Quiz 1 PP

Muller v. Oregon 1908 (Brandeis Brief)

Changed approach - instead of just using legal theory, Brandeis submitted over 100 pages of sociological, medical, and statistical data documenting physical and societal harms of working long hours on women. extra-legal evidence such as social science data can now be used for judicial decision making. Oregon upheld the law limiting women’s hours.

Brown v. Board 1954

Brown elevated behavioral science to level of interpreting fundamental constitutional rights - used “doll test” - psych study showing that segregation caused black children to feel inferior and internalize prejudice. = shift to legal realism - young kids shown black and white dolls and asked to categorize them as either good or bad.

Risk assessment: Sentencing

identifying low risk defendants who can be diverted from prison into probation, help judges determine type/length of a sentence

Risk Assessment: Pretrial

estimate likelihood that a defendant will fail to appear in court or commit a new offense before trial, informing whether they should be released on bail/recognizance, or detained

Risk Assessment": Parole

use risk scores to evaluate readiness for release and set the required intensity of post-release supervision

Risk Assessment: Corrections

Guided by the Risk-Needs-Responsivity (RNR) model, correctional facilities use these tools to allocate scarce resources, ensuring that intensive rehabilitation, cognitive behavioral therapy, and job training are targeted at higher-risk individuals who benefit the most from them

Actuarial Approaches vs. Unaided Professional Judgement

actuarial approaches outperform “clinical” judgment bc these people are human with cognitive biases, fatigue, inconsistency, etc. actuarial tools are statistically weighted and empirically validated. When professionals override these tools, the predictive accuracy decreases.

Trade offs: Risk assessment Calibration

Calibration: A risk assessment tool is perfectly calibrated if a specific risk score corresponds to the exact same probability of reoffending across all demographic groups. For example, if both a Black defendant and a White defendant receive a score of 7, they both have a 60% probability of recidivating BUT This changes the balance of false positive rates: This definition of fairness requires that the tool mistakenly classifies non-reoffending individuals as "high risk" at the exact same rate across all demographic groups.

Incompatibility: If an algorithm is properly calibrated to accurately reflect risk probabilities, the group with the higher base rate of offending will inevitably experience a higher false positive rate. Equalizing the false positive rates requires intentionally uncalibrating the tool, making its actual risk estimates inaccurate for one or both groups.

Trade offs: Risk Assessment

The Human Judgment Question: Human judgment cannot bypass this problem. The trade-off between calibration and error-rate balance is a fundamental mathematical constraint of base rates, not a unique flaw of computers. If a human judge is accurately calibrated in their risk estimations, their decisions will mathematically generate the exact same unequal false positive rates across demographic groups.

What We Know: Despite widespread concerns that risk assessments exacerbate systemic racism by relying on socio-economic variables or criminal histories correlated with race, empirical evidence indicates that adopting these tools generally does not increase existing racial disparities compared to the status quo of human judgment. In many jurisdictions, implementing validated risk assessments has successfully driven down overall jail and prison populations for all demographics without widening the racial gap. Research also shows that well-designed tools achieve comparable predictive validity for both Black and White individuals.

What We Don't Know: We do not fully know the isolated, causal impact of these tools across all jurisdictions because outcomes are heavily dictated by local implementation. The ultimate effect on racial disparities depends on where policymakers draw the threshold for "high risk" and how frequently local judges choose to override the algorithm. Furthermore, it remains incredibly difficult to definitively separate "true" criminal offending from "label bias"—the reality that historically over-policed minority neighborhoods have higher baseline arrest rates for the exact same underlying behaviors.

Unstructured Clinical Judgment

At the least structured end, decision-makers rely on discretionary, subjective evaluations without standardized checklists, explicit scoring guidelines, or formal weighting. This approach suffers from low inter-rater reliability, vulnerability to cognitive bias, and lower overall predictive accuracy.

Structured Professional Judgment (SPJ)

Positioned in the middle of the spectrum, SPJ tools (such as the HCR-20) require assessors to evaluate a standardized, evidence-based list of risk factors. However, rather than using a rigid mathematical formula to generate a final score, the clinician retains discretion to weigh the factors holistically and assign a final risk category (e.g., low, moderate, high)

Actuarial Risk Assessment Instruments (ARAI)

At the most structured end, actuarial tools (such as the Static-99 or VRAG) use explicit, empirically derived algorithms to assign numeric weights to specific inputs. The items are added together to generate a total score, which directly translates to a statistical probability of reoffending or violence based on norming samples. Discretion is completely removed from the scoring process.

