Thursday, June 12, 2008

Brendan O' Flaherty notes a problem with Hanna Rosin's piece on Section 8

In an email, he writes:

The Rosin article advances a fascinating hypothesis about why murder or crime rose in Memphis and in the rest of the country. It gets one important detail wrong. Crime did not rise in Memphis or in the rest of the country.

Here is a fairly long time series for UCR murders in Memphis (UCR excludes justifiable homicide, law enforcement, and murders that did not occur in the jurisdiction of the Memphis Police Department):

1990 195
1991 189
1992 176
1993 198

1994 159
1995 181
1996 161

1997 138 Rosin says the trouble started here
1998 115
1999 118
2000 147

2001 158
2002 149
2003 126
2004 107

2005 138 they stopped moving people out of the projects here
2006 149
2007 129 preliminary

If you treat murders as Poisson variables, only the 2004-2005 and 1999-2000 increases approach 2 se, but murders are really stuttering Poisson, which has greater variance. So going from 1996 until today, which is the natural comparison for the effect of project demolition, you have a decrease from 161 to 129. You can slice it many different ways, but my reading is essentially nothing happened. Remember also that the projects get emptied out long before demo. Since the title of the article is murder, i think it's very fair to look at murder.

What about nationally? Everybody knows that murders have come down a lot since the mid 90s nationally. The national rate stabilized around 2000 and has had only minor random blips since. The new data for 2007 show a small drop of 2.7% from 2006, but this is not significant to me. Rosin makes a big deal of the increase in cities 500k-1 m. Again, murders went down in these cities 2006 to 2007.

Rosin also cites the Police Executive Research Foundation report. This was about the change in murder in a non-random sample of large cities 2005 to 2006. Murder went up in about half of those cities, it went down in the other half, and the authors chose to report on all the increases. Half go up and half go down is what you expect from pure white noise.


Brendan says more, including the fact that Jens Ludwig has a regression that more convincingly ties the crime rate in New York to the success of the Yankees than the work that ties murder in Memphis to Section 8.

3 comments:

Anonymous said...

I heard on the radio a few days ago that L.A. was having a spike in crime because of the heat, plus the laker game. Other's theorized that it was the economy. Freakonomics.

Matthew said...

I really don't know whether the Memphis data is meaningful, but I think focusing on crime totals is a mistake. The thrust of that article seemed to me to be that the crime was following the people from the projects to where they moved. Clearly the distribution of crime is a separate from the total number of crimes, and even less would it seem connected to the total number of murders.

Anonymous said...

"...a fascinating hypothesis about why murder or crime rose in Memphis and in the rest of the country."

A rise in national murder rate is certainly not the hypothesis of the Rosin article. Nor general crime rate.