Disparity and Discrimination in the Criminal Justice System

A statistical gap between groups is the starting observation, never the conclusion. Disparity simply names a difference in rates: if one group is stopped, arrested, or sentenced to prison at a higher rate than another, a disparity exists by definition.

Disproportionality is a related but distinct measure, comparing a group's share of an outcome, such as arrests, against that group's share of the general population. A group can make up 13 percent of a population and 30 percent of arrests for a given offense category; that is disproportionate contact, and researchers in juvenile justice call this specific measure disproportionate minority contact.

Neither disparity nor disproportionality is, by itself, evidence of discrimination, which is a claim about the mechanism: that a decision-maker treated a person differently because of race, class, gender, or another protected characteristic, holding the legally relevant facts constant.

Proving discrimination requires isolating the effect of group membership from every legitimate factor a decision-maker was entitled to weigh, such as offense severity, weapon involvement, or prior convictions. That is a hard empirical problem, because administrative data rarely captures every legitimate factor a police officer, prosecutor, or judge actually used.

Alfred Blumstein's 1982 study of United States prison populations illustrates the method criminologists still use to separate these claims. Blumstein compared the racial composition of state prison populations against the racial composition of people arrested for the same offenses, reasoning that if arrest and imprisonment tracked each other closely for a given crime category, the disparity in imprisonment for that category was better explained by differential arrest, and behind that potentially differential offending, than by decisions made later in the process.

He found that the explained share was high for homicide and robbery and much lower for some drug offenses, meaning later-stage decisions did more work in explaining the racial gap for those offenses.

This decomposition logic, comparing the disparity at each stage against the disparity carried in from the stage before it, is the tool this topic uses throughout. A gap that first appears at sentencing, after controlling for the offense and criminal history a person was arrested and charged with, points toward sentencing-stage discrimination or bias far more directly than a gap that was already present in arrest data before sentencing ever began.

Policing is the entry point, and stop-and-search or stop-and-frisk programs are the most studied example. In the United States, Terry v. Ohio (1968) set the constitutional standard that lets police stop and pat down a person on reasonable suspicion, a lower bar than probable cause.

Studies of large city police departments running aggressive stop programs, most prominently New York City in the 2000s and early 2010s, repeatedly found that Black and Hispanic residents were stopped at rates far exceeding their share of the population, while the rate at which stops of any group turned up a weapon or contraband was low and did not differ enough by group to explain the gap in stop rates alone.

Pretrial detention and bail decisions are the next stage, and they matter enormously because a person detained before trial is more likely to plead guilty, more likely to be convicted, and more likely to receive a custodial sentence, independent of guilt. The United States relies heavily on secured money bail, so ability to pay, not just assessed risk, determines release; this is the central objection driving the American bail reform movement.

England and Wales instead apply the Bail Act 1976, which creates a general presumption in favor of bail that the prosecution must rebut against specific statutory exceptions. India's pretrial detention runs through the Bharatiya Nagarik Suraksha Sanhita, 2023, in force since 1 July 2024, which replaced the Code of Criminal Procedure, 1973, and government prison statistics have for years shown that undertrial prisoners, people awaiting trial rather than serving a sentence, make up a majority of India's total prison population.

Charging and plea decisions sit largely in prosecutorial discretion, which is harder to study than policing because charging files rarely record every factor a prosecutor weighed. Where researchers have obtained case-level prosecutorial data, several studies have found that, controlling for the arrest charge and criminal history, defendants from some racial groups are more likely to be charged with offenses carrying mandatory minimums or to be denied diversion programs, though findings vary by jurisdiction and study design.

Sentencing disparity is the most litigated stage. The United States Sentencing Commission was created by the Sentencing Reform Act of 1984 to reduce judge-to-judge variation through federal sentencing guidelines, but the guidelines themselves built in a stark disparity: the 100-to-1 sentencing ratio between crack and powder cocaine offenses, which fell disproportionately on Black defendants, stood for over two decades until the Fair Sentencing Act of 2010 reduced it to 18-to-1.

