Parole Judges, Bias and Noise

There’s a famous study of parole judges that find their decisions are influenced by the timing of their breaks. On average, these judges would deny about 2/3rds of all parole requests. Just after a break, the study found the judges would approve nearly 2/3rds of all requests, while in the period leading up to the break the odds of approval dropped to 0%.  What is really going on here? To understand, we need to grasp the concepts of bias and noise.

When we think of each individual case, we can ask how in-need or not-in-need of a break the judge probably feels. At the individual decision level, that feeling may bias the outcome. When we think of a full day’s worth of cases at the collective level, we can note the overall odds of denial at 65% and then consider how those odds shift throughout the day. The lumpy distribution of results is what we call noise. Sometimes bias can explain away noise, but sometimes it can’t.  Maybe needing a break changes decisions, but maybe it’s something else altogether. 

Daniel Kahneman told Tyler Cowen that “bias has been overestimated at the expense of noise.” As a Nobel Prize winner for his work on decision-making, Kahneman has made a career of categorizing our mental mistakes. Recently his work has started to focus on noise, hence his statement. He explained to Cowen that when we are looking for bias we are looking for causal relationships (this causes that). When we are looking for noise we are looking at statistical relationships (this correlates with that). We have to be very careful not to confuse the two. 

While our instinct after reading the parole judge study may be to rethink the break schedule to de-bias the uneven results, Kahneman’s new work suggests maybe not. Most of the time there’s no one variable to rule them all, and noisy data makes everything a lot more complicated. Bias suggests needing a break may cause less-fairness, but noise asks us to look for what else might be correlated. If we only focus on bias and ignore noise, we could end up doing more harm than good. 

Whether it’s in our own decision making or those of our clients, it pays to step back and see how bias and noise fit into the equation. From recognizing patterns in stocks prices to seeing religious figures on a grilled cheese sandwich, our brains love to jump to conclusions. As professionals, our job is to help people slow down and focus on using a process that their future self will be happy they employed. No judge wants to find out he did a poor job because he was hangry. Likewise, no client wants to find out they made a poor decision because they acted too impulsively and we weren’t able to help them. By learning to think about bias and noise together we can keep our humility in a complex world and focus on the small details we can control.