A newspaper headline says robberies fell 35% in a year. Is that because there were fewer robberies, or because fewer victims phoned the police? Both can be true, and which one is true changes what the government should do. The figure that looks so firm is the end of a chain of human decisions: someone decided to act, someone decided the act was a crime, someone decided to report it, someone decided to record it.
That chain is what this course teaches you to see. It does not say crime is "just a social construct" (a person who has been mugged knows better) and it does not treat official numbers as a plain count. It asks why societies have rules at all, why some people break them, why some people and places get caught more, and what punishment does. The topic is one where readers arrive with strong views, so this piece gives the evidence with its dates and keeps the facts apart from the arguments about what to do.
Émile Durkheim, writing in The Rules of Sociological Method (1895), made a claim that sounds shocking and is meant to be dry: crime is normal. Every known society has it. A society with no crime at all, he said, is not imaginable, because even a community of saints would treat small faults the way we treat crimes. Crime, on this view, marks a boundary: the moral rules of a group are made visible when someone breaks them and the group reacts. Durkheim did not call crime good, and he did not mean any particular amount was fine. A rate that is unusually high for a society is a sign that something is wrong with it.
His book Suicide (1897) was the method in action. Suicide looks like the most private act, but its rates differ systematically between groups and stay stable inside each society from year to year. He argued the rates reflect two forces, social integration and moral regulation, and named four kinds: egoistic (too little integration), altruistic (too much), anomic (too little regulation, as in sudden economic booms and busts) and fatalistic (too much). Anomie, a state of normlessness, became his lasting contribution. Critics note that his data were crude, that official suicide figures vary by how coroners record deaths, and that the famous Protestant-Catholic gap is weaker and more varied than he said. The lesson that survived is the method: look at rates, and ask what in the society produces them.
Later theories divide into families depending on what they take for granted. Strain theories assume everyone wants roughly the same things and ask what happens when the means are blocked. In "Social Structure and Anomie" (1938), Robert Merton argued that a society that preaches success for all but spreads the legitimate means unevenly pushes some people toward other routes. He listed five ways people respond, and a person can use different ones in different parts of life.
Does the person accept the society's goal (say, financial success)? Do they use the approved means (say, a steady job)?
Merton's typology, 1938. His example for "retreatism" has been criticised, and the scheme describes a US-style success goal; other societies may have different goals.
Without JavaScript: accepting both goal and means is conformity; accepting the goal but not the means is innovation (which includes crime); rejecting the goal but keeping the means is ritualism; rejecting both is retreatism; replacing both is rebellion.
Learning theories say offending is learned like other behaviour. Edwin Sutherland's differential association (1939) held that people become offenders by learning, from the people they are close to, more definitions favourable to breaking the law than against. Sutherland also coined "white-collar crime" (1939), pointing out that criminology had studied the poor while ignoring offences by the respectable. Control theories flip the question. Travis Hirschi (Causes of Delinquency, 1969) argued that the interesting thing is not why people offend, since temptation is common, but why most do not. His answer is the social bond: attachment to others, commitment to conventional goals, involvement in conventional activities, and belief in the rules. Later, Gottfredson and Hirschi (1990) proposed low self-control as the common cause; many criminologists accept that self-control matters and dispute that it is the whole story.
Marxist and conflict theorists, and feminist criminologists, ask the questions the others skip: who writes the laws, whose harms count, and why women's offending and women's victimisation were ignored for so long. Each school has its own best evidence, and the field has not picked a single winner. Most researchers use several: strain for pressure, learning for pathways, control for restraint.
A different approach stops asking about offenders and counts where and when. Lawrence Cohen and Marcus Felson's routine activity theory (1979) says a crime needs three things to meet in space and time: a motivated offender, a suitable target and no capable guardian. Change daily life (more homes empty in the daytime, more portable goods) and crime changes without anyone's character changing. Concentration is the other pillar. In Minneapolis, Lawrence Sherman and colleagues (1989) found that about 3% of addresses produced half of all calls to the police. David Weisburd (2015) argued this is something like a law: in city after city, a few percent of street segments account for about half the crime, and the pattern holds over time. It connects to Robert Sampson's findings on neighbourhoods in Cities, Communities & Social Networks (35/81): collective efficacy and the physical setting both matter. It also shows why a city's crime rate is a poor guide to the risk on your own street.
