We read the research so you don’t have to.
First, thanks to all who sent in suggestions of possible studies to include in this edition of What We’re Reading. A word to the wise: We are much more likely to include your research if it is not locked behind a paywall. Why hasn’t someone figured out a way to make academic research more broadly available on the internet? As Yosemite Sam used to say, there’s gold in them thar hills – at Vital City, we believe that a great deal of important knowledge never makes it out of academia and into the hands of those who could turn these ideas into action. Expensive paywalls are a big part of the problem. People have been advocating for “open access” for decades now, without a lot to show for it. If you are in the academic publishing game, please feel free to disabuse us of our naivete. Or better yet — please send us the necessary passwords so that we can access all of your journals without paying a king’s ransom.
On to this winter’s edition…
As always, these are quick takes on recent research. The usual caveats apply: We are not social scientists and we are not licensed to practice social science in any state. We are not checking math or methodology, but we are reading the studies so you don’t have to.
One of our themes this time out is discretion — always a vexed topic. In recent years, as racial disparities in the criminal justice system have garnered increasing attention, many reformers have worked to limit the discretion of police officers, prosecutors and judges on the theory that if frontline practitioners are left to their own devices, they will make decisions that betray either explicit or implicit bias.
While promoting better results is a worthy goal, of course, it is impossible to eliminate discretion from any system designed and operated by humans. And would we even want to? People will always be confronted with idiosyncratic situations and complicated fact patterns that do not conform to preexisting rules and regulations. Arguably, the only way to achieve good outcomes in such cases is for those with discretion to exercise good judgment.
In a working paper for the National Bureau of Economic Research, economists Felipe M. Gonçalves and Steven Mello look at traffic enforcement in Florida, where over one-third of all speeding citations are issued for exactly 9 miles per hour over the posted limit — just below a $75 increase in the associated fine. According to Gonçalves and Mello, “The dramatic bunching in the speed distribution suggests systematic manipulation by officers. Specifically, the distribution implies the practice of speed discounting, where officers observe drivers traveling at higher speeds but write down nine MPH on the citation as a form of lenience.”
What accounts for this beneficence? Gonçalves and Mello believe the answer is simple: a distaste for traffic court. Drivers who receive higher fines are more likely to contest their tickets, which would require the ticketing officer to appear in court.
Gonçalves and Mello document that, in the case of Florida traffic safety, discretion comes with a cost: Harsher fines reduce a motorist’s future traffic offending and likelihood of crash involvement, although the size of the effect isn’t large (about 1.6 percentage points). They suggest that if officer discretion were eliminated and all offending drivers received the harsher penalties, public safety would be improved, although again the effect size isn’t enormous (about 2%).
The Bottom Line: “Increasing the harshness rate for the average driver improves safety, and allocating harsh fines to motorists who typically receive lenience when officers use discretion further reduces the reoffending rate, because these motorists are especially responsive to harsh fines. Note that this is exactly the opposite of what would be expected under safety maximization by officers, which would predict the smallest treatment effects for the motorists typically issued lenient fines.”
In another working paper for the National Bureau of Economic Research, Victoria Angelova, Will S. Dobbie and Crystal Yang (all from Harvard) present findings from a quasi-experimental study examining judicial decision-making in a large mid-Atlantic city that employs a pretrial risk assessment to predict the likelihood of defendant misconduct and make recommendations to judges about whether to release or detain defendants. As Angelova, Dobbie and Yang report, judges frequently ignore the recommendations made by the risk-assessment instrument, providing an opportunity to assess the performance of judges against the instrument as well as against each other.
The findings confirm what has become the conventional wisdom in the field: Algorithms outperform human judges. While this is bad news for those who seek to resist the inexorable rise of our robot overlords, the fine print reveals at least one kernel of good news for human decision-makers. The researchers found that 1 out of 10 judges outperformed the algorithm when they made discretionary overrides. As the authors dryly observe, this suggests that “a human can still add value to the decision-making process.”
What distinguishes the high-performing judges from their peers? According to the authors, it is largely a matter of being able to distinguish signal from noise. In the case of bail decisions, the “noise” includes factors that are not predictive of risk (such as whether a defendant has an out-of-state license) and that are irrelevant to the case at hand (such as being influenced by having heard a case in which a different defendant was arrested for a serious violent offense while on pretrial release). Perhaps most tellingly, the low-performing judges in the study placed far more importance on demographic factors such as race, while the high-performing judges placed greater emphasis on “non-demographic factors such as mental health, substance abuse and financial resources.”
