Tweets and punishment: UC study finds social media better than U.S. census at predicting crime

What’s a better predictor of petty crime than population figures?

Tweets.

University of Cincinnati geography researchers found that Twitter was a better gauge of where petty crimes were likely to occur than population data from the U.S. Census Bureau.

The reason? Census figures show only where people reside while Twitter can tell you where people are spending their waking hours, lead author and UC graduate assistant Minxuan Lan said.

“Crimes like theft are more likely to be related to where people really are, not where they live,” he said.

Researchers call this the “ambient population.” And since crime rates are typically calculated using census figures, social media could help law enforcement understand why crimes are happening in less populated areas, Lan said.

The study was published in the journal Sustainability.

The project demonstrates UC's commitment to research as part of its strategic direction called Next Lives Here.

Students in UC’s College of Arts and Sciences examined a year’s worth of theft statistics along with public social media posts. They weren’t interested in social media content, just the geotagged locations.

They applied a mathematical model called negative binomial regression to study the ambient population derived from tweets as well as its “spillover effect” on crimes of theft. The spillover effect detects information on Twitter users’ mobility embedded in their tweets.

They found that tweets outperformed census numbers in analyzing spatial theft patterns. Tweets showed a significant spillover effect on thefts, unlike the census figures.

“It may apply to any city,” Lan said. “We use public geotagged tweets to measure the dynamic population distribution. We found a correlation between tweets derived from the ambient population and theft.”

The results are consistent with previous geography research that examined correlations between social media and crime. For example, researchers at the University of Tennessee Knoxville found a correlation between petty crime and social media check-ins on Facebook and tweets on days when the Tennessee Volunteers played home football games.

UC geography professor Lin Liu said data from social media has its limitations. Unlike census figures which attempt to account for everyone, Twitter isn’t representative of the entire population. Users are both younger and more likely to live in urban areas than the general population.

“I go to the extreme of saying all social media data are biased. Bias related to age is the biggest issue. Young college-age people take the lion’s share,” he said.

Likewise, just 4% of Twitter users share their locations when posting, which presents additional limitations. And users often share their locations selectively, the study found.

UC study finds tipping point in deforestation

Geography study finds hot days lead to wildfires

Environment underlying cause of refugee crisis, UC professor says

New UC map shows why people flee

UC map identifies Ohio's opiate hot zones