What Causes Crime?

What a criminological theory is, and what would count as evidence

The American Society of Criminology describes its members as pursuing knowledge about the measurement, etiology, consequences, prevention, control, and treatment of crime and delinquency. Etiology means causation, so this lesson's question is a defining object of the discipline.

A criminological theory is not a hunch about why people offend. It is a general explanation that names its concepts, states how they relate, and predicts specifically enough to be shown wrong. The working standard is the ordinary standard of social science: a theory must be falsifiable; its concepts must be operationalized, meaning translated into something measurable; and researchers must rule out spuriousness, where two things move together only because a third is moving both. Ice cream sales and murder rates rise together, and nobody thinks the ice cream is doing the work.

What counts as evidence, then, is not a memorable case. A theory gains support when its predictions hold across independent samples, across different measures of the same concept, and against designs that make competing explanations implausible. It loses support when the relationship vanishes once a plausible confounder is controlled, or when it holds for one offense type but not the others it claims to cover. Theories are therefore supported for particular offenses and particular populations, not universally. No single theory explains crime.

Four levels of explanation, and why they are not automatically rivals

Explanations of crime operate at different scales. An individual-level account asks why this person and not that one; a situational account asks why this moment, this target, this place; a community-level account asks why offending concentrates in some neighborhoods; a structural account asks why rates differ across societies, eras, and social positions. Criminology texts draw much the same distinction as macro versus micro: macro-level explanations address group rate differences and point toward social structures, while micro-level explanations address differences among individuals and point toward social processes.

Levels change what a theory is answering. "Why did this burglary happen?" and "Why does this city have a higher burglary rate than that one?" are different questions, and an answer to one is not automatically an answer to the other. So theories at different levels need not be competitors: a structural account of how opportunity is distributed and an individual account of who takes an available opportunity can both hold. They become rivals only when they predict different things about the same outcome, measured the same way, in the same population.

The classical school and deterrence

Modern deterrence theory descends from two Enlightenment writers. Cesare Beccaria published Dei delitti e delle pene - On Crimes and Punishments - in 1764, attacking cruel and arbitrary treatment of the accused and arguing that penalties should be proportioned to harm, set by legislatures, and applied equally. Jeremy Bentham, the English utilitarian, held that punishment must be justified by its effects. The 2014 National Research Council review summarizes their shared framework: deterrence has three ingredients - severity, certainty, and celerity, or swiftness - and severity alone cannot deter, because a penalty unlikely to arrive threatens little. Beccaria thought probability mattered more than cruelty.

Research has largely borne that out, and it is the most policy-relevant empirical result in the unit. The National Institute of Justice's summary of the deterrence literature states that the certainty of being caught is a vastly more powerful deterrent than the punishment; that increasing severity does little to deter, partly because people know very little about the sanctions attached to specific crimes; and that police deter chiefly by strengthening the perception that offenders will be caught. The National Research Council reached the same place from the sentencing side, concluding that the incremental deterrent effect of increases in already-lengthy prison sentences is modest at best.

The NRC adds a qualification that is easy to lose: essentially all the evidence on certainty concerns the certainty of apprehension, not the certainty of imprisonment once a person is convicted. Keep one more distinction straight. Deterrence is a person not yet caught deciding not to offend; incapacitation is a confined person being unable to offend.

Positivism, biology, and a history that has to be named

Where the classical school assumed a free and calculating chooser, positivism assumed that offending has measurable causes. Positivism in this sense is a commitment to measurement, objectivity, and causal inference, and that commitment is now simply how criminology works. Its early substantive content, however, was largely wrong and did severe harm.

Cesare Lombroso, an Italian physician, published the first edition of Criminal Man in 1876 and claimed that a large share of offenders were born criminals - evolutionary throwbacks identifiable by physical features. Kaufmann and Vestad document the method: the book opens with cranial measurements of sixty-six skulls sorted by circumference, offered as proof that criminal skulls were abnormal. Charles Goring later added claimed mental deficiencies, and in the United States H. H. Goddard used intelligence testing to sort people for institutionalization, deportation, or sterilization.

This was not a fringe. The National Human Genome Research Institute records that eugenics - the scientifically erroneous theory that humans can be improved by selective breeding - drew popular, elite, and governmental support in many countries, and that at least 60,000 people were involuntarily sterilized under thirty states' laws by the 1970s, disproportionately Latinx, Native American, African American, poor white, and disabled people. In Buck v. Bell, 274 U.S. 200 (1927), the Supreme Court upheld Virginia's compulsory sterilization statute over a lone dissent, and Justice Holmes wrote for the Court that society may prevent the manifestly unfit from reproducing rather than wait "to execute degenerate offspring for crime." A hereditary theory of crime was written into American constitutional law.

Contemporary biosocial criminology is a different enterprise and must not be described in the old vocabulary. There is no crime gene. Research reports modest correlations between certain genetic variants, neurochemical and hormonal measures, and antisocial behavior - correlations whose direction is often unclear, since testosterone rises after aggressive behavior as well as before it, and whose effects depend heavily on environment. Anything stronger than a conditional, probabilistic statement goes past the evidence, and the paragraph above is what stronger statements have been used for.

Psychological approaches

Psychological accounts locate causes in relatively stable individual differences: personality, self-regulation, cognition, development. Sheldon and Eleanor Glueck concluded from mid-century data that there is no single criminal personality; later work found instead a cluster of interrelated traits correlated with offending, including impulsivity, low self-control, low empathy, and difficulty learning from punishment. The Southern Oregon criminology text reports Caspi and colleagues (1994) finding two broad traits, constraint and negative emotionality, to be robust correlates of delinquency.

