Broken Windows Theory

Broken windows theory is the hypothesis, published by James Q. Wilson and George Kelling in 1982, that visible disorder — smashed windows, graffiti, public drinking — signals that nobody is watching, weakening informal social control and inviting serious crime. The policy inference is that attending to minor disorder prevents major offenses. The metaphor did much of the persuading, which is itself a lesson about Naming.

The supporting case has two parts. Field experiments, most prominently a 2008 series in Groningen, found that people littered and stole more in settings seeded with disorder cues — direct evidence for the psychological mechanism. Separately, several analyses of order-maintenance policing report associated crime reductions. Kelling maintained that the theory prescribed restoring order alongside residents, not mass pedestrian stops.

The critical case is equally specific. Reanalyses of the Moving to Opportunity experiment, which effectively randomized neighborhood disorder, found no effect on individual offending, and some replications of the cue experiments came out weak or null. The larger problem is attribution: violent crime fell sharply across nearly every large American city in the 1990s, and in several other countries, including places that never adopted the policy. The counterfactual is unrecoverable, and rival explanations — incarceration, demographics, the crack market's decline, lead abatement — fit the same curve.

The implementation costs are documented separately and little disputed. Enforcement-heavy programs concentrated stops and arrests on young men in poor and minority neighborhoods, producing court debt and records with long consequences, and one city's stop-and-frisk program was ruled unconstitutional as practiced. A screen with a low hit rate puts most of its burden on innocent people, a question of Sensitivity and Specificity; and once stop counts became the performance measure, Goodhart's Law did the rest. A hypothesis compatible with every outcome is weak on Falsifiability before anyone disputes its data.