In my public policy course this week, we used a Harvard Kennedy School case entitled “A Rising Storm: Eric Garner and the Explosive Controversy over Race & Policing,” which introduced questions about criminal justice, violence, and race in New York City.

In my own work based on national survey data, I find that being Black raises the odds of being mistreated by the police almost five-fold (4.6 times) when considering gender, education, age, English-language proficiency, household income, housing type, county-level income, and any mental health diagnosis. However, the Kennedy School case also draws our attention to the 75% drop in homicide and 50% drop in rape in New York City between 1990 and 2013, which require explanation and assessment.

I suggested that we all use mental models (simplified representations of the world) to think about such issues. Our models may be more or less precise and articulated in our own heads. If we want to know what’s best to do, we are obliged to clarify our models to ourselves and seriously consider whether alternatives might be wiser.

I asked students to consider models of different forms and types:

• A root-cause model explains many phenomena as the result of one underlying cause. Just as you should expect a weed to return unless you pull it up by its roots, so you should expect a social injustice to recur unless you remove its roots–that is the implication of the metaphor. “Radical” comes from the Latin for “root,” and proponents of root-cause models often feel that they are the radicals.

In this case, the root might be white supremacy, invoked to explain police bias, the distribution of wealth and poverty, the unequal impact of public schools, and other current realities.

Classical Marxism recommends a different root-cause model: the economic system always gives rise to the state, and a bourgeois state employs police to protect the economic elite. Substantial reform is impossible without a revolution. Finally–as a student astutely noted–some libertarians might offer a root-cause model in which the underlying problem is state power, always fundamentally violent (regardless of the economy); police violence is a predictable manifestation.

• A cyclical model is built of components that affect each other, often reciprocally or in loops. For instance, perhaps racial discrimination worsens poverty, poverty increases the prevalence of crime, and crime exacerbates poverty by raising costs for victims and perpetrators and by discouraging investment. Those elements might be only a few parts of a more elaborate cyclical model.

This kind of model resists the notion of a “root” and instead encourages us to “break the cycle” by acting on vulnerable links. For example, reforming police union contracts would not address white supremacy, but it might break a specific cycle that involves impunity for violent officers, and that change could have positive effects across a connected system.

• An organizational model would treat the NYPD as an entity with a mission, budget, personnel, and outcomes. We might presume that by changing the organization, we can get different outcomes. For instance, it may matter what the NYPD measures when it assesses its employees. Should officers be promoted for making many arrests for minor offenses, or not?

We might consider two variants of an organizational model:

• In one, the NYPD is fundamentally a bureaucracy, per Max Weber. It is made of people who have clear responsibilities within a hierarchy. Bureaucracies are supposed to make reliable, predictable decisions. However, some degree of discretion is inevitable, and bureaucracies strive to handle human choices by either (a) minimizing discretion or (b) ensuring that bureaucrats are as professional as possible. Professionalism means competence and trustworthiness. Per Michael Lipsky (1969), police are “street-level bureaucrats,” faced with constant discretionary decisions. If data show that actual police officers’ choices are biased or otherwise detrimental, then they should either (a) lose discretion or (b) become more professional as a result of better hiring and training.
• In another variant of the organizational model, the NYPD is a public agency. The people of New York vote for elected leaders, who appoint senior police officers as their agents. If we object to the outcomes, then (a) a majority of voters have the wrong beliefs or values, or (b) the electoral system is flawed so that elected leaders don’t represent the people, or (c) those leaders’ will is being frustrated by their agents.

• In a genealogical model, the NYPD and related institutions (such as the New York City Public Schools) derive from predecessors. Like you and me, these organizations have ancestors that are responsible for much that’s true about them today. Among the NYPD’s predecessors were slave patrols that arose in 19th century America to prevent enslaved people from escaping to freedom. However, institutions typically have many ancestors, not just one, and the NYPD could also be traced back to village constables in England or to law enforcement bureaucracies in 19th century France and Germany. (After all, the word “police” is French.) The point of a genealogical model is to uncover historical causes that may require recompense, reparation, and repair.

• In a behavioral model, you might think of human beings as a species that has proclivities to violence (including sexual violence) as well as tendencies to cooperation and care. You might think that mass societies with high degrees of anonymity will permit violence unless it is surveilled and deterred. Relatedly, you might think of peace and social order as collective goods that pose dilemmas at large scales. (Why should individuals sacrifice to protect strangers against violence?) In that case, police departments might represent solutions to a problem of collective action. This analysis is not necessarily conservative–in the sense of protective of the status quo–because the 30,000 armed police officers of the NYPD represent at least an implicit source of violence. A behavioral model might suggest that the police also need surveillance and deterrence. And we might consider alternative ways of achieving the collective good of peace, without armed officers.

• In an interest group model, the population of New York City is configured into many organized groups, although some people (such as unlicensed street vendors like Eric Garner) may not have effective organizations. Groups gain power from numbers and/or money. Among the most relevant groups in this case are the police unions, civil rights organizations, the city’s Democratic Party and specific political campaigns, and business interests. The reality on the streets is the result of competition and negotiation among interest groups. The best way to change outcomes is to form or strengthen groups that reflect the interests that concern you.

Different types of models can certainly be merged. That said, a model should not be excessively complicated, because the point is to enable wise judgement. A huge page of symbols and arrows will not yield clarity.

Also, there is a risk of letting our initial assumptions drive everything, so that we go looking for any components that confirm what we already thought. (A genealogical argument here, a bit of root-cause rhetoric, a specific proposal for breaking a vicious cycle ….) I think we are more likely to learn something new by following the logic of a particular model to its conclusion and then seriously considering alternatives to it.

See also: social education as learning to improve models; making our models explicit; police discrimination, race, and community poverty; the political economy of policing; professionals as grizzled veterans or as reflective learners; what must we believe?; and Complexities of Civic Life.