Parole and release decisions, the final discretionary stage, have received comparatively less empirical attention but show a similar pattern in the limited studies available: disparities that persist even after controlling for the offense and institutional conduct record.

A gap of a few percentage points at any single stage can look minor in isolation. The concept of cumulative disadvantage, developed in life-course criminology by Robert Sampson and John Laub in their 1993 study Crime in the Making, describes how early disadvantages compound over a life course because each one shapes the opportunities and decisions that follow it.

Applied to a single criminal case rather than a life course, the same logic explains why disparities that look modest at each of six or seven decision points, from stop through parole, can compound into a much larger gap in the final outcome, because each stage's population is filtered by every stage before it.

The mechanism is straightforward once stated. If a group is somewhat more likely to be stopped, and among those stopped somewhat more likely to be arrested, and among those arrested somewhat more likely to be detained pretrial, and among those detained somewhat more likely to be convicted, and among those convicted somewhat more likely to receive a custodial sentence, then the ratio comparing that group's final imprisonment rate to a comparator group's rate multiplies the ratios at each stage together rather than adding them.

Six stages each carrying a modest 10 to 20 percent relative gap can combine into a final disparity several times larger than any single stage would suggest.

This is also why a defensible decision at each individual stage does not guarantee a defensible system overall. A police department, a bail statute, a prosecutor's office, and a sentencing court can each point to a legitimate, case-specific reason for a given decision, and still preside jointly over a system whose net effect is a large disparity, because none of the individual decision-makers is accountable for, or even able to observe, the compounding across the stages the others control.

Cumulative disadvantage analysis is also why reform efforts that target a single stage in isolation, such as sentencing guidelines alone, often move the aggregate disparity less than expected. If the gap entering sentencing already reflects compounded disadvantage from policing, bail, and charging, tightening judicial discretion at sentencing narrows only the portion of the gap generated at that last stage, leaving the compounded portion from earlier stages untouched.

The differential offending account holds that observed disparities in arrest and imprisonment largely track real differences in the rate at which groups commit the offenses that draw the heaviest enforcement, particularly serious violent crime.

Blumstein's 1982 decomposition, discussed above, is the classic evidence cited for this view, and it explained a large share of the racial gap in imprisonment for homicide and robbery specifically. Critics note that this account explains disparities in recorded offending, which is itself shaped by where police patrol and whom they stop, so it cannot rule out differential enforcement upstream of the arrest data it relies on.

The differential enforcement account holds that disparities arise from where and how law enforcement resources are deployed, independent of any underlying difference in offending. If patrol intensity, stop thresholds, or charging aggressiveness vary systematically by neighborhood demographics rather than by crime rate alone, disparities in official statistics will appear even where true offending rates do not differ, or differ by less than the enforcement gap suggests.

The stop-and-frisk evidence described earlier, where stop rates diverged sharply by race while the rate of finding contraband barely did, is generally read as evidence for this account rather than differential offending.

The implicit bias account, drawing on social psychology research including the Implicit Association Test developed by Anthony Greenwald, Debbie McGhee, and Jordan Schwartz in 1998, holds that officers, prosecutors, and judges can hold automatic associations between social groups and criminality or dangerousness that shape split-second or discretionary decisions without the decision-maker consciously intending discrimination or even being aware of the effect.

This account is attractive because it does not require assuming deliberate bad faith, but it is also harder to test directly against case outcomes, since implicit associations measured in a laboratory setting do not map cleanly onto a specific real-world charging or sentencing decision.

The structural or legitimacy-based account, associated with procedural justice research such as Tom Tyler's 1990 book Why People Obey the Law, locates disparity further upstream, in residential segregation, concentrated poverty, and the erosion of trust between communities and legal institutions, which then shapes both the areas police patrol most heavily and the willingness of residents to cooperate with investigations.

On this view, disparity is a downstream symptom of unequal social conditions rather than a defect that can be isolated to any single decision point, which is also why this account is the hardest of the four to test or to remedy with any single procedural fix.

Race is the most studied axis of disparity, but class and gender operate alongside it rather than separately from it. Class shapes outcomes through mechanisms that look procedural rather than explicitly discriminatory: the ability to retain private counsel instead of relying on an overburdened public defender, the ability to post secured bail and avoid pretrial detention, and the ability to absorb a fine rather than face a custodial alternative for nonpayment.