Validated Tools in comparison with Unstructured Judgement

• Superiority Over Unstructured Judgment: Virtually all validated tools—whether actuarial or SPJ—outperform unstructured clinical judgment in predicting future reoffending and violence.

• Comparable Utility Across Tools: Different validated tools generally display remarkably similar levels of predictive accuracy (typically yielding moderate-to-large effect sizes, with AUC values clustering around 0.65 to 0.75).

Why are Validated Tools comparable in predictive accuracy?

• Shared Latent Constructs: Most validated instruments measure the same underlying construct (e.g., general antisocial propensity, behavioral dysregulation, or criminal lifestyle), even if the specific questions or items differ slightly across tools.

• Dominance of Static Factors: Criminal history and past violent behavior are the strongest predictors of future behavior. Because most validated instruments incorporate these core historical indicators, they tap into the same variance.

• Statistical Redundancy: Key criminogenic risk factors are highly correlated with one another. Swapping one validated variable for another adds little unique predictive value, leading to a ceiling effect in statistical prediction.

Risk Factor: Fixed Marker

A risk factor that cannot be changed, either inherently or through intervention (e.g., age at first arrest, criminal history, or biological sex).

Risk Factor: Variable Marker

A factor that can change over time or through intervention, but changing it does not alter the risk of the outcome (e.g., age naturally increases over time, but manually altering an indicator correlated with age does not change underlying risk).

Risk Factor: Variable Risk

A factor that can change and whose change correlates with a shift in outcome probability, but it has not yet been demonstrated experimentally that changing this factor causes the reduction in risk.

Risk Factor: Causal Risk (Dynamic Criminogenic Need):

A variable risk factor that, when successfully modified through intervention, directly leads to a demonstrated reduction in the probability of future reoffending (e.g., active substance abuse, antisocial attitudes, or association with criminal peers)

Distinction: Using risk assessment for predicting risk vs reducing recidivism

• Predicting Risk: If the goal is purely classification or prediction, static fixed markers (like past convictions) are highly efficient and reliable. However, relying solely on fixed markers creates a static assessment that can never demonstrate whether an individual has successfully rehabilitated.

• Reducing Recidivism: If the goal is risk reduction, fixed markers are unhelpful because they cannot be changed. Treatment programs must target causal risk factors. Assessing dynamic needs ensures that correctional resources are directed toward modifiable behaviors, allowing agencies to track changes in risk levels over time as interventions take effect.

Which risk factors: Bail, Security level, Parole eligibility

RISK PREDICTION

• Accurately estimate the overall probability of reoffending to inform detention or supervision levels.

• Fixed Markers & Variable Markers (Static, historical variables offer maximum predictive stability).

• Actuarial Instruments (ARAI) focused on static factors (e.g., Static-99). Efficient, highly reliable, and stable over time.

RISK REDUCTION

• Identify specific targets for intervention and measure whether treatment successfully reduces offender risk over time.

• Causal Risk Factors (Dynamic criminogenic needs that can be directly remediated through treatment).

• Dynamic Actuarial or SPJ Instruments assessing dynamic needs (e.g., LSI-R, COMPAS, HCR-20).

Definition: Disproportionate Minority Contact and Confinement (DMC)

overrepresentation of racial and ethnic minorities at every key decision point within the juvenile and adult criminal justice systems relative to their proportion in the general population

Degree of disparity between different races (DMC)

Historically and throughout the era of mass incarceration, Black Americans have been incarcerated in state and federal prisons at rates approximately 6 to 7 times higher than White Americans (and Hispanic Americans at rates roughly 2 to 3 times higher). For Black men born during the peak of mass incarceration, the lifetime risk of imprisonment was approximately 1 in 3, compared to about 1 in 17 for White men.

Offense types for DMC:

While Black overrepresentation is present across violent crime categories, DMC is most severely skewed in non-violent offenses—specifically drug offenses, weapons charges, and low-level public order violations—where law enforcement discretion is highest

Causes of DMC

1. Drug and Crime policies of 80s/90s (malign neglect)

• disparate impact on black communities was foreseeable and documented at the time. politicians knew they would disproportionately incarcerate black americans

2. Differences in Violent crime rates

• black americans commit serious violent offenses at higher rates than white americans (bc of structural inequity: poverty, segregation, bad education, limited econ. mobility)

• BUT violence differences between groups only account for 50-70% of black/white disparity in prison populations for violent crimes

• CANNOT explain vast racial disparities in drug offenses, property crimes, or low-level public order arrests

3. Bias

• unconscious racial biases in judges, law enforcement, that associate blackness with higher criminality

• discretionary biases have accumulative affect: police more likely to stop and search black people, prosecutors offer less favorable deals/diversion, judges set higher bail and assign longer sentences

Why was the War on Drugs so bad?

