Subjects and lessons to explore.

Law

Criminal Justice & Criminology · Policing

Police Discretion

13 min read 5 practice questions
10 key termsSourcesReport an error
Want it in plain words first? Jump to Eli explains — the same idea, no jargon.
On this page 9 sections
  1. In 30 seconds
  2. Why this matters
  3. The college version
  4. Eli explains
  5. Worked example
  6. Key takeaway
  7. Quick check
  8. Study tools
  9. Sources & references

In 30 seconds

is an officer's authority to choose among lawful options: to stop a car or let it go, to warn or cite, to arrest or release, to charge someone or route them toward treatment. It exists because criminal codes reach far more conduct than any agency can process, budgets are finite, and statutes need judgment to apply to real situations. The live question in research and policy is not whether officers have discretion but how it is guided, recorded, and reviewed.

Why this matters

Discretion is where the criminal justice system actually begins. Officers decide which of the many technically enforceable events become cases at all, and every later actor works from that filtered set. Understanding how discretion operates lets you read stop and search statistics carefully: neither waving away documented disparities nor treating any gap as settled proof of discrimination. It also explains why most reform proposals target written policy, supervision, data review, and oversight rather than trying to abolish judgment, which cannot be done. If you can tell a legal limit from a policy limit from an empirical finding, this literature becomes readable rather than a shouting match.

The college version

Discretion is built into enforcement, not bolted onto it

Start with a counting problem. Every jurisdiction's criminal code and traffic code together describe far more conduct than any police agency has the officers, hours, or budget to act on. Writing the law is cheap; enforcing it is not. Joseph Goldstein's 1960 article in the Yale Law Journal gave this situation a useful geometry. Total enforcement would mean acting on everything the substantive criminal law describes. Police can never reach that perimeter, because constitutional rules, statutes, and court decisions carve out a no-enforcement area in which officers have no authority to invoke the criminal process at all. What remains is the zone: the conduct police are legally permitted to pursue. Goldstein's point was that even full enforcement is not a realistic expectation. Definitions of offenses and of procedural limits are ambiguous at the edges, and the day-to-day pressures on an agency make complete enforcement of the remainder impossible. So officers select. That selection is police discretion, and it is not an escape hatch from the law; it is the ordinary condition of applying law to a world that produces more violations than the system can absorb. This lesson describes arrangements in the United States, and it is educational material rather than legal advice. Enforcement authority, agency policy, and oversight structures vary considerably from state to state and department to department.

Why the widest routine discretion sits at the bottom of the organization

Most bureaucracies concentrate discretionary authority at the top. Policing inverts that picture for routine decisions. A patrol officer working alone decides, in seconds and usually without a supervisor present, whether an encounter happens at all. Goldstein emphasized that decisions not to invoke the criminal process are of extremely low visibility: a warning issued and forgotten, a car not pulled over, a dispute settled on a sidewalk. Nothing is filed, so nothing can be reviewed. Decisions to invoke the process generate paper and therefore generate scrutiny from prosecutors, judges, and defense counsel. Non-decisions generate neither, which is why they escape the checks the rest of the system relies on. Michael Lipsky generalized this pattern in his 1980 book on , later reissued in an expanded thirtieth anniversary edition. Frontline public servants who deal directly with the public, he argued, hold substantial discretionary authority while carrying heavy caseloads, working toward ambiguous agency goals, and operating with inadequate resources; they cope by rationing what they provide and by routinizing cases into familiar types. The routines they build become, in practice, what the agency does. Read that way, the officer on the street is not deviating from policy when she exercises judgment. She is producing the agency's real policy, one encounter at a time.

Where discretion operates: the decision points

Discretion is easier to study when you break an encounter into the choices it contains. Whether to initiate contact at all. Whether to extend the contact into a search. Whether to close it with no action, a verbal or written warning, a citation, or an arrest. Whether to use force, and how much. Whether to route a person toward services rather than charges. Each choice is separately measurable, and the outcomes differ. In the Bureau of Justice Statistics' 2022 Police-Public Contact Survey, among residents age sixteen or older whose most recent police contact was as a driver in a traffic stop, about 9 percent reported no enforcement action, 44 percent a warning, 43 percent a ticket, and 4 percent a search or arrest. Those are self-reported figures from a nationally representative supplement to the National Crime Victimization Survey, not agency records, and the distinction matters when comparing them to administrative data. The referral decision is newer as a formal option. A 2026 conjoint survey experiment published in Health and Justice presented 329 officers in five agencies running pre-booking diversion programs with forced choices about whom to send to treatment instead of jail; the officers' choices tracked legal characteristics such as criminal history severity, drug quantity, and whether another offense had occurred, and also varied with the age, sex, and race of the person described.

