Bias built in to crime prediction

May 27, 2016

Predictive policing seems to be playing a bigger role in court decisions these days. People charged with crimes can be given a risk score based on priors and their background, which represents a fuzzy likelihood that they commit a crime again. ProPublica investigates the reliability of these scores, using data from Broward County, Flordia, between 2013 and 2014

The score proved remarkably unreliable in forecasting violent crime: Only 20 percent of the people predicted to commit violent crimes actually went on to do so.

When a full range of crimes were taken into account — including misdemeanors such as driving with an expired license — the algorithm was somewhat more accurate than a coin flip. Of those deemed likely to re-offend, 61 percent were arrested for any subsequent crimes within two years.

So tons of uncertainty.

Then, when you compare scores for blacks and whites, taking into account the types of crimes committed, there appeared to be a bias towards scoring blacks at higher risk. Troubling. I suspect this is related to the mentioned factors like employment and education.

But still, with so much uncertainty case-by-case, should these numbers — that I can only imagine appearing nearly concrete in the court room — even be part of the equation of an individual’s future?

More on the ProPublica analysis here.

Favorites

This is an American Workday, By Occupation

I simulated a day for employed Americans to see when and where they work.

Visualizing the Uncertainty in Data

Data is an abstraction, and it’s impossible to encapsulate everything it represents in real life. So there is uncertainty. Here are ways to visualize the uncertainty.

Unemployment in America, Mapped Over Time

Watch the regional changes across the country from 1990 to 2016.

The Best Data Visualization Projects of 2014

It’s always tough to pick my favorite visualization projects. Nevertheless, I gave it a go.