1. symmetrical use/asymmetrical enforcement

• use and sell at comparable rates, with white people surpassing in absolute numbers

2. targeted policing

• heavy saturation in inner city areas, buy-and-bust operations, stop and frisk, open air drug markets targeted in inner cities

3. correctional impact

• As a result, Black individuals were arrested for drug crimes at 3 to 5 times the rate of White individuals and were far more likely to receive mandatory prison sentences upon conviction. Drug enforcement directly drove mass incarceration without meaningfully reducing drug availability, usage, or addiction rates.

Risk Assessment: Benefits

if used in justice system, can

• lower prison population without increasing crime

• increases consistency

• opposes “just deserts” (retributive justice) that focuses on just what the offender actually did vs. their risk of offending again

Black male statistics from class

• if a black man was born in the 70s and didn’t finish highschool, 70% risk of federal incarceration

• racial disparities decreased by 40% in last 2 decades: 8:1-5:1

Tonry’s recommendations for reducing racial disparities

a. radically reduce prison population

b. shift drug policy from street policing to prevention, treatment, tolerance

c. reduce racial profiling by the police (such as stop and frisk)

d. reduce weight of criminal record in sentencing

Why social science for crime prevention?

• looking at lifespan perspective (age + risk), economics, behavior

• what an influence policies, regulatory measures, laws, etc

• two ways to approach crime from pub pol perspective: preventing within lifespan, and public safety

2010-2026

• 2018: First Step Act: first mjor federal reform, reduced crack/cocaine disparities

• brianna taylor/george floyd

• 2021-2022, serious increase in violent crime

• 2022-now: justice reform not over, but decentralized

3 diff. simulations for black incarceration - Tonry 2006

1. reduce black incarceration by 10%: racial disparity goes from 6:1-5:1 100k red. in black prisoners

2. cut 2006 prison pop in half: 6:1-5.5:1, 500k red in blakc prisoners

3. return to 1980 rate: 6:1-6:2, 700k fewer black prisoners

AKA problem is mass incarceration, systemic volume

Discrimination cause

discretion + bias

• disparate attention causes disparate outcomes = differential selection

• reducing discretion more useful than bias training

Dark figure of crime

knwon to vicim but not reported to police, or “victimless” like sellung drugs

strengths/weaknesses of UCR and victimization surveys

• UCR: twice a year, but reporting late/erratic, doesn’t get age/race crossover, emits crimes that go unreported to police, standardized

• victimization surveys: small sample sizes, environmental characteristics of neighborhood not considered, every 3.5 yrs, excludes homicides, arson, commercial burglaries, victimess crime, relies on memory, excludes unhoused individuals, captures unreported crime directly from victims

crime trends

• 1970s-80s boomer bulge, crime increasing: social change, urban decay, crack epidemic

• The 1980s Trough: Driven primarily by demographic aging. As the large "Baby Boomer" generation aged out of their peak offending years (ages 15–24), the relative proportion of high-risk young males in the general population declined, temporarily blunting crime growth

• 90s: violent crime rate peaked in 1991-93 (fallen by half since then)

• The Early 1990s Spike: Driven by the rapid expansion of the crack cocaine market in urban centers. Sold in low-cost single doses in open-air markets, crack organizations recruited large numbers of young, non-adult males to sell drugs. To protect cash, inventory, and territory in an illicit market lacking legal dispute mechanisms, these young sellers acquired handguns in unprecedented numbers. This created an explosive, age-specific surge in juvenile and youth homicides, even as violent offending among adults aged 25 and older remained flat or declined

• decline from 1992 peak

• pandemic bump

incarceration trends

• 1920s-1970s: stable, 110/100k

• 1960s goldwater election, change from “rehab” mindset

• 70s increasing exponentially 6-7% each year

• 
  

Correlational Studies

• Design Characteristics: Researchers measure two or more variables as they exist naturally, without intervening or attempting to alter the environment.

• Allowable Conclusions:

  • Description: Summarizes patterns, trends, or distributions across variables (e.g., documenting that higher educational attainment correlates with higher political participation).

  • Prediction: Enables forecasting an outcome based on knowledge of a predictor variable (e.g., using a person's income level to predict their probability of voting).