What research associates with discretionary choices

Law sets the outer boundary. Fourth Amendment doctrine on stops, searches, and warrants is the subject of a separate lesson, but its practical effect here is simple: legal rules define what an officer may do, and leave wide room inside that boundary for what an officer will do. Research on what fills that room has been running since the field observation studies of the 1960s. Complainant preference shows up early and persistently; Goldstein documented a department that declined to enforce felonious assault laws when the victim would not sign a complaint, which effectively let the complainant rather than the statute decide whether a felony was a felony. Suspect demeanor has been studied and contested for decades. Worden and Shepard's 1996 reanalysis of the Police Services Study data reported that the association between a disrespectful or hostile demeanor and arrest held up, after other researchers had questioned it. Situation matters as well. The National Academies' 2018 consensus report on proactive policing observes that where policies widen the scope for officer discretion, they may increase the potential for cues such as race to affect behavior, and it summarizes psychological work indicating that task complexity, time pressure, fatigue, and fear push people generally toward superficial cues and hunches. The committee is careful to add that most of that work is laboratory research rather than observation of officers in the field. None of these findings say that officers decide arbitrarily. They say that seriousness and evidence do not explain everything, and that the unexplained remainder is where the argument lives.

Measuring disparity: four tools and their limits

Disparities in stop and search decisions are documented, and the tools for probing them are worth learning precisely. Pierson and colleagues compiled 94,778,505 traffic stops from 21 state patrol agencies and 35 municipal police departments and published the analysis in Nature Human Behavior in 2020. The first tool is a benchmark: compare stops to some population at risk. Their per-capita figures showed black drivers stopped more often than white drivers and Hispanic drivers less often, but the authors immediately noted that residential population is a poor denominator, since it ignores time spent driving, driving behavior, and the fact that drivers often are stopped where they do not live. The second tool, the , compares who is stopped at the same clock time in darkness versus daylight, exploiting seasonal and daylight-saving shifts. Black drivers made up a smaller share of stops after dark in both state patrol and municipal data, which the authors read as suggestive of bias in stop decisions while listing ways lighting and violation type could distort the result. The third tool is the , attributed to Becker: compare hit rates, the share of searches that find contraband. Across state patrol stops, contraband was found in 32.0 percent of searches of white drivers, 29.4 percent of black drivers, and 24.3 percent of Hispanic drivers, even though black and Hispanic drivers were searched roughly twice as often. The outcome test suffers from : if the underlying distributions of risk differ, one race-neutral search threshold can still yield unequal hit rates, so the test can flag bias that is not there and miss bias that is. The fourth tool, the threshold test, estimates the evidentiary bar applied to each group; the authors inferred lower thresholds for black and Hispanic drivers, while stating that thresholds are not identified by the data alone and depend on model assumptions.

What the disparity evidence does and does not settle

Here the honest answer is that researchers disagree, and the disagreement is about causation rather than about the existence of the gaps. The National Academies' 2018 committee stated two conclusions directly. First, large racial disparities in the volume and nature of police-citizen encounters are likely whenever police concentrate enforcement on high-risk people or high-risk places, which many proactive strategies do by design. Second, existing evidence does not establish conclusively whether, or to what extent, those disparities reflect statistical prediction, racial animus, implicit bias, or other causes. The committee paired that with the observation that the history of racial injustice in American criminal justice, together with ethnographic work on the effects of policing in non-white communities, makes the causes a priority for research and policy. Two things follow for a student. Reporting a disparity is not the same as explaining it, and refusing to explain it is not the same as denying it. Deployment is the clearest confounder: if patrol hours are concentrated in some neighborhoods, residents there will accumulate more encounters regardless of how any individual officer decides. That is a real effect on people's lives, and it is a different mechanism from an officer treating two similar drivers differently.