The Bottom Line: “We find that the judges in our setting underperform the algorithm when they make discretionary overrides, increasing pretrial misconduct by an average of 2.4 percentage points at the judges’ existing release rates (a 15% increase from the mean). This finding indicates that the typical judge in our setting is less skilled at predicting misconduct than the algorithm and that we could substantially decrease misconduct rates by automating release decisions. But this average impact masks substantial variation in the judges’ performance compared to the algorithm. The negative average impact of human discretion on the accuracy of decisions is explained by the 90% of judges who underperform the algorithm when they make a discretionary override. In fact, nearly 70% of the judges make override decisions that are no better than random — that is, they could achieve a lower pretrial misconduct rate by flipping a coin or using a random number generator. At the same time, we also find that 10% of the judges outperform the algorithm.”
How do we get people to exercise their discretion to make good choices? In recent years, some behavioral economists have suggested that subtle “nudges” can help make a difference. For example, placing fruit near the cash register where customers make impulse purchases might help reduce the consumption of unhealthy snacks.
While nudge thinking certainly has its skeptics, there are still many researchers interested in testing the application of these ideas. One recent effort was led by Alex Chohlas-Wood, the executive director of the Stanford Computational Policy Lab, who with a team of researchers conducted a randomized controlled trial with 5,706 clients of the Santa Clara County Public Defender Office.
Half of the group received a series of automated reminders before their assigned court dates and half did not. Reminders were sent seven days, three days, and the day before a required appearance. In the control group, 12.1% of clients received a bench warrant at their first scheduled court date, compared to 9.7% for clients in the treatment group. While this was a relatively modest effect size, it still counts as a victory for the idea that subtle nudging can help achieve better outcomes.
The Bottom Line: “As states attempt to reduce the number of people they incarcerate, many are looking to court reminders as a way to increase court appearances and reduce jail time. With an average marginal cost of roughly 60¢ per defendant per case, our results suggest that a text message reminder program can be an effective and relatively inexpensive way to increase appearances and decrease incarceration.”
This job market paper from Rutgers economist Xichen Kong, which was promoted on Twitter/X by Arnold Ventures’ (and Vital City contributor) Jennifer Doleac, seeks to add to the growing literature about the impact of marijuana dispensaries on crime, a field that is evolving rapidly as cannabis legalization advances in the U.S.
Using publicly available data from Chicago, Kong finds that the establishment of a recreational marijuana dispensary leads to a 30% increase in crime within a 500-meter radius of the facility. Kong theorizes that this finding may confirm both the broken windows theory and the routine activity theory, which suggests that crime is the product of motivated offenders and suitable targets converging at a location where there are no “eyes on the street.” That is, marijuana dispensaries might increase local disorder and attract both would-be offenders and potential victims.
The Bottom Line: “For both property and violent crimes, we observe an increase in incidents between 200 meters and 300 meters from a marijuana dispensary, followed by a decrease around 600 meters. This pattern suggests that recreational marijuana dispensaries tend to attract criminal activity from areas farther away toward locations in closer proximity to the dispensary.”
In this paper, Sam Peltzman, an emeritus professor at the University of Chicago and an editor at the Journal of Law and Economics, takes a deep dive into the General Social Survey, which, since 1972, has asked American adults this basic question: “Taken all together, how would you say things are these days — would you say that you are very happy, pretty happy, or not too happy?”
Peltzman teases out the following:
- Gender: “For the post-2000 years males and females have been about equally happy.”
- Age: “Happiness is flat until the late 50s, increases until the late 70s and then falls part way back.”
- Marital Status: “Marital status is and has been a very important marker for happiness…the married population is over 30 points happier than the unmarried.”
- Race: “Whites are around 15 points happier than Blacks. But the gap has been narrowing steadily — from around 25 points in the early 1970s.”
- Education: “College grads have been 9 or 10 points happier than high school dropouts and 5 or 6 points happier than high school grads since the 1970s.”
- Income: “The happiness difference between the 75th and 25th income percentiles exceeds 20 points. Moreover, these differences have been widening.”
- Geography: “Cities remain the least happy places, but they have entirely escaped the post-2000 national decline in happiness.”
- Politics: “Conservatives are around 9 points happier than liberals and 7 points happier than moderates overall.”
- Trust: “People who trust others or the federal government are distinctly (around 20 points) happier than the more wary; However, trust, especially in the government, has declined substantially over time.”
The Bottom Line: “Overall, the population is reasonably happy even after a mild recent decline.”
As always, please send comments and suggestions to firstname.lastname@example.org.