Explaining the Crime Drop

Beginning in the early 1990s, crime fell across the rich world — a large, sustained, international decline that continued for two decades and shows in both recorded crime and victimisation surveys (see 10.2.1), so it is not a recording artefact.

Almost nobody predicted it. Few can fully explain it. And it is one of the most instructive episodes in the discipline — not only for what it says about crime, but for what it says about how little the social sciences can predict, and how confidently they nonetheless explain after the fact.

The forecasts, and what happened instead.

In the early 1990s, the expert consensus in the United States was that crime was about to get much worse.

Prominent criminologists warned of a coming wave of youth violence. A demographic bulge of young people was arriving; a term was coined — the "super-predator" — for a generation of remorseless young offenders that was said to be on its way; and the projections were for a bloodbath. These were not fringe voices. They were leading figures, and the forecast shaped policy — harsher sentencing, more prisons, the trying of juveniles as adults.

The opposite happened.

Violent crime in the United States fell, and kept falling, for roughly two decades. Homicide — the anchor measure (see 10.2.1), so this is not a recording effect — roughly halved. Property crime fell further. And the fall was not confined to the country that had predicted a wave. It appeared, at broadly the same time and to broadly similar degrees, across most of the rich world — in countries with completely different policing, different incarceration rates, different drug markets, different demographics and different economies.

Read the two facts together, because their combination is the whole puzzle.

The forecast was confidently wrong — by capable people, using reasonable-looking demographic reasoning, in the direction of catastrophe. And the thing that actually happened was international, which means the explanation cannot be any single country's policies, because the same fall occurred in countries that did the opposite of each other.

A country that tripled its prison population and a country that did not both saw crime fall. A country that adopted aggressive policing and a country that did not both saw crime fall. Whatever caused it operated across borders, across policy regimes, and across political systems — which rules out most of the explanations offered for it, because most of them are national.

And here is the part that should stay with you. After the fall, dozens of confident explanations appeared, each claiming the credit — for a policy, a policing method, a leader. Almost all of them are national explanations for an international event, and almost all of them are therefore wrong, or at most a small local contribution to a global pattern. The crime drop is a graveyard of confident causal claims, and reading it is training in how to hold one.

An international effect requires an international cause, and that constraint eliminates most candidates.

This is the single most useful analytical move in the lesson, and it is 7.5.4's comparative logic applied ruthlessly.

If the crime drop happened in many countries at once, at similar times and magnitudes, then any explanation must be something that changed in all of those countries at once.

Which immediately eliminates: this country's sentencing reform, that country's policing strategy, the other country's drug policy, a particular leader, a particular law. These are national, and the event was international, so at most they are local contributions — they might explain why the fall was slightly larger here than there, but they cannot explain the fall itself, because it also happened where they were absent or reversed.

What survives the test is a short list of things that changed across the rich world in roughly the same period: technology, markets, environmental exposures, cohort characteristics, and broad economic and social shifts. The candidates that pass the international test are the ones worth taking seriously, and even they are contested.

And the meta-lesson: the confidence of an explanation is unrelated to its survival of this test. The loudest claims — that a policy or a politician caused the drop — are precisely the ones the international pattern refutes, and the surviving candidates are the ones nobody campaigns on because they are undramatic and give no one credit.

The candidates, tested

The national explanations that mostly fail the test.

More police and more prison. The United States massively expanded incarceration over the period, and it is intuitive to credit this. The evidence is that it contributed modestly at most, and cannot be the main cause, for two reasons: the fall happened in countries that did not expand incarceration (Canada's crime fell like America's with stable imprisonment), and the incapacitation effect shrinks sharply as imprisonment expands (locking up the highest-rate offenders reduces crime; locking up ever-lower-rate offenders reduces it far less, at rising cost). A national policy that also occurred where it was absent is not the international cause.

Better policing. Particular policing innovations — data-driven deployment, focused strategies — spread in this period, and some have real evidence of local effect. But they were national and various, and the fall was international and uniform, so they are at most a contributor. The strongest claims made for specific policing methods have not survived scrutiny of the size the advocates claimed.

The economy. Crime did not track the economy in the expected way — it kept falling through recessions, which a pure economic explanation cannot accommodate. The relationship between the economy and crime is real but weak and complicated (see 8.6.2 on the same problem for the dependency thesis), and cannot carry the drop.

