In a previous post, I described some cases of the use of discretion in law enforcement from Barry Lam’s book Fewer Rules, Better People. But while citing individual cases can be useful for illustrating an idea, coming to a decision on whether or when discretion should take precedence over legalism can’t be made by citing individual anecdotes. What is needed is a second-order consideration of what kinds of circumstances make it more likely for discretion to bring about better results than legalism in general, rather than pointing to individual instances of rules or discretion bringing about good or bad outcomes.

A proponent of legalism might try to suggest that we use rules informed by the most up-to-date research, counting on behavior guided by those rules to outperform discretion. Lam describes the results of one such attempt at research and how it was integrated into policy. While according to the Supreme Court, law enforcement has no obligation to provide protection or services to any citizen, there is an exception in most jurisdictions – domestic violence:

As a result of well-intentioned activism, a majority of states and jurisdictions have made the use of force legally mandatory in cases of domestic violence. Police have no discretion, for they can be arrested, jailed, or fined for failing to arrest a suspect.

How did this come to be? It started with the work of Lawrence Sherman, a criminologist in Minneapolis, who was asked to help work out the best approach for domestic violence calls during the early 1980s. These calls tended to have one of three outcomes – either the police made an arrest, or they attempted to bring the situation to a resolution with the parties, or they would separate the people involved by having one or both parties stay with family or friends for 24 hours until tempers had cooled. The question was which of these actions produced the best results. Rather stunningly, Sherman managed to convince the police department that the only way to answer this would be to carry out each of these approaches at random:

The Minneapolis police would have to do the equivalent of rolling a three-sided die each time they received a domestic violence call. The officers would have to perform their assigned intervention on the call regardless of what they learned or encountered on that call. What if it was a minor call because someone punched a wall and scared the kids? If the die read “arrest,” you had to arrest. What if the suspect gave a serious beating to an elderly parent? If the die read “don’t arrest, only separate them,” you would have to do that.

What were the results of this experiment?

After months of conducting the experiment, Sherman found that, of the people who were randomly assigned “arrest,” 10 percent of them were rearrested for domestic violence within six months. Of those who received mediation, it was about 18 percent. And for people who were separated for twenty-four hours, it was over 20 percent.

So, the data seemed to show that arresting suspects led to greatest reduction in future incidents of domestic violence. This had a swift impact:

Within a few years, twenty-eight states passed laws that made it mandatory for police officers to arrest someone in a domestic violence dispute, imposing a $1,000 fine or one-year prison term if they didn’t do so. These “shall arrest” laws even expanded to include the category of violating a restraining order (or protection order) related to domestic violence. It became widespread US policy that selective discretion applies to all crimes except domestic violence.

This had a popular political coalition to support it as well – Republicans in the Regan Era were very big on being tough on crime in general, while Democrats and feminists were keen to push for stronger police action for domestic violence in particular.

However, the long-term results have not been what the initial advocates might have hoped for:

Unfortunately, forty years of empirical data shows that there is no difference between domestic violence rates in states with mandatory verses discretionary-arrest policies. This is even though twice as many people are arrested for domestic violence in mandatory-arrest states than in discretionary-arrest states…This is partly because states with mandatory-arrest policies result in more dual arrests: officers use far less discretion trying to determine who is at fault or the primary aggressor, so they arrest everyone in the dispute. The result is that mandatory-arrest states have two to three times more people with an arrest record, filling jails and being taken out of work and family care while the domestic violence rates remain the same.

Even more surprisingly, states with mandatory-arrest policies for domestic violence tend to have worse long-term outcomes for victims of domestic violence:

Meanwhile, mandatory-arrest policies lead to significantly higher murder rates of spouses. There are 35 percent fewer murders of spouses in states with discretionary-versus-mandatory arrest policies. Even outside death by homicide, women whose partners were arrested have a much higher premature death rate than women whose partners were not.

Here it’s worth pointing out that Sherman, the aforementioned criminologist, had a much more restricted interpretation of his findings in the Minneapolis experiment than policymakers:

“So arrests worked best from the Minneapolis experiment, in that city, in that context,” concluded Sherman, and he reported these finding to the Minneapolis department.

But the context of time and place makes a huge difference as to the outcomes:

Sometimes arrest increased domestic violence recidivism, sometimes it had no effect. The issue turned out to be complicated. Whether arrested deterred future violence depended on a host of other factors local to a community, whether the community was affluent or improvised, whether the household was impoverished or solidly middle class, and others. Sherman never intended the Minneapolis study to lead to a blanket policy, and his subsequent studies revealed just how ineffective and self-defeating that policy was.

Decades of additional research has provided data on all kinds of ways particular variables are correlated with long-term outcomes of arrest or other approaches. But this more robust and detailed collection of data, Lam argues, still can’t do the work that the policy-over-discretion crowd would want from it. We might say “in a cases with X, Y, and Z variables, arrest leads to better outcomes 75% of the time.” But this does not imply that for any given XYZ case, there’s a 75% chance for arrest to bring about the best results. There might be XYZ cases where arrest is almost certainly going to make things worse. These kinds of data can inform decision-making, but they can’t themselves be what actually makes a decision in a given case – that will have to be made on the ground by a particular person. And making the best possible decision in that case will depend on the proverbial man on the spot using the full amount of information at hand relevant to that specific case – information that cannot be known or possessed by any given rule-maker:

There really are better or worse ways to do the job. And doing the job better requires knowing some background facts as well as the facts of each situation.

And this is the key mechanism Lam identifies for expanding the scope of discretion over rules for the “on-the-street” level bureaucrats and enforcers. The people directly on the scene are the ones most likely to have the best and most relevant information for each case, and will be able to use that local knowledge to greater effect than simply following a blanket policy made by a high-level policymaker operating according to statistical aggregates and detached social theory:

Real life provides a lot more information than can ever be reckoned with in the statistical norms. Maybe there are children or elderly parents in the home. Maybe there is hunger involved or a firearm. How can this information be ignored while decided whether this is one of the 75 percent or one of the 25 percent of exceptions to the rule? In circumstances like these, do we want officers to walk into a situation with discretion or a mandate?

Deciding that an officer must treat every unique circumstance, not according to the particular facts of time and place relevant to that circumstance, but according to a statistical flowchart written up in an office by a policymaker who can’t possibly have all the relevant information each situation requires, has led to great harm:

The utopian social engineer dreams that a single easy-to-follow rule, laid out in advance and executed without exception, will solve a particular social problem. The merely optimistic engineer dreams that at the very least, the rule will outperform discretion, the acts of thousands of individuals making thousands of decisions based on the thousands of micro-situations they encounter. The “shall arrest” rule was a forty-year experiment about whether the complex social problem of domestic violence admits of a blanket solution. In hindsight, to think there could have been such a one-size-fits-all solution seems naive.

But Lam’s concern with legalism goes beyond the inability of high-level rules to capture the information relevant to on-the-ground level circumstances. He also believes that legalism comes with a significant moral cost, making us into worse people. I’ll review that aspect of his argument in the next post.