One unrepaired window tells everyone nobody's watching — and that signal travels a lot further than a city block.
Criminology & Social Systems · Organizational CultureBroken Windows Theory holds that small, visible signs of disorder or neglect — a broken window left unrepaired, litter that nobody picks up, a minor rule violation nobody corrects — signal that no one is watching or that nobody cares, which lowers the perceived cost of further disorder and invites more of it, sometimes escalating into much more serious problems.
"One un-repaired broken window is a signal that no one cares, and so breaking more windows costs nothing."
Political scientist James Q. Wilson and criminologist George L. Kelling introduced the idea in their article "Broken Windows," published in The Atlantic Monthly in March 1982, partly illustrated by a retelling of psychologist Philip Zimbardo's 1969 abandoned-car field study. New York City's Transit Police under William Bratton began applying it to subway graffiti in 1990, and Bratton carried the same "quality of life" approach into the NYPD under Mayor Rudy Giuliani starting in 1994; a 2001 study by Kelling and William Sousa credited it with part of the city's crime decline. The theory has been seriously contested since: Robert Sampson and Stephen Raudenbush's 1999 research in Chicago (American Journal of Sociology) found a neighborhood's "collective efficacy" — not visible disorder itself — better predicted crime, and a 2015 meta-analysis by Anthony Braga, Brandon Welsh, and Cory Schnell found that community-focused disorder policing produced modest reductions while aggressive, zero-tolerance enforcement of individual minor offenses (the approach behind the stop-and-frisk controversies of the 2000s) did not. The cleanest surviving experimental support actually comes from outside policing altogether: a 2008 Dutch field study published in Science.
Use this compass to orient yourself before you reach for the model. It tells you what field it comes from, what it travels with, and where the real tension sits.
Criminology and urban sociology in origin, but it now functions as a general theory of how a small, visible signal of neglect communicates "no one is watching" — used well outside policing, in software engineering, management, civic design, and personal habits.
The Matthew Effect (a small early advantage compounds into a large one) and Path Dependence (an early condition shapes everything that follows) — all three describe how a small starting signal snowballs, just in different directions.
Fixing the literal window — the physical or environmental signal — is cheap, fast, and reasonably well supported. Using the theory to justify aggressive enforcement against the people standing near the window is where it becomes most contested, least supported, and most prone to misuse.
Deciding where to spend small, fast maintenance effort — in a codebase, a shared space, a team's norms, or a public square — before a visible lapse compounds into something far more expensive to undo.
Work through these four steps in order. Each one builds on the last.
0 of 4 steps opened
A window stays broken for weeks. Graffiti sits on a wall unremoved. A small rule — no bikes locked to that fence, no spam in this chat, no skipped code review — gets violated once and nobody says anything. None of these events cause serious harm by themselves. What they communicate is the real event: that whoever is supposed to be maintaining order here either hasn't noticed or doesn't care enough to act.
Wilson and Kelling's claim was never that disorder directly causes serious crime. It's that visible, unaddressed disorder changes what observers believe about the odds of being noticed or corrected. Their 1982 article leaned on a retelling of a 1969 field study by psychologist Philip Zimbardo: an abandoned car left intact for days in a calm Palo Alto neighborhood, versus one in the Bronx that was stripped within 24 hours — until Zimbardo smashed one of the Palo Alto car's windows himself, after which passersby joined in and dismantled it within hours. The story is vivid, but later scholarship (historian Bench Ansfield's 2020 study in American Quarterly) found that Wilson and Kelling's retelling oversimplified Zimbardo's actual methods and findings. Treat the car story as a memorable illustration of the proposed mechanism, not as rigorous proof of it — the real evidence lives elsewhere.