A demographic shift. The aging of the population reduces crime (offending is heavily concentrated in youth). But the timing and size do not fit — the fall was too large and too synchronised to be demographics alone, and it fell among the young too.

These are the explanations most often offered, and the international test demotes all of them to contributors.

The candidates that pass the international test — and are still contested.

Environmental lead. The most striking of the international hypotheses. Leaded petrol was introduced across the rich world and phased out across the rich world, on a broadly similar timetable — and lead exposure in early childhood impairs the developing brain in ways associated with impulsivity and aggression. The lag between the phase-out and the crime drop is roughly a generation — the time for lead-free children to reach the peak offending ages — and the timing fits across multiple countries.

The evidence is genuinely suggestive and genuinely contested. The cross-national timing correlations are striking; some quasi-experimental studies using variation in lead exposure find effects on later behaviour; and the mechanism is biologically plausible. And the criticisms are serious: much of the evidence is ecological (see 7.6.4 — correlations at the population level), the effect sizes needed to explain the whole drop are large, and the studies vary in quality. The honest position: a real candidate that passes the international test, with suggestive but not decisive evidence, probably a genuine contributor of uncertain size.

Security improvements. The most mundane and possibly the most important. Cars, homes and goods became much harder to steal across the rich world at roughly the same time — immobilisers in cars, better locks, alarms, and the falling value and rising traceability of stealable goods. The "security hypothesis" has strong evidence for the property-crime fall specifically: vehicle theft, which is well measured (small dark figure — see 10.2.1), fell dramatically and tracks the spread of immobilisers closely. This passes the international test — the security technology diffused everywhere — and has better evidence than most for its part of the drop. Its limitation is that it explains property crime better than violence.

And a cluster of broad social changes — the maturing of drug markets after the crack epidemic (which passes for the US but not everywhere), the spread of mobile phones and later the internet changing how young people spend time, and cultural shifts in youth behaviour (young people across the rich world drink less, fight less and offend less than previous cohorts, for reasons not fully understood). These are real, international, and hard to measure precisely.

The honest synthesis: the crime drop was probably caused by several international factors operating together — security improvements (strong for property crime), possibly lead (suggestive), changing youth behaviour and technology (real but hard to quantify) — with national policies contributing at the margins. No single factor explains it, the field does not agree on the weights, and the confidence of any monocausal explanation is a sign it is wrong.

The long view

The crime drop sits inside a much larger and even more surprising finding.

Violence, measured by homicide over the very long run, has fallen dramatically — by orders of magnitude from the medieval period to the present in the societies with the best records. Estimated homicide rates in medieval Europe were many times — in some estimates dozens of times — higher than today's, and the decline is one of the best-documented long-run trends in social science, visible precisely because homicide is the anchor measure (see 10.2.1).

The explanations offered are large and contested. A long civilising process (see the Elias tradition) in which the internalisation of self-restraint spread; the monopolisation of violence by states, which suppressed private feud and vengeance; the growth of commerce, which made cooperation more profitable than predation; and the strengthening of institutions of law. Each is plausible, none is proven, and the honest report is that the decline is far better established than any explanation of it.

And this is the frame that should hold the crime drop. The recent fall is a small, recent wobble on a very long downward line. Which means the intuition that the world is becoming more violent — held by most people in most rich countries, most of the time — is, on the best evidence, precisely backward (see 9.6.1 on the mean-world perception; the media environment produces the belief, the data refutes it). We live in one of the least violent periods in recorded human history, and almost nobody believes it.

Two cautions this course must give. The long-run decline is a claim about interpersonal violence measured by homicide, not about all harm — it does not count war deaths in the same series, and the twentieth century's state violence is a separate and enormous story. And the trend is not guaranteed to continue — long declines can reverse, and reading a downward line as a law would be exactly the error 7.9.1 warns against. The finding is that violence has fallen, robustly and surprisingly; it is not that it must keep falling.

Because the crime drop is the clearest available lesson in the limits of social prediction and the cheapness of after-the-fact explanation.

The forecast was confidently, catastrophically wrong, made by capable people using reasonable reasoning, and it shaped harsh policies whose justification evaporated. And the confident explanations that followed the actual fall are mostly national explanations for an international event — which the comparative test refutes, and which persist because they give someone credit.

Three questions to carry.

Is the explanation national or international? For any account of the crime drop, ask whether the proposed cause changed in all the countries where crime fell. If it is national, it cannot be the main cause of an international event — this single question demolishes most of the confident claims.