Another turn in the 1950s and 1960s moved attention from the act to the reaction. Edwin Lemert (1951) separated primary deviance, which may be fleeting and explained away, from secondary deviance, which happens once a person takes the label on as an identity and others treat them accordingly. Howard Becker's Outsiders (1963) put the idea sharply: deviance is not a quality of an act but the result of other people applying a rule and a label to someone. The point is not that nothing wrong happens. It is that who gets labelled depends on power, visibility and who is watching.
A well-known field test of label effects is Devah Pager's audit study in Milwaukee (2003). Pairs of young men with matched resumes applied for the same entry-level jobs. Among white applicants, 34% of those with no record got a callback, against 17% of those with a drug conviction on their record. Among Black applicants it was 14% with no record and 5% with one. The record cut callbacks by about half for both, and a Black applicant without a record did slightly worse than a white applicant with one. Separate follow-up studies of young people (for example the Rochester Youth Development Study, Bernburg and Krohn, 2003) found that formally processed youths later offended more than similar youths who were not, though critics point out that the youths who were processed may have differed in ways the studies could not measure.
The same logic runs through data. Where police look, they find; where they find, records pile up; where records pile up, police are sent. Kristian Lum and William Isaac (2016) showed this by simulating a predictive-policing tool on Oakland's drug-offence data: because it learned from past arrests, it kept sending officers to the same neighbourhoods, though survey-based estimates suggested drug use was spread much more widely. Ensign and colleagues (2018) described such "runaway feedback loops" in general terms. The tool below is a toy version. It does not show that real police departments behave like this; it shows what happens if records drive patrols and nothing else corrects them.
A toy model: 20 offences a day across two neighbourhoods; the chance an offence is recorded rises with the share of patrols sent there, and patrol share follows (past records)^k. k = 1 is "in proportion to records", k above 1 concentrates patrols on the "worst" area. Invented parameters, expected values, no noise.
Without JavaScript: if both neighbourhoods have the same true offending (50%) but A starts with 55% of past records, patrols proportional to records leave A with about 52% after a year, and patrols that concentrate on the "hot" area (k = 2) push it to about 76%.
Criminologists call the crime that never reaches the record the dark figure (from the German Dunkelfeld). Police-recorded crime counts only what is reported and then logged. Victimisation surveys fill in the rest: since the 1970s in the United States (the National Crime Victimization Survey) and since the early 1980s in England and Wales (now the Crime Survey for England and Wales), households are asked what happened to them, whether or not they told anyone. The two sources often disagree in level and sometimes in trend. In 2014 the UK Statistics Authority withdrew its official "National Statistics" label from police-recorded crime in England and Wales after doubts about recording practice.
Whether victims report depends on the crime and on trust. The US Bureau of Justice Statistics found that 42% of robberies were reported to police in 2023, down from 64% in 2022, and 72% of motor-vehicle thefts in 2023, down from 81%. If the true number of robberies had stayed the same, police figures would have shown a drop of about a third that year with no change in the street. Homicide is the exception that proves the rule: a body is hard to hide, so it is the best-measured crime. The UN Office on Drugs and Crime estimated about 458,000 homicide victims worldwide in 2021, with the highest regional rates in Africa and the Americas and the lowest in Europe and Asia.
The 1,000 is invented and held fixed so you can see the reporting effect alone; the 64% and 42% are BJS's reporting shares for robbery (Criminal Victimization, 2023). A real change in reporting would not leave the true count unchanged.
Without JavaScript: with 1,000 real robberies, a 64% reporting rate gives 640 reports; at 42% it gives 420, a fall of 34%, with no change in actual robbery.
Reporting also varies by country, by how much people trust the police, and by what the law counts. So comparing the "crime rates" of two countries from police records mostly compares their police and their definitions. This is why serious comparisons use victimisation surveys or homicide.