Structuring discretion instead of pretending to remove it

Because discretion cannot be eliminated, agencies and courts try to channel it. Written policy is the first instrument, and its coverage is uneven. Bureau of Justice Statistics data from the 2016 LEMAS survey show that 84.8 percent of local police departments had a written directive on motor-vehicle stops and 71.3 percent had one on stop and frisk; because large agencies are likelier to have policies, those directives covered 95.6 percent and 85.7 percent of officers respectively. Training and supervision are the second instrument, and the Justice Department's 2017 findings report on the Chicago Police Department illustrates what investigators look for: supervisor span of control, whether supervisors work consistently with the same officers, and whether anyone reviews what officers actually do. Recording technology is the third. By 2016, 47 percent of the country's 15,328 general-purpose agencies had acquired body-worn cameras, and about 86 percent of adopters had a formal camera policy. The evidence that cameras change behavior is mixed: the National Institute of Justice reports that a review of 70 studies found no consistent or no statistically significant effects, earning a No Effects rating from CrimeSolutions across use of force, arrests, stops, and complaints, while individual evaluations range from Effective in Boston to No Effects in Washington, D.C. Early intervention systems are the fourth. As the Justice Department described them in the Chicago report, such a system tracks officer conduct, flags risk of future problems, and triggers non-disciplinary responses such as additional training or counseling; Chicago's, the Department found, existed in name only.

Accountability structures and their legal limits

Four accountability mechanisms are worth distinguishing. Internal investigation is the baseline: in 2016, 88.4 percent of local police departments had a written directive on investigating employee misconduct. External criminal investigation requirements are separate and uneven; 59.4 percent of departments required an outside law enforcement body to investigate a firearm discharged at a person, and 82.6 percent required one when force caused death, though those requirements were markedly less common in the largest agencies, and the BJS question excluded civilian review. Civilian oversight bodies sit outside the agency and form a third layer; they are created locally, and the BJS figures just cited say nothing about them, because the survey question expressly excluded civilian reviews. Fourth is federal structural litigation. Under 34 U.S.C. section 12601, enacted as part of the Violent Crime Control and Law Enforcement Act of 1994 and originally codified at 42 U.S.C. section 14141, it is unlawful for a governmental authority to engage in a pattern or practice of law enforcement conduct that deprives people of federal rights, and the Attorney General may sue for equitable and declaratory relief to eliminate it. The Civil Rights Division notes that a single incident is usually not enough, that it cannot investigate federal agencies, and that its settlements typically require data collection, independent oversight, force review, and better training and supervision. Individual damages run on a different track: 42 U.S.C. section 1983 allows suit against officials acting under color of state law, but in Harlow v. Fitzgerald (1982) the Supreme Court held that officials performing discretionary functions are shielded unless their conduct violated clearly established rights a reasonable person would have known, and in Pearson v. Callahan (2009) it held unanimously that courts may address that clearly-established question first without deciding whether a right was violated at all. Described as doctrine and nothing more, means many suits over discretionary decisions end on that question rather than on the merits.

Eli, the EliExplains learning guide

Eli explains

The same idea, in plain words

Explain it like I’m 10

Imagine writing down every rule you would like people to follow in a city, then hiring far too few people to watch for all of it. The people you hired will have to pick what they act on. They are not cheating when they pick. Picking is the job. But once you notice that picking is the job, some questions get sharper. Which picks get written down and which vanish? Who checks the ones nobody wrote down? If two people do the same thing and get different results, is that because the situations differed, because the officers differed, or because one neighborhood gets watched more closely than another? Researchers have built careful statistical tools to attack that last question, and the tools are genuinely good, and they still cannot fully settle it.

Picture it like this

Think of a lifeguard at a crowded public pool. The posted rules forbid running, glass bottles, diving in the shallow end, and breath-holding contests. If she blew the whistle for every violation she would never stop whistling, and she would miss the child actually going under. So she chooses. What she consistently whistles for, and what she consistently lets slide, becomes the pool's real rulebook, no matter what the sign says.

Where the picture stops working

The comparison breaks down in ways worth naming. A lifeguard cannot arrest anyone, cannot search a bag, and carries no weapon; the worst outcome of her judgment is usually an argument. Her decisions are not governed by constitutional law, no court reviews her whistle, and no statute authorizes a federal lawsuit over how she manages the shallow end. Police discretion carries consequences that reach into liberty and life, which is exactly why so much machinery has been built around it.

Worked example

An officer clocks a car ten miles per hour over the limit at 9 p.m. and stops it. That single stop contains at least four separable decisions. First, whether to stop at all, since many similar cars pass unstopped. Second, whether the encounter stays a conversation or extends into a search, which requires a legal basis. Third, the outcome: nothing, a warning, a ticket, or an arrest. Fourth, whether anything is written down. Using the 2022 Police-Public Contact Survey, the aggregate answer to the third question was about 9 percent no action, 44 percent warning, 43 percent ticket, and 4 percent search or arrest. Now break those outcomes out by race, as BJS does: 49.0 percent of white drivers reported a warning versus 34.7 percent of black drivers and 35.9 percent of Hispanic drivers, while 8.6 percent of black drivers reported a search or arrest versus 3.1 percent of white drivers. Those are self-reports, they are one year, and they do not control for stop reason, location, or driving record. They tell you the outcomes differ. Deciding why requires the veil-of-darkness, outcome, and threshold tools, and even those leave the causal question open.

