimes I guess. And even then, its not a huge decrease. I wonder if
> SSRS or AMMA can provide a heavy duty crime fighting effect. :)
> 
> By the way, I assume each diamond on chart II is one week. If so,
> there was a sizable drop in HRA crimes in week 9 and 10, AFTER the
> course. Which sort of disturbs the immediadiacy of the ME and that
> "This shows that usually the violent crime levels were directly
> proportional to temperature — and therefore that violent crime could
> be accurately predicted from the previous pattern in the data."
> 
> Of course you could postulate a lag effect -- which I have for the
> financial markets impacts -- but then you have to discount the drops
> in weeks 1 and 2. Which occurred by the way, just after a large spike
> pre-course. So much for very strong temperature correlated crime trends.
>


Reply via email to