Crime Score as Weapon of Math Destruction

What is Walk Score/Crime Grade?

In the past 10 years Walk Score has gone from a cool GIS startup to a popular Redfin widget, helping people plan their moves and real estate purchases based on a neighborhood’s perceived walkability. Recently Walk Score has begun beta testing Crime Grade, a measure of crime risk near an address.

How does it work?

Walk Score uses a patented system to score locations by amenity access and pedestrian friendliness including population density, block length and intersection density (Walk Score Methodology). Crime Grade‘s letter grade is calculated using a location’s per capital crime rate ranked with other rates around the city.

Why is it problematic?

Jeff Speck’s General Theory of Walkability states:

“…to be favored, a walk has to satisfy four main conditions: it must be useful, safe, comfortable, and interesting.”

While Walk Score does a pretty good job of measuring useful walks with distance to amenities, it is less effective at measuring a walk’s safety, much less the more subjective aspects of comfort and interest.

Crime Grade is scored based on reported crime: crimes reported to the local police. This is potentially very different from the measure most important to potential walkers; perceived safety. While it is useful to know what crimes are reported to the police, what keeps walkers from a neighborhood is really their perception of what crimes are committed there. Like many crime statistics, measures of perceived crime and disorder are very connected to the racial and economic demographics of the neighborhood (Sampson and Raudenbush 2004). In fact, the most powerful predictor of perceived disorder is neighborhood demographics, stronger than observed disorder and even reported crime (ibid). Crime Grade has the potential to become another in a long line of Weapons of Math Destruction, algorithms that hard code our implicit biases about race and class with potentially drastic consequences.

Potential Solutions

My research attempts to tease out some of the racial, gender and class dynamics in perceived walkability. Using data collected about people’s perceptions of sidewalk interactions I intend to analyze racial/gender and class differences.