Crime rate down, bad graphs and misleading headlines up

Freakonomics links to a lovely assortment of media reactions to the latest preliminary crime statistics released by the FBI. All the stories noted are overreactions, but the worst, by far, is NPR : "U.S. Violent Crime Rises at Pace Unseen in 10 Years."

Before examining the flaws in the NPR report, here is a chart of the U.S. violent crime rate since 1985, using data culled from the the 2004 FBI UCR report, extrapolating the figures for 2005 from the most recent report (which showed only the percentage of change, not the underlying numbers).

We easily see that violent crime has fallen drastically since 1993, and that the current rate, about 4.7 violent crimes per person, is barely over half what it was at its 1991 peak of 7.6. The 2.5% "jump" heralded by NPR represents a change from 4.65 violent crimes per 1,000 people to 4.77.

So there we have the actual data, presented accurately in light of long term trends. Now let's look at how NPR charts the data.

This chart tells us that the violent crime increase from 2004-2005 is as big as the entire east coast of the United States! Edward Tufte has written extensively about chart junk and lie factors in statistics. Graphical irrelevancies, such as the background image of the United States here, harm the reader's comprehension of the data being presented. The scale of the rate change suggested by the map supports NPR's hysterical (though technically accurate) headline, and leads the reader to (mistakenly) believe that we are suddenly in the midst of a vast and unprecedented crime spree which threatens to wipe out the many gains made in fighting crime over the past 30 years.

NPR now breaks down the FBI data based on increases within different-sized cities. It appears that moderately large cities have quite the crime wave! Their 8.9% increase stretches from Pocatello, Idaho to Blandford, Massachusetts. Yet the actual data (The FBI's release on Monday did not include the actual data, only the trend numbers. I am extrapolating the actual data from the 2004 final UCR report) shows that this apparently gargantuan increase really means that the number of violent crimes reported in these cities (which in 2004 had a total population of 14,642,880) increased from 135,783 in 2004 to an estimated 147,867 in 2005. It's 4,092 kilometers from Idaho Falls to Bradford. On NPR's graph, each kilometer on the map represents 2.9 new violent crimes. Hardly an explosion. Extrapolated to a daily average per each of the 22 cities in that category, we see an increase from 16.9 violent crimes per city per day to 18.4.

The good news is that crime remains down considerably over the peaks in the late 1970s, which lasted into the mid 1990s. The very small increase reported from 2004 to 2005 justifies experts exploring for causes of the increase, which could range from more comprehensive data reporting to causes more worrisome. But the data does not justify alarmist reporting suggesting that a new crime wave has broken out. Part of the fault lies with the FBI, which should have provided greater context for the new numbers. But NPR and other news sources are supposed to provide the context and more in-depth analysis to help us understand official numbers. Here, they failed.