None of these traits is criminal in itself, and most people who score high on them never offend. Treat a personality correlate as you would any risk factor: it shifts a probability across a population and predicts very little about a particular person - which is true of nearly every correlate in this field.

Social structure and social process: the two families

Most of contemporary criminology sorts into two families, and the rest of this unit is organized around them. Social-structural theories ask why crime rates differ across groups, places, and periods, and look to how a society distributes opportunity, disadvantage, and collective capacity. Social-process theories ask how a particular person comes to offend, and look to learning, attachment, control, and official reaction.

Here is the map, one sentence each, because each has its own lesson. Strain theory, following Merton in 1938, argues that a disjunction between culturally promoted goals and the legitimate means available for reaching them produces pressure toward crime. Social disorganization, from Shaw and McKay's Chicago research, treats crime as a product of neighborhoods whose social institutions have lost the capacity to regulate behavior. Social learning theory, in Edwin Sutherland's differential association, holds that criminal behavior is learned in interaction within intimate personal groups. Labeling theory examines the social construction of crime and how being labeled and treated as a criminal shapes what a person does next. Routine activity theory moves the question off offenders entirely: an offense occurs when a motivated offender and a suitable target converge in time and space without a capable guardian, as Cohen and Felson argued in 1979. They sit at different levels - two mainly structural, two mainly processual, one situational - which is why they are as often complementary as competing.

The age-crime curve and desistance

One regularity here is about as solid as anything criminology has. Offending rises through late childhood, peaks in the teenage years - NIJ puts the peak at roughly ages fifteen to nineteen - and declines through the twenties and beyond. NIJ calls this bell shape universal in Western populations while noting real variation inside it: violence peaks later than property crime, girls peak earlier than boys, and the curve is higher and wider for young men in the most disadvantaged neighborhoods. NIJ's deterrence summary reports a steep decline at about age thirty-five, and the National Research Council notes that even highly active career offenders typically desist in their thirties - one reason selective incapacitation is hard to implement, since high-rate offenders are easy to identify in retrospect and hard to identify in advance.

What the curve means has been argued about for forty years. Travis Hirschi and Michael Gottfredson put the relationship at the center of criminological theory in "Age and the Explanation of Crime" in 1983; NIJ's 2021 desistance volume records their position that longitudinal data were unnecessary because the pattern held across time and place, against the criminal-career researchers' insistence that following the same people over time was essential. Eleanor and Sheldon Glueck had already run some of the first longitudinal panels in criminology, following the same men over decades and showing that behavior improves over time even among serious offenders. Robert Sampson and John Laub - whom NIJ calls pioneers of life-course criminology - extended that tradition, defining desistance in their 2003 work as the absence of new arrests up to age seventy.

That work is why desistance is now treated as a process rather than an event. NIJ, citing Laub and Sampson, summarizes the sociological side of the literature as pointing to common life events such as stable employment and marriage, and NIJ's study-group review reports strong evidence that for males marriage and stable work foster desistance while unstructured time with peers is associated with persistence. One caution matters especially: desistance can be underway while offending is still occurring, which is why NIJ urges measuring criminality - the propensity - and not only behavior.

Correlation, causation, and why research design carries the weight

Almost every interesting claim about crime begins as a correlation, and the correlations are nearly all confounded, reciprocal, or both. Places with more police record more crime partly because police are what record crime. Incarceration rates respond to crime rates at the same time crime rates may respond to incarceration. The National Research Council makes the point directly about state panel studies: how much incarceration reduces crime depends on how strongly incarceration rates themselves depend on crime rates, and after weighing that literature the committee concluded it could not arrive at a precise estimate, or even a modest range of estimates, of incarceration's effect on crime. That is a striking result for one of the field's most studied questions, and a warning about the rest.

So design carries the weight. Randomized experiments, natural experiments in which exposure is assigned by something unrelated to the outcome, and longitudinal panels following the same people or places over time are what let a criminologist argue for causation rather than association - and why the age-crime debate turned, since a cross-sectional snapshot cannot distinguish a population aging out from individuals changing.

The same discipline is required when the variables are income or race, and what is measured should be stated. In the 2023 National Crime Victimization Survey, the violent victimization rate was 39.0 per 1,000 persons age twelve or older in households earning under $25,000, against 15.7 per 1,000 in households earning $200,000 or more; it was 26.9 per 1,000 for non-Hispanic Black respondents and 22.5 for non-Hispanic white respondents, against a national rate of 22.5. But measurement is not neutral. Police-generated statistics are shaped by where police are deployed, and the National Academies' 2018 proactive policing report concluded that large racial disparities in police-citizen encounters are likely wherever police concentrate on high-risk people or places, and that existing evidence does not conclusively establish whether those disparities indicate statistical prediction, racial animus, implicit bias, or other causes. Report the measured differences, name the measurement problem, and do not present a causal account of them as settled. Recorded crime is not total crime either: BJS states that the system does not respond to most crime because so much is never reported, and the FBI's 2023 national figures rest on voluntary agency submissions.

What follows for policy, and what does not

A theory, even a well-supported one, does not hand you a policy. The deterrence research supports raising the perceived certainty of apprehension - but which agency does that, at what cost, by what methods, with what effect on the people policed, and with what consent from the community are separate questions no deterrence finding answers. Values enter where the evidence stops, and honest analysis marks that line.

This lesson is educational material about criminology and the United States criminal justice system. It is not legal advice and takes no position on any policy question. Crime is defined and measured under state as well as federal law, and both vary by jurisdiction.