Jimmy Li over at the worthwhile Traiberman-Li blog has this discussion on decisionmaking, from which I excerpt this section:
In one of their studies, K&E looked at the bail decisions of five San Diego judges (each of whom did not know they participating in a study when they made their decisions). As K&E mention, most states have explicit guidelines that judges are supposed to follow in making their bail decisions (a “bail decision” is a decision on whether to grant bail, and at what price). At the time of the study, for example, California’s guidelines emphasized factors like “dangerousness,” “risk of non-appearance,” and “community ties of the defendant.” To what extent did factors like these actually influence judges’ decisions?
To answer this question, K&E developed and tested many decision models, each of which took into account different factors, and to varying degrees. A simple model, for example, might make a decision based purely on the defendant’s criminal history (e.g. “If any criminal history, no bail”), while a more complex model might include four or more differentially-weighted factors.
When K&E compared their models’ predictions to the judges’ actual decisions, they found that judges’ behavior was “characterized by a remarkable (almost offensive?) simplicity.” In setting bail, for example, judges’ decisions could be fully predicted by just two factors: the recommendation of the prosecuting attorney and the recommendation of the defense attorney (with much more weight granted to the former). The two attorneys, in turn, based their recommendations off of just one factor: the severity of the crime.
This means that when it comes to setting bail, many of the factors that the California guidelines enumerated and all of the factors that the judges (when interviewed) claimed to consider didn’t make into judges’ actual decision functions. Instead, they took into account just one factor, and they did so indirectly, through their reliance on the prosecution and defense.