“Murder Numbers Decrease” is the headline in today’s Austin-American Statesman. Public safety reporter Ciara O’Rourke reports on the Austin Police Department’s homicide data released for 2013. Her report indicates there were 24 homicides in Austin in 2013 and that was the least amount in the last three years. Has the homicide rate in Austin really gone down?
Today’s headline reminds me of Temple University professor of mathematics John Allen Paulos’ 1995 book: A Mathematician Reads the Newspaper in which he takes stories like today’s piece and profiles the issues with the way data are reported by journalists who don’t understand statistics. Following Dr. Paulos’ approach, let’s take a look at the Austin Homicide data reported and see if we agree with the story.
Let’s begin with the description that there are fewer murders than the previous three years. The table below shows the number of homicides each year from 2010-2013. The story only included data in the text of the story; the table was created by me.
|Table – City of Austin Homicides 2010-2013|
|Year||Number of Homicides|
Looking at the last three years, 2013 did have fewer homicides. Does that mean anything? We really don’t know. This is an example of summary statistics presented in a format that doesn’t give us enough data to interpret nor is it displayed in a way that helps us to appreciate change. From the headline, you would think there is improvement, but in the article, “police officials” recommend “caution against conclusions from the city’s homicide statistics in any given year, because the numbers fluctuate.” In other words, APD understands a little bit about variation.
In their annual Crime and Traffic Report, APD does two things that did not occur in the Statesman story.
(1) APD reports the data using the industry standard rate per 100,000 population. This is important because according to the City of Austin Demographer, the US Census data shows an Austin population growth of more than 2% in all but one of the last seven years, which means annual jumps of approximately 20,000 residents. So, 2008 – which also had 24 homicides, but 92,225 fewer residents – had a murder rate of 3.19 versus 2013 with a rate of 2.84
(2) APD presents the data over time, in a times series chart. The line chart below comes from the 2012 Crime and Traffic Report. It does not include 2013.
Looking at this chart tells us a lot more. Considering 2013’s homicide rate is 2.84 per 100,000, we might conclude it’s one of the lowest rates for a year in recent record. That said, if we look at the homicide rate over the last decade, we might also conclude that the rate is vary similar and within normal variation. In other words, while this year’s rate is lower than the last three years, it is not out of the norm and not really a decrease in the overall homicide rate. As you will see in a minute, a rate of 1.38-6.27 homicides per 100,000 would be completely predictable.
The decreased homicide rate in 2013 may not be newsworthy of a front-page story, but looking at Austin’s homicide rate using all data available in the FBI Crime Statistics database dating back to 1985 does show something interesting. See the Statistical Process Control (SPC) Chart below.
From 1985 through the mid-90s, the homicide rate was a mean of 8.8 and was more variable than today. Then something changed toward the end of the 90s that reduced the homicide rate, significantly reduced the variation, and has sustained to today. What was that change? I don’t know. Could it have been a change in law enforcement activity, a difference in the availability of types of weapons, a shift in drug activity, or maybe it was the transformation of the EMS system that treats trauma patients, which experienced two major system evolutions at that time. Or maybe it was a combination of things. I do not know, but something did change in the direction of goodness and in spite of rapid growth, Austin continues to sustain a low rate of homicides and that’s good too.
So, next time you read a story in the paper that discusses data and says something is better or worse comparing just a few data points, remember Dr. Edwards Deming’s advice: “When you have two data points, it is very likely that one will be different from the other.”