One thing that most of these analyses do that really bothers me is that they propose an analysis of the crime decline, but not an analysis of the previous rise. It is hard to imagine any useful analysis coming out such historical blindness. I don’t expect much of Vox, but the Brennan Center report acknowledges the history, and then discards it. Vox itself says that #3 broken-windows policing is a myopic theory, but fails to note that all of its analysis is myopic. Also, international comparisons are very important. Canada and much of Europe saw homicide double 1965-1980 and then return to baseline. America is unique in having an extra peak around 1990, the crack epidemic.

Crack (#12) is not an explanation, but the international and regional comparison shows that it is exogenous. Surely the crack epidemic has a separate explanation, layered on top of the reasons common across countries for the doubling and halving. Hypotheses that try to explain both at once should be looked upon with suspicion. One might make the same argument to ignore the American 1980 cocaine peak. By the metric of homicides, it was just as high and just as abrupt as the crack peak. But it was a uniform crime wave, while the the crack epidemic was concentrated among the young. The rise in homicide by and of youth (say, 15-25) was far faster in the crack epidemic than in the cocaine wars. Anyhow, it hardly matters whether one includes the 1980 peak because that just pulls the peak back five years, which is well within the variation of peak years in Europe.

The homicide rate today is about the same as it was in 1960-1965. The obvious guess is that the same thing caused the rise and fall. One should be suspicious of explanations that ignore this symmetry. That applies to almost all of Vox’s items, but 1, 2, 3, 7, 8, 9, 11, and 15 are particularly asymmetric. I think #5 and 6, too. #6, the economy, is what the Brennan Center gives the most weight.

Steve Sailer turns the vague #4 “improved policing” into a symmetric, but still quite vague hypothesis, that the police were hamstrung or demoralized by the Warren Court reforms, but eventually adapted to them, or were just brought out of their funk by public opinion. Maybe even “broken-windows policing” is a return to the old ways. Warren only had direct power in America, but there are vague claims that the general pattern spread throughout Europe.

But I know that the explanation cannot be perfectly symmetric. The homicide death rates for 15-25 are maybe the same as in 1968 (much higher than 1965), but for 25-35 and 35-45, they are maybe 2/3 the 1968 rates. So the overall average is like 1965, but the distribution is different. That’s pretty weird and I don’t know what it means. Probably we need at least two causes, one increasing crime among the youth (gangs?) and one decreasing overall. Relevant graphs, 1 2 3, discussed above.

And those demographic trends argue against many explanations. In particular, the lead #16 and abortion #15 theories prediction exactly the opposite of what happened, so I thoroughly reject them.

I think that #10 Alcohol consumption does have the right symmetry, but the changes are small on the historical scale.

If all you want to do is explain the crack epidemic, #12 crack and #13 “gangs have gotten less violent” are symmetric. They are definitely the answer, in some sense, but they aren’t much of an answer. If you are interested in a larger time frame, crack is irrelevant, but gangs might be. It would be useful to know if gang violence has declined more or less than other violence from my baseline of 1980. But even if I did know, that would only be a hint, not an answer. Compared to 1960, surely gangs are more violent today. So I endorse the opposite of #13 for the long term.

Age demographics #14, is the first theory I whole-heartedly endorse. It appears to me to explain 20% of the rise and 20% of the fall. The Brennan Center report calls it 5% during the 90s and nothing during the 00s. Partly that is because of the demographics of the crack epidemic, but even considering that, I do not see how they can get such a small number. It would be useful to do a comparison with Germany, which had a baby boom much later than the winners.

Going back to #1, incarceration, I reject this theory because it is asymmetric, doesn’t explain any other country, and doesn’t really match the timing. But the one thing going for it is that it that the raw incapacitation version of the theory does match the changing demographics of crime. (Indeed, it is the only theory that does, except for the opposite of #13 gangs). People in prison are systematically older than prisoners, explaining why crime has fallen so much more among the older than the younger. More likely, other policing reforms cause more and better arrests that train people out of crime. But such theories are vague and hard to test.

Report comment

People already have plenty of reasons to reject the individual hypotheses. Giving them a new one to reject to joint hypothesis on the grounds that it is multifactorial fails to address the basic problem that people are failing to reject at all on the grounds that they already have.

Report comment

It looks like about 20% of the run up and the run down is due to a changing age profile. The normal homicide rate has been cut by 55% since the twin peaks of the cocaine wars and the crack epidemic. But the red line has only been cut by 45%.

The numbers are from CDC’s Wonder database. It only goes back to 1968, while I’d really like to go back to 1965 or 1960. I’m using it to get demographic data, which ought to be available elsewhere. The database is mainly for cause of death, which suggests other graphs. Here is a graph of homicide death rates for the biggest groups. As in the graph in the other comment, homicide victimization rates for a demographic are similar to, but not quite the same as the rate of committing homicide. Younger groups are being killed at the same rate as in 1968, while older groups are being killed much less often. In particular, in 1968, people aged 35-45 were almost twice as likely to be killed than those aged 15-20, while in 1995 it was the opposite. Today they’re at parity.

Report comment

Report comment

I have churned away at training and test data myself. It takes the tiniest amounts of ‘misconduct’ to get the garbage generator going,

Huh! I once recall spending 6 months trying to get some modelling results out for a project and failing utterly.
Admittedly I only really had 12 data points for the LHS.

Report comment

Unfortunately, it turned out that large segments of the American population don’t react to welfare the way Swedes do, instead displaying an almost immediate rise in crime.

Seems a little like begging the question in a thread about what causes increases and decreases in crime.

Report comment

Report comment

Looking at the 2000 survey figures for total victimization rates, we have:

“- Above 24% (victim of any crime in 1999): Australia, England and Wales, the Netherlands and Sweden
– 20%-24%: Canada, Scotland, Denmark, Poland, Belgium, France, and USA
– Under 20%: Finland, Catalonia (Spain), Switzerland, Portugal, Japan and Northern Ireland. ”

“The risk of having a car stolen was highest in England and Wales (2.6% of owners had a theft), Australia (2.1%), and France (1.9%). Japan, Switzerland, Catalonia, the USA, Finland, and the Netherlands show risks of 0.5% or less. ”

http://www.unicri.it/services/library_documentation/publications/icvs/publications/index.htm

Report comment

Report comment

Report comment