Chris forwarded me this article on the disagreement between economists Lott and Levitt on the topic of crime. In summary, Lott thinks that carrying guns helped reduce crime in some places (and supports this assertion by data). On the other hand, Levitt thinks that legalizing abortion helped reduce crime (and also supports this assertion by data). What's notable about the extent of this disagreement is the defamation lawsuit described in the article.
The type of analysis I like a lot is Phil Henshaw's detailed analysis of what happened with the crime in the NYC. Although this picture of data by precinct (reproduced below) seems to give a lot of support that it was neither guns nor abortions, but a political/police crackdown, along with a tremendous amount of complexity around it. Phil's time series of crime from 1960's to now also provide an interesting explanation for the original increases in crime: the crime bubble's rise was through the events between 1968 and 1972 (although the pre-1965 data might invalidate this observation). In the present context, I should also mention Gladwell's tipping points.
Those who wish to check state-level data for themselves can use
I should also point to Gladwell's take on this issue (which resembles mine), and the Freakonomics response, followed by Gladwell's response to the response.
How do you disentangle the abortion hypothesis from the Giuliani hypothesis? The missing abortion kids would have been beginning their prime crime years about the same time the "police crackdown" began.
I would like to see precinct level data for Jersey City or Philadelphia for the same time period before I am going to give the mayor all the credit.
One of the great things about reading time series for the stories they tell about complex events is that they open up lots and lots of stories you'd otherwise not suspect. The NYC precinct data curves for murder rates through the crack epidemic shown above (75 overlapping traces) does not show you how they should be grouped to display the different courses and timing of events at the larger scale. After experimenting with a few other ways, it seems that aggregating the date for each of the 5 boroughs displays the behavior of 5 cohesive communities with very different responses to the same invading vector.
[http://www.synapse9.com/cw/crimewave_nys2.htm]
[http://www.synapse9.com/cw/nyc-boromrate-dr.jpg]. There's much more to be done, but I haven't gotten the interest of the criminologists at John Jay or the NYC PD, either of which might be able to make the GIS data available for tracing the local complex system dynamics.
Responding to Anonymous: Looking at the raw data at [usdoj.gov] shows a spike of youth delinquency peaking at around 1991, which slowly tapered off. Before the availability of legalized abortion, there was no such spiking.
But yes, the data in this blog entry only show that a rapid crackdown did work in NYC. The reduction of crime was slower elsewhere, only stabilizing around 2000.
Aleks,
I think the graph would be clearer if it plotted something normalized rather than raw counts. As is, so much of the visual impact is cluttered by the variation between precincts, but what you really want to focus on is variation over time.
Phil's graph does show the simultaneous rapid decrease in all the problematic precincts: something one would expect when trying to improve the performance of police with limited resources. Smoothing or aggregating would obscure this intervention. I guess it's best to observe the data at the level where the actual causal interventions happen.
But it is clear that 1994 interventions only explain a part of the variation.
Aleks,
My suggestion was normalizing (e.g., counts per population, or even something similar like normalizing each precinct by dividing by total murders from 1987-1992), not smoothing or aggregating. I agree that smoothing or aggregating would not be such a good idea.
P.S. I would think it would be helpful to see data from several different cities, as well. I think some people have done this, actually. Perhaps Jeff Fagan would have some helpful comments.
I've read quite a bit about both hypotheses and have examined data looking at these trends. I don't think Lott or Levitt have it right. IMO, Steve Sailer's critique of Levitt's work is dead-on, and Levitt has admitted to some methodological errors after this.
As for Lott's work, I think he can make this case only by tweaking his statistical model to intentionally produce this. Other researchers have used his data (the data that didn't mysteriously dissapear in a hard-drive crash with no backup copies of any sort) and have produced coefficients completely in the opposite direction by tweaking the model in a similar way. If the hypothesis is correct, then I don't think one needs an overly-complex model to explain such a simple and parsimonious theory that should have such a robust effect.
Steve Sailer, as far as I can tell through his bewildering posts at the Freakonomics site, is a certified nut, in addition to being a troll.
It wasn't Guiliani. Don't belive the hype. End of crack era. Murders dropped. Still high in certain areas today. End of story.
A delayed reply... LifelongResident is quite right! When the data is aggregated by community (i.e. independent social units rather than precincts) you see the clear peak and then following collapse of the crack epidemic crime wave as the main event. The numbers of murders starting fall like a rock 3 years before Guiliani was elected. His innovations may have reflected a police department geared up for a war but finding less than expected to do! It shows that you can use statistics to find a phenomena, but if you don't find the phenomena, the statistics dont tell you much. http://www.synapse9.com/cw/crimewave_nys2.htm