Get the Lead Out II

There is an absolutely serious hypothesis that the sharp drop in violent crime in the USA since the early 1990s is due to the EPA and, in particular, to the shift from leaded to unleaded gasoline roughly two decades earlier.

Kevin Drum notes that James Q. Wilson ascribes about half of the total decline to lead exposure. The hypothesis makes sense, lead poisoning too mild to cause other symptoms is very strongly correlated across individuals with violent behavior.

The hypothesis has clear implications for violent crime in other countries, since the USA banned lead early. The UK went unleaded in 1986 12 years after the USA. The USA violent crime peak came in 1995. So the predicted UK crime peak would be in 2007.

In get the lead out I, I thought the USA went unleaded (for new cars) in 1973 so with an arithmetic fail, I predicted 2008 for the UK. Notably, the post is dated 2008 so I had only a very tiny bit of data.

Now they have the number of reported violent crimes for two more years 2008/9 and 2009/0 (I have no idea why they report by a period other than the calender year).

I have a huge ugly pdf with the data I need and much much more.

The bottom line is that the total number of violent crimes was basically identical in 2004/5 2005/6 and 2006/7 then declined about 17% by 2009/10. The predicted peak of 2007 corresponds about as precisely to the data as is conceivable. This really was an out of sample prediction (except I forgot the year of the US requirement that new cars use unleaded but that’s my fault not the fault of the lead hypothesis).

A hypothesis has yielded a correct out of sample prediction about violent crime rates. This isn’t just extrapolating the trend, the prediction was made right at the time of the predicted peak.

The problem is that I have too much data. I have an illegible screen shot, but I will just retype the relevant numbers after the jump.

update: My post with the prediction (and mis remembered date of unleaded in the USA) is here. I would have just edited without admitting it was an update, but Mark Sadowski has already commented on the post (I thought he was more into geld than lead).

The numbers
Total Violence Against the Person Offences

1998/99 502,778
1999/00 581,034
2000/01 800,913
2001/02 850,326
2002/03 845,078
2003/04 967,228
2004/05 1,048,095
2005/06 1,059,583
2006/07 1,046,187
2007/08 961,099
2008/09 903,447
2009/10 871,712

Tags: , Comments (26) | |