Back in July, the Planet Money Podcast had an interesting episode about using big data to change how bail is set.
[The researchers] create something called a risk assessment algorithm. This algorithm basically predicts how likely somebody is to show up to court and how likely they are to commit a crime. The algorithm just looks at nine factors:1
- Does the defendant have a prior conviction for a crime of violence?
- Has that conviction occurred within the last X number of years?
- Has the defendant had a failure to appear, or bench warrants issued in prior cases?
- How old was the defendant at the time the crime was committed?
What I like the most about this story is the juxtaposition of the complex and the simple. Millions of cases are ultimately distilled into a small set of understandable heuristics. It’s the type of information processing that human brains excel at, just hopefully with a reduced susceptibility to cognitive biases.
They only read four. ↩