  • Causation: Cannot establish causality. Correlational designs cannot rule out confounding variables (a third variable driving both $X$ and $Y$) or reverse causality ($Y$ causing $X$ rather than $X$ causing $Y$)

Quasi-Experimental studies

• Design Characteristics: Researchers evaluate the impact of a intervention, treatment, or exogenous policy change across groups or time periods, but without random assignment. Common methodologies include difference-in-differences, regression discontinuity, and interrupted time-series designs.

• Allowable Conclusions:

  • Description & Prediction: Accurately maps pre- and post-intervention trajectories across non-equivalent groups.

  • Causation: Supports plausible or conditional causal inferences, provided key design assumptions hold (e.g., the parallel trends assumption in difference-in-differences). However, because assignment to treatment is non-random, selection bias or unobserved temporal shocks can still threaten causal conclusions.

Experimental Studies (Randomized Controlled Trials)

• Design Characteristics: The investigator directly manipulates the independent variable (assigning treatment vs. control/placebo conditions) and uses random assignment to place units into conditions.

• Allowable Conclusions:

  • Causation: Enables definitive causal statements ($X$ causes changes in $Y$). Random assignment ensures that treatment and control groups are probabilistically identical on all pre-existing characteristics (both observed and unobserved) prior to treatment. Consequently, any post-treatment difference in the outcome variable can be mathematically attributed to the manipulation of the independent variable

Internal Validity

Internal Validity: The degree to which a study accurately demonstrates a true cause-and-effect relationship between the independent variable and the dependent variable, free from the influence of confounding variables, methodological artifacts, or alternative explanations

External Validity

The degree to which the findings of a study can be generalized beyond the specific sample, setting, measurement, or temporal context of the original experiment to other populations and real-world environments

Examples of High validity

Examples of High Validity

• Unusually Strong Internal Validity: A double-blind, randomized laboratory experiment in psychology testing a cognitive intervention, conducted in a soundproof room with computerized stimulus timing, identical lighting, and a standardized placebo control. By strictly controlling the physical environment, researcher interaction, and participant assignment, the study eliminates virtually all competing explanations for observed changes in the dependent variable.

• Unusually Strong External Validity: A nationally representative, multi-site field experiment evaluating a voter turnout campaign across 50 diverse municipalities, using public voter registries and real-world elections. Because the trial takes place in natural electoral conditions with real voters across varied demographics and geographic regions, the findings easily generalize to actual political behavior in the broader electorate (though internal validity is slightly lower than a lab due to real-world noise)

Independent vs. Dependent variables

• Independent Variable (IV): The variable hypothesized to be the cause, predictor, or antecedent. In an experimental study, it is the variable directly manipulated by the researcher (e.g., receiving a specific job-training program vs. a control package). In observational research, it is the variable conceptualized as driving or explaining variation in the outcome.

• Dependent Variable (DV): The variable hypothesized to be the effect, outcome, or criterion. It is measured by the researcher to determine whether and how much it varies in response to changes in the independent variable (e.g., employment status or hourly wage 12 months post-intervention)

Cause of crime decrease since mid 90s

• Crack Market Stabilization: Formalized territory boundaries, market maturation, and a strong cultural anti-crack norm developing among younger generations.

• Economic Growth: Strong employment markets in the late 1990s provided legitimate opportunities for young adults.

• Policing & Community Infrastructure: Shifts toward data-driven policing (e.g., CompStat), targeted gun enforcement, and expansion of local community-based organizations.

• Environmental Factors: Reductions in environmental lead exposure (which adversely impacts neurodevelopment and impulse control).

• Incarceration: Expansion of prison populations provided an initial incapacitation effect (though with diminishing returns)

Nature and Effect of War on drugs

• Depended on heavy criminalization, mandatory minimum sentences (such as the federal 100:1 crack-vs.-powder cocaine disparity), and aggressive police saturation targeting low-income minority neighborhoods.

• Effect: Failed to meaningfully alter drug usage or market availability. Instead, it exponentially inflated arrest and imprisonment rates—particularly for young Black males—and inadvertently escalated street violence by destabilizing drug markets and incentivizing handgun arming

Blumstein effect of incarceration on crime

• Overall Effect: Expanded incarceration exerts a modest overall crime-reduction effect, but it operates under diminishing marginal returns. As the prison population grows, each additional incarcerated individual yields progressively smaller reductions in overall crime.

• Attributable Crime Drop: Empirical consensus estimates that the expansion of incarceration accounted for only 10% to 25% of the major crime drop seen in the 1990s and 2000s. Non-carceral social, economic, demographic, and policing factors accounted for the remaining 75% to 90%