Key takeaway

Discretion is the ordinary condition of enforcing a criminal code broader than any agency's capacity, and it concentrates at the lowest rank where it is least visible. The serious work is in structuring and reviewing it, and in reading disparity evidence for exactly what it measures.

Quick check

3 questions here, of 5 in this lesson’s practice set. Answers stay hidden until you check.

Question 1 of 3foundational

What term did Joseph Goldstein use in 1960 for police decisions not to invoke the criminal process?

Choose an answer, then check it.
Question 2 of 3intermediate

Why did Goldstein conclude that full enforcement is not a realistic expectation?

Choose an answer, then check it.
Question 3 of 3advanced

Searches of one group turn up contraband as often as searches of another. What does the outcome test alone establish about bias in those search decisions?

Choose an answer, then check it.
Practice all 5

Keep learning

Ready to build on this? Continue to the next lesson.

Practice this lesson
Study tools & related lessonsYou’ll learn to · Common mistakes · Easily confused · Key vocabulary · Related

You’ll learn to

  • Define police discretion and explain why it is a structural feature of enforcement rather than a defect in it.
  • Distinguish Goldstein's total enforcement, full enforcement, and no enforcement zones and locate discretion within them.
  • Identify the decision points at which discretion operates and the factors research associates with those decisions.
  • Analyze what benchmark, veil-of-darkness, outcome, and threshold tests can and cannot establish about disparity.
  • Evaluate mechanisms that structure discretion, including written policy, cameras, early intervention systems, and pattern-or-practice litigation.

Common mistakes

  • Treating discretion as a loophole or an abuse to be eliminated.

    Discretion follows from a criminal code broader than any agency's capacity and from statutes that require judgment to apply. The realistic policy question is how to structure and review it, not how to remove it.

  • Reading a raw stop-rate or search-rate gap as proof of discrimination.

    Raw rates depend on the benchmark you pick. Residential population ignores driving exposure and deployment patterns, which is why researchers built the veil-of-darkness, outcome, and threshold tests instead of stopping at the raw numbers.

  • Concluding that equal hit rates prove searches were unbiased.

    That is the infra-marginality problem. Groups with different underlying risk distributions can produce equal or unequal hit rates under one identical, race-neutral threshold, so hit rates alone settle neither direction.

  • Assuming body-worn cameras have been shown to reliably change officer behavior.

    NIJ reports the evidence is mixed; a review of 70 studies found no consistent or no statistically significant effects, while individual city evaluations range from Effective to No Effects.

  • Confusing the different accountability tracks with one another.

    Section 1983 damages suits target individuals and can end on qualified immunity; 34 U.S.C. 12601 gives the Attorney General a civil action for equitable relief against an agency's pattern or practice and cannot recover damages or prosecute anyone.

Easily confused

Discretion vs. Discrimination

Discretion is the authority to choose among lawful options. Discrimination is choosing on a prohibited basis. Every discriminatory decision is an exercise of discretion, but the overwhelming majority of discretionary decisions are not discriminatory, so demonstrating discretion demonstrates nothing about bias.

Outcome (hit rate) test vs. Threshold test

The outcome test compares how often searches succeed and is vulnerable to infra-marginality. The threshold test uses a Bayesian model to estimate the evidentiary bar applied to each group from search and hit rates together, which addresses infra-marginality but makes the result depend on model form and priors.

42 U.S.C. 1983 suit vs. 34 U.S.C. 12601 action

Section 1983 lets an injured individual sue an official for damages, where qualified immunity applies. Section 12601 lets the Attorney General sue a government for equitable and declaratory relief to end a pattern or practice; it produces institutional change, not compensation.

Internal affairs investigation vs. Civilian oversight body

Internal affairs sits inside the agency and reports through its chain of command; civilian oversight sits outside it. BJS's 2016 survey asked only about external investigations by another law enforcement or criminal investigative body and expressly excluded civilian reviews, so those national figures tell you nothing about how common or how empowered civilian oversight bodies are.

