## Thursday, September 3, 2015

### Are Murder Rates Spiking?

Recently the New York Times wrote on the subject of crime statistics. The article, "Murder Rates Rising Sharply in Many U.S. Cities," was backed up by this table of numbers:

I've seen a lot of pundits adding their two cents to the Times story, but nobody seems to be criticizing the awful data analysis in the article.

First problem: The title of the Times's graph says "Rising Murder Rates," but none of the numbers in the graph are rates (per capita). Of course, if the cities' populations were fairly constant over the past year, then the bar graph would still be fairly accurate. Nevertheless, you shouldn't say "Murder Rates" and then not chart murder rates.

The much larger problem is that the data shown here in no way demonstrate the article's thesis. For one thing, what is Philadelphia doing in this list? A four percent increase is just noise. Contrary to the article's thrust, the data for Philadelphia show convincingly that there is no spike in the murder rate, at least for that large city.

As one commenter on the Times's website said,
Math. The NYT should learn how to apply it. Crime rates have been historically low the last few years. They have been declining for the better part of a quarter century. Could this be the start of a new trend? Or is it a blip from previous multi generation lows? No one knows for sure but headlines like this are misleading clickbait.
To justify a story like the one the Times wrote, what you want to see is a line plot showing the murder rate in each city over the past five, ten, or twenty years. I made such a line plot for New Orleans, using data from the City of New Orleans Police Department(The numerical values are listed further below in an Appendix; on the plot itself, I have omitted the vertical scale in order to focus your attention on the pattern of fluctuations.)

Seen in context, the blip for 2015 is more noise than signal. So, while the Times article might well contain some interesting sociological analysis, what the article doesn't do is explain the data for New Orleans—because there is nothing in the New Orleans data to explain.

Maybe if you repeated this exercise for Milwaukee instead of New Orleans, the results would be strikingly different. Even if that were so, it wouldn't change the fact that New Orleans doesn't belong in the list (to say nothing of Philadelphia). In any case, a historical line plot—not the graph the Times presents—is the kind of picture you should be looking at in order to grasp a situation like this. Why didn't the Times do that analysis to begin with?

Appendix: Data for the line plot

I did this quickly, let me know if there are errors.

Murders by Year (sources: 1, 2, 3, 4, 5)

2010     175
2011     199
2012     193
2013     156
2014     150
2015*   189

*The 2015 value is the Times value, 120, multiplied by 365/232. The end-date range noted in the Times graphic is 8/11–8/31, the average of which, 8/21, is 232 days into the calendar year.

Population by Year (source)

2010     1,189,866
2011     1,214,235
2012     1,228,375
2013     1,241,949
2014     1,251,849
2015*   1,270,759

*The linked source gives populations for 2010–2014. The population for 2015 is the value obtained from a linear fit to the preceding five years (r^2 = 0.97).

Murder Rate by Year (by dividing)

2010     14.7
2011     16.4
2012     15.7
2013     12.6
2014     12.0
2015     14.9