Taking it to the Tweets: Statistics proves Twitter a powerful tool in forecasting crime

"My initial hypothesis was that there would be no correlation between Twitter use and crime. After all, people don't share with the world that they intend to or have just committed a crime," said Gerber. "What they do share are things like social events or outings that could lead to criminal activity." Gerber chose Twitter over other social media platforms for its openness and the fact that anyone can access GPS-tagged tweets generated in a given area.

His statistical method involved collecting more than 1.5 million public tweets tagged with Chicago-area GPS coordinates spanning January to March of 2013, as well as crime records covering the same period and geographic area. After dividing and mapping out tweets and crime records onto a grid and identifying common topics of discussion (e.g., sports, restaurants, and entertainment) appearing in tweets, Gerber combined conclusions from this analysis with older forecasting models to predict crimes over the next month. The result of his combined method was more precise, accurately predicting 19 out of 25 crime types.

"Some cities that utilize such methods as a basis for resource allocation have seen dramatic decreases in crime," said Gerber. As for the causal connection between tweets and crimes, Gerber admits his method cannot answer that. Even so, it's gaining attention from police departments all over the United States, including Chicago and New York City. His work could further assist departments in resource allocation, deciding where and when to deploy officers.

Gerber co-directs UVA's Predictive Technology Laboratory, which uses data to create predictive models with the goal of promoting better decision making. In addition to applying models to the field of policing, the lab is also conducting research in other important fields like health care and the military.