Key vocabulary

Police discretion
The lawful authority of an officer to choose among permissible courses of action in a given encounter, including the option of taking no formal action at all.
Full enforcement
Goldstein's term for the zone of conduct that police are legally permitted to pursue once constitutional, statutory, and procedural limits have been subtracted from the whole criminal code.
Low-visibility decision
A choice, typically not to invoke the criminal process, that leaves no record and therefore cannot be reviewed by prosecutors, courts, or supervisors.
Street-level bureaucracy
Lipsky's term for public agencies whose frontline workers deal directly with the public, hold wide practical judgment, and effectively make agency policy through their routines.
Benchmark problem
The difficulty of choosing a defensible comparison population when judging whether a group is stopped or searched disproportionately.
Veil-of-darkness test
An analytic method comparing the racial composition of drivers stopped at the same clock time in darkness versus daylight, on the theory that darkness obscures a driver's race.
Outcome test
An analytic method that compares the share of searches yielding contraband across groups, treating a lower success rate as a sign that a lower evidentiary bar was applied.
Infra-marginality
The reason success-rate comparisons can mislead: when two groups have different underlying risk distributions, a single race-neutral decision threshold still produces unequal success rates.
Pattern-or-practice investigation
A federal civil inquiry under 34 U.S.C. 12601 into whether an agency systematically deprives people of federal rights, resolvable by court-enforceable equitable relief rather than damages.
Qualified immunity
A judicially created defense shielding officials sued for damages over discretionary acts unless the conduct violated clearly established law a reasonable person would have known.

Sources & references

  1. Police Discretion Not to Invoke the Criminal Process: Low-Visibility Decisions in the Administration of Justice, 69 Yale Law Journal 543 (1960) — Joseph Goldstein; Yale Law School Legal Scholarship Repository (openYLS)
  2. Street-Level Bureaucracy: Dilemmas of the Individual in Public Services, 30th Anniversary Expanded Edition (2010) — Michael Lipsky; Russell Sage Foundation (publisher's book page)
  3. A large-scale analysis of racial disparities in police stops across the United States, Nature Human Behavior 4(7):736-745 (2020) — Emma Pierson, Camelia Simoiu, Jan Overgoor, Sam Corbett-Davies, Daniel Jenson, Amy Shoemaker, Vignesh Ramachandran, Phoebe Barghouty, Cheryl Phillips, Ravi Shroff and Sharad Goel; Nature Human Behavior (Springer Nature)
  4. Proactive Policing: Effects on Crime and Communities (2018), Chapter 7, Conclusions 7-1 and 7-2 — Committee on Proactive Policing, Committee on Law and Justice, National Academies of Sciences, Engineering, and Medicine; The National Academies Press
  5. Contacts Between Police and the Public, 2022 (NCJ 308847) — Bureau of Justice Statistics, U.S. Department of Justice
  6. Local Police Departments: Policies and Procedures, 2016 (NCJ 254826, August 2020) — Connor Brooks, Bureau of Justice Statistics, U.S. Department of Justice
  7. Body-Worn Cameras in Law Enforcement Agencies, 2016 (NCJ 251775, November 2018) — Shelley S. Hyland, Bureau of Justice Statistics, U.S. Department of Justice
  8. Research on Body-Worn Cameras and Law Enforcement — National Institute of Justice, Office of Justice Programs, U.S. Department of Justice
  9. Conduct of Law Enforcement Agencies (Special Litigation Section) — U.S. Department of Justice, Civil Rights Division
  10. Investigation of the Chicago Police Department (findings report, January 13, 2017) — U.S. Department of Justice, Civil Rights Division, and U.S. Attorney's Office for the Northern District of Illinois
  11. 34 U.S.C. 12601 - Cause of action (pattern or practice of unconstitutional law enforcement conduct) — Office of the Law Revision Counsel, U.S. House of Representatives
  12. 42 U.S.C. 1983 - Civil action for deprivation of rights — Office of the Law Revision Counsel, U.S. House of Representatives
  13. Harlow v. Fitzgerald, 457 U.S. 800 (1982) — United States Reports volume 457, Library of Congress digital collection
  14. Pearson v. Callahan, 555 U.S. 223 (2009) — United States Reports volume 555, Library of Congress digital collection
  15. Demeanor, Crime, and Police Behavior: A Reexamination of the Police Services Study Data, Criminology 34(1):83-105 (1996) — Robert E. Worden and Robin L. Shepard; Criminology (American Society of Criminology / Wiley)
  16. Somebody's going to emergency, somebody's going to jail: pre-booking diversion and officer discretion, Health & Justice 14(1) (2026) — Kate McLean, E. Suiter, N. Shneibaum and B. L. Miller; Health & Justice (Springer), via PubMed Central

EliExplains lessons are original prose written from the open, credible references above. See Copyright & Licensing.

Researched 2026-08-18

Educational content only. It is not medical, legal or professional advice. Found an error? Tell us.