Does it fit the timing and the offence? Security improvements fit property crime and the timing well; lead fits the generational lag; economic explanations fit the recessions badly. A cause has to match the shape of the thing it explains.

And what is the base rate of the intuition? Most people believe crime is rising when it is falling and that the world is more violent than ever when it is among the least violent in history. The gap between the perception and the data is itself a finding — about the media environment, and about how badly intuition tracks aggregate social change.

One closing observation, which is the deepest lesson of the Part so far. The super-predator forecast was not a random error. It was produced by taking a real risk factor — a demographic bulge of young men — and reasoning forward from it with confidence, ignoring everything that could intervene. It is 7.6.2's lesson (a risk factor is not a destiny) and 7.6.3's (do not update strongly on a projection) and 7.9.1's (hold beliefs in proportion to evidence) — all failing at once, in public, with policy consequences.

And the honest response to the crime drop is not to reach for the explanation that flatters your politics. It is to say: crime fell, internationally, for reasons that are partly understood and contested; no single factor explains it; and the episode should make everyone — forecasters and explainers alike — considerably more humble about how much of social change anyone can see coming or account for afterward. That humility is not a failure of the discipline. It is one of its more mature achievements.

From the early 1990s, crime fell across the rich world for two decades, in both recorded and survey measures — and the expert forecast had been the opposite, a coming wave of youth "super-predators" that shaped harsh policy and never arrived.

The decisive fact is that the fall was international — the same drop in countries with opposite policies on incarceration, policing and drugs — which means the cause must be international, and most offered explanations are national and therefore refuted or demoted to contributors.

The national explanations mostly fail the test. More prison (occurred where absent; incapacitation shrinks as it expands); better policing (national and various, event international and uniform); the economy (crime fell through recessions); demographics (too large and synchronised). Each is at most a marginal contributor.

The candidates that pass the international test are still contested. Environmental lead — leaded petrol phased out across the rich world a generation before the drop, biologically plausible, suggestive but largely ecological evidence. Security improvements — cars and goods made harder to steal everywhere at once, strong evidence for the property-crime fall (vehicle theft, well measured, tracks immobilisers), weaker for violence. And broad social and technological change in youth behaviour, real and hard to quantify. The synthesis: several international factors together, weights disputed, no monocausal account credible.

The drop sits inside a much larger finding: homicide has fallen by orders of magnitude over the very long run — medieval rates many times today's — explained by contested large theories (civilising process, state monopoly of violence, commerce, law), with the decline far better established than any explanation. Which means the near-universal intuition that the world is more violent than ever is, on the best evidence, backward — with the cautions that this measures interpersonal violence not war, and that the trend is not guaranteed to continue.

And the episode's deepest lesson is about prediction and explanation: a confident forecast catastrophically wrong, and confident after-the-fact explanations mostly refuted by the comparative test — training in the humility the field earned the hard way.

The crime drop — the large, sustained, international fall in crime from the early 1990s, visible in both measures.

The super-predator forecast — the confident, catastrophically wrong 1990s prediction of a youth crime wave.

The international test — the requirement that an explanation for an international event be something that changed across all the affected countries.

Incapacitation effect — the crime reduction from imprisoning offenders; shrinks sharply as imprisonment expands to lower-rate offenders.

The lead hypothesis — the claim that the phase-out of leaded petrol reduced later offending a generation on; suggestive, largely ecological.

The security hypothesis — the claim that harder-to-steal cars and goods drove the property-crime fall; strong for property crime.

Long-run decline in violence — the order-of-magnitude fall in homicide over centuries; robust, its explanation contested.

Civilising process — the proposed spread of internalised self-restraint; one contested explanation of the long decline.

Mean-world perception — the belief that violence is rising when it is falling (see 9.6.1).

One — apply the international test. Take any explanation you have heard for the crime drop and ask whether the proposed cause changed in every country where crime fell. Most fail immediately.

Two — check the forecast. Look up the early-1990s crime predictions for the United States and compare them with what happened. Note that capable people made them with confidence.

Three — separate property and violence. For the security hypothesis, check whether it fits property crime better than violence. A cause should match the shape of what it explains.

Four — find your intuition's base rate. Before looking, guess whether violent crime in your country is higher or lower than a generation ago. Then check.

Five — take the long view. Look up estimated medieval homicide rates and compare them with today's. Then reconcile the number with your sense of whether the world is getting more dangerous.