None of these mechanisms names race or class directly, yet each one systematically disadvantages poorer defendants, and because race and class correlate strongly with each other in most of the jurisdictions studied here, class-based disadvantage and racial disparity are difficult to separate cleanly in aggregate statistics.

Gender disparities run in a less straightforward direction. Research going back decades has documented what is sometimes called the chivalry hypothesis: women, especially for less serious offenses, have historically received more lenient treatment than men at charging and sentencing, including higher rates of diversion and non-custodial sentences for comparable conduct.

An opposing strand of research describes an evil woman hypothesis, where women who violate gender expectations, such as those charged with violence or offenses involving their own children, can face harsher treatment than men charged with comparable offenses. Both patterns can hold simultaneously across different offense categories within the same system.

The intersection of race, class, and gender is where the interaction matters most. Kimberlé Crenshaw's 1989 legal scholarship introduced intersectionality as a term for the way overlapping categories of disadvantage produce experiences that a single-axis analysis, looking at race alone or gender alone, misses.

Applied here, a leniency effect measured for women overall can mask a much smaller or absent effect for women of color, or a pattern where the leniency effect for women is concentrated among first-time, non-violent offenders and disappears for repeat or violent offenders, groups in which women of color are frequently overrepresented in the underlying population studied.

Practically, this means a study finding that a given jurisdiction shows no aggregate racial sentencing gap can still conceal a gap that appears only within a specific gender or offense subgroup, and a study finding leniency toward women overall can still conceal harsher treatment of a specific subgroup of women.

Disaggregating disparity analysis along more than one axis at a time, though it requires larger datasets than most agencies routinely publish, is now standard practice in rigorous sentencing research for exactly this reason.

Wrongful convictions offer a distinct lens on disparity, because they are a case where the criminal process reached the wrong answer entirely, and that error is not evenly distributed. The Innocence Project, founded in 1992 by Barry Scheck and Peter Neufeld at the Cardozo School of Law in New York, used DNA testing to secure exonerations in cases that had already gone through conviction and, in many instances, appeal.

Analyses of the resulting exoneration cases have repeatedly found that eyewitness misidentification is the single largest contributing factor, and that Black men are overrepresented among exonerees relative to their share of the general population, a pattern researchers connect to cross-race identification errors, and to some of the same enforcement and charging disparities documented at earlier stages of this topic.

Reform responses have targeted several of the stages described above, with mixed evaluation evidence. Bail reform efforts, including New Jersey's near-elimination of cash bail for most offenses under legislation that took effect in 2017, replaced secured money bail with a risk-based pretrial assessment for most defendants; early evaluations found large reductions in the pretrial jail population alongside no consistent increase in failure-to-appear or new-offense rates, though the risk-assessment tools themselves have drawn separate criticism for potentially encoding the same historical disparities they were meant to remove.

Sentencing guidelines were adopted in several jurisdictions specifically to reduce judge-to-judge sentencing variation, and evaluations generally find they narrow disparity attributable to which judge a defendant happened to draw, though, as the crack and powder cocaine example shows, guidelines can also formalize and entrench a disparity if the underlying rule itself treats comparable conduct differently.

Body-worn cameras, now standard equipment in many police departments across the jurisdictions covered here, have produced more modest and mixed results in rigorous evaluations: some studies find reductions in use-of-force incidents and civilian complaints, while others find little effect on stop or arrest disparities specifically, because a camera records a decision without changing the threshold that produced it.

Data transparency requirements, which compel agencies to publish stop, arrest, and outcome data broken down by demographic category, do not by themselves reduce disparity, but they are a precondition for the stage-by-stage evidence this topic relies on; without published, case-level data, researchers cannot run the decomposition analysis that separates differential offending from differential enforcement in the first place.

For that reason, several reform coalitions have prioritized mandatory data publication as a first step ahead of, and a prerequisite for, more contested substantive reforms such as sentencing guideline changes or use-of-force policy revisions.