New York City's "quality of life" policing in the 1990s, and Kelling and Sousa's 2001 study crediting it with part of the city's crime drop, made broken windows famous. But Sampson and Raudenbush's 1999 Chicago research found that a neighborhood's collective efficacy — whether neighbors trust each other enough to intervene — predicted crime better than how much visible disorder it had, and other researchers pointed to confounding factors like the national crime decline, a stronger economy, and other policing changes happening at the same time. The cleanest supporting evidence comes from psychologists Kees Keizer, Siegwart Lindenberg, and Linda Steg, who ran real field experiments in Groningen, the Netherlands, published in Science in 2008: in an alley with a graffiti-covered wall, 69% of cyclists littered a flyer stuck to their handlebars, versus 33% at the same alley with a clean wall; when bikes were already improperly locked to a fence, 82% of people violated a posted "no entry" rule by cutting through anyway, versus 27% when the fence was clear. A 2015 meta-analysis by Anthony Braga, Brandon Welsh, and Cory Schnell later found that policing strategies which cleaned up physical and social disorder through community problem-solving produced modest crime reductions, while aggressive, zero-tolerance enforcement against individuals for minor offenses did not.
The well-supported part of the theory is environmental and fast to act on: fix the small, visible signal of neglect quickly, because the signal itself — not necessarily the specific violation — is what spreads. The poorly supported, and most criticized, part is treating the theory as license to crack down harder on the people near the disorder instead of on the disorder itself. Keep those two things separate in any domain you apply this to, and you keep the useful 20% of the idea without inheriting its most damaging misuse.
Same signaling pattern, six different rooms. Click a tab.
Andy Hunt and Dave Thomas named a chapter of The Pragmatic Programmer (1999) "Software Entropy" and used broken windows directly: a codebase with one obvious, unaddressed mess (a bad hack, a failing test left skipped, inconsistent naming) signals that quality standards here are optional, and other contributors start leaving their own messes too. Teams that fix the first small lapse the day it appears tend to keep the whole codebase cleaner than teams that let "we'll clean it up later" accumulate.
New York City's Transit Police began an aggressive subway graffiti-removal program in 1990 under William Bratton, cleaning each car before it returned to service rather than letting tagged cars keep running. The contested part came later, when the same logic was extended into aggressive stop-and-frisk enforcement of minor offenses — the piece of the theory's legacy that drew the most criticism and the least empirical support.
A manager who lets one missed deadline, one sloppy report, or one person showing up visibly late to every meeting go unaddressed finds that standard quietly becomes the team's real norm, regardless of what the written policy says. Correcting the first small slip — calmly, quickly, and consistently — keeps everyone else's bar where it's supposed to be.
A storefront with a cracked sign, a dead window display, and trash piling up outside reads as unsupervised and unwanted — which is exactly the effect Keizer, Lindenberg, and Steg measured experimentally: littering roughly doubled (from 33% to 69%) at an alley with graffiti on the wall compared with the same alley clean. A landlord or property manager who fixes the obvious cosmetic damage within days, not months, keeps the signal from compounding into vacancy and vandalism.
In a forum, Discord server, or group chat, the first spam message, slur, or rule-breaking post that sits unremoved for hours tells every other member the posted rules aren't actually enforced. Communities that remove the very first violation quickly, visibly, and consistently see far less follow-on rule-breaking than ones that let the first instance slide "just this once."
One unwashed dish left in the sink, or one pile of mail left unsorted on the counter, tends to attract a second and a third far faster than an empty counter does — the same signal-of-neglect mechanism operating at the scale of a single household. Fifteen seconds spent handling the first stray item is often cheaper than the hour it takes to deal with the pile it would otherwise become.
This one uses real published numbers, not an illustrative estimate. Keizer, Lindenberg, and Steg ran a field experiment in a Groningen, Netherlands bicycle alley: researchers stuck a flyer on parked bikes' handlebars, then watched whether riders littered it on the ground or carried it away, comparing a wall with graffiti against the same wall freshly cleaned.
Why the gap runs this direction: nothing about the alley's actual rules or supervision changed between conditions — only the visible presence of one unaddressed norm violation (the graffiti) did. The researchers' broader point, confirmed across six separate experiments in the same paper, is what they called "cross-norm inhibition": once people observe one norm being violated and unaddressed, they become measurably more likely to violate a completely different, unrelated norm themselves. The littering isn't caused by the graffiti directly — it's caused by what the graffiti implies about whether anyone is enforcing anything here at all.
Pick the situation you're facing, and build a starting prompt you can take to any AI assistant to help you apply this model well — and avoid its most common misuse.
Keep this model on hand as a reusable skill for yourself or an AI assistant.