Societies punish in different amounts. The World Prison Brief (Institute for Crime & Justice Policy Research at Birkbeck, University of London) collects the number of people held in prisons per 100,000 residents for nearly every country. The chart below reproduces a selection as read from the World Prison Brief's ranking on 2 November 2026; each country's reference year differs (it is shown on its WPB page) and figures are revised regularly.
Selected countries, World Prison Brief highest-to-lowest ranking, accessed 2 November 2026. Each country has its own reference year. China's WPB figure may cover sentenced prisoners only, so it understates total detention. El Salvador's figure follows a state of exception declared in March 2022.
Without JavaScript: El Salvador 1,659; Cuba 794; United States 542; Türkiye 504; Rwanda 489; Brazil 439; Mexico 202; England and Wales 140; China 119; Germany 69; Norway 54; India 36; Nigeria 36; Japan 33; Yemen 14 (per 100,000, WPB ranking read on 2 November 2026).
Read it with care. Rates are not crime rates: a country can imprison many people for few crimes or the reverse. They depend on what is criminalised (drug laws are a large share in some countries), sentence lengths, pre-trial detention (large in much of Latin America and South Asia), and whether the state records detention outside the prison system. Different readers draw different conclusions. Those who argue for imprisonment point to incapacitation and deterrence; those who argue against point to costs, to what a record does to a person's life (see Pager above) and to evidence that longer sentences deter little compared with the certainty of being caught. The El Salvador case shows the split sharply: its government says the state of exception collapsed gang violence, while human-rights organisations document arbitrary detentions and deaths in custody. Both claims can be checked against evidence and neither settles the question of what should be done.
The sociological stance is to take each policy claim as an empirical question. Pick an approach to see what supporters say, what evidence exists and what critics say.
Without JavaScript: the approaches are more officers, hot-spot patrols, order-maintenance ("broken windows"), community and procedural-justice policing, body-worn cameras, alternative responders, and reducing the role of police.
Two cautions run through all of them. First, most US studies are about the US and may not carry over to Mexico City, Lagos or Glasgow. Policing means different things in different places: Japan's neighbourhood koban boxes, Latin America's "mano dura" (firm hand) policies, Britain's "policing by consent", which is often attributed to Robert Peel but whose attribution historians question. Second, "does it reduce crime?" is only one question. Whether people are treated fairly, whether the burden falls evenly and what it costs are separate questions, and people weigh them differently.
Crime is a social fact in two senses: every society has rule-breaking, and every crime figure passes through human choices about reporting and recording. The main theories ask different questions (what pressures, what is learned, what restrains), crime is concentrated in a few places, labels can become causes, and punishment rates differ enormously for reasons beyond crime itself. Policing debates are best handled by asking what the evidence for each approach is, and which questions it answers.
Try this: the next time you see a crime statistic in the news, write down who must have reported it, who recorded it and what else could have changed besides the amount of crime.
Sources. Durkheim, Rules of Sociological Method (1895) and Suicide (1897); Merton, American Sociological Review 3 (1938); Sutherland (1939, 1949); Hirschi (1969); Gottfredson and Hirschi (1990); Cohen and Felson, ASR 44 (1979); Sherman, Gartin and Buerger, Criminology 27 (1989); Weisburd, Criminology 53 (2015); Lemert (1951); Becker (1963); Pager, American Journal of Sociology 108 (2003); Bernburg and Krohn, Criminology 41 (2003); Lum and Isaac, Significance 13 (2016); Ensign et al., PMLR 81 (2018); BJS, Criminal Victimization, 2023 (September 2024); UNODC, Global Study on Homicide 2023; UK Statistics Authority (2014); World Prison Brief (prisonstudies.org, accessed 2 November 2026). The policing summaries draw on Sherman and Weisburd (1995), Braga et al.'s Campbell reviews of hot-spots policing, Kelling and Wilson (1982), Sampson and Raudenbush (1999), Chalfin et al. (2022), Ariel et al. (2015), Yokum et al. (2019), Dee and Pyne (2022). Interactive numbers are computed live; the two-neighbourhood and funnel models use invented parameters.