In 1969, a Stanford psychologist named Philip Zimbardo parked two cars on two streets. Both were the same kind of car in the same condition: no license plates, hood up, plainly abandoned. One went to a block in the Bronx. The other went to Palo Alto, California. The difference in what happened next is the reason you have heard the phrase “broken windows,” and it is also the reason to handle that phrase carefully.
🧠 Two Cars, Two Streets
The Bronx car was attacked within minutes. The first people to arrive were not, as Zimbardo had half expected, a pack of teenagers. They were a family: a father, a mother, and a young son, who removed the radiator and the battery. Within twenty-four hours everything of value had been stripped from the vehicle. Then the windows went, then the parts, then the upholstery, and eventually children were using the shell as a playground.
The Palo Alto car sat untouched for more than a week. What finally broke the stillness was Zimbardo himself, walking up with a sledgehammer and taking out a window in front of passersby. People joined in soon after. That last detail is contested, though: the car had by then been moved onto the Stanford campus, and the first people to follow him were reportedly his own students.
His note about who actually did the damage is the part that dissolves the easy explanation. Most of the adult vandals in both cities, he recorded, were well dressed, clean-cut, and entirely respectable looking. Vandalism was not tracking character. It was tracking what the street appeared to permit.
⚙️ How It Works
In March 1982, James Q. Wilson and George L. Kelling published an essay in The Atlantic called “Broken Windows” that turned the observation into a theory. Their central image: if one window in a building is broken and left unrepaired, the rest will soon follow, because a single unrepaired window is a signal that no one cares. And once no one cares, breaking another one costs nothing.
The mechanism is a loop rather than a straight line, and it runs in five steps.
One. The setting is anonymous. On a street full of strangers, you cannot read the local rules off other people, and you cannot tell how likely anyone is to hold you to them. So you read the environment instead. People take their cue from the place before they decide how to act. That is the premise of everything that follows.
Two. Disorder broadcasts an absence of responsibility. A window nobody fixed, litter nobody swept, graffiti layered three times over. The message is not “crime is allowed here.” The message is “nobody is in charge here.” That distinction is the whole thing, and it runs on the same switch that diffuses responsibility in a crowd.
Three. The cost of stepping over the line gets repriced. Once “nobody is in charge” has been established, the odds of being seen and held to account feel lower. The first person acts.
Four. The first act leaves fresh evidence. The car is more broken. There is more litter. There is another layer of graffiti. Each one is a new signal that nobody is in charge.
Five. The loop feeds itself. A neighborhood does not collapse in a single event. It collapses as a chain that keeps fueling its own next link. Which is exactly why the prescription is not a grand cleanup but a fast repair: Kelling and Coles suggested fixing the window within a day, a week at the outside.
🤔 The Twist
The small mechanism has real experimental support. In controlled field experiments run in 2007 and 2008, Kees Keizer and colleagues at the University of Groningen arranged city locations to be either orderly or visibly disorderly, then watched secretly to see who behaved differently. Their conclusion, published in Science, is blunt: one kind of disorder, like graffiti or litter, really does encourage another, like taking something that is not yours.
The large claim did not fare nearly as well. In 2007, Bernard Harcourt and Jens Ludwig showed in a paper titled “Reefer Madness” that regression to the mean explains the precinct-level crime changes in 1990s New York all by itself. In other words, some of the most cited evidence for the policy was simply the data settling back after a spike. In a 2006 paper in the University of Chicago Law Review the same two authors pointed out that the violent crime decline began in 1990, under Mayor David Dinkins, before the new policing took hold, and that crime was falling across the entire United States in cities that never adopted the approach. They also followed a federal program that moved public housing tenants into more orderly neighborhoods. The theory predicts less crime from those tenants. They committed crimes at the same rate.
Then there is the cleaner explanation. Robert Sampson and Stephen Raudenbush argued that disorder and crime are not a causal chain at all, but two outputs of a shared upstream factor they called collective efficacy: cohesion among residents, plus shared expectations about controlling public space. Their follow-up work went further. What counts as disorder is itself socially and racially encoded in the eye of the observer, which means it cannot function as a neutral, objective input to the model.
One footnote about the theory’s own sources is worth carrying around. Bench Ansfield has shown that the 1982 Atlantic essay rested its central claim on exactly one source: Zimbardo’s abandoned car. Wilson and Kelling did not merely lean on it. They removed Zimbardo’s own framing about inequality and anonymity and kept the slogan.
🔗 Why It Matters To You
The loop is the useful part, and it maps onto any shared space, including digital ones. An unmaintained error page, a comment section that will absorb anything, a “What’s New” section frozen in the past year. Each one says what the unrepaired window said. And it compounds: the first piece of junk invites the second. The cheap intervention is not the quarterly cleanup, it is response speed. Remove the first one, visibly, and the loop never gets its second turn.
For diligence work, the theory is a warning worth taping to the wall. Walking a factory floor and reading the state of the equipment, the storefront, the office kitchen is genuinely informative. Sampson’s point is that the story people tell about it is usually wrong. It is probably not “a messy floor causes bad outcomes.” It is more likely “wherever management attention is absent, the floor gets messy and other things go wrong too.” Treating the mess as the cause produces a confident, well-organized inference pointing the wrong way. That is the correlational slide waiting inside every site visit.
For fiction, the loop is a gift. A city or a family does not need a catastrophe to fall apart. It needs one window nobody fixes, and then each chapter adds one more piece of evidence. What makes a collapse feel inevitable is visibility, not scale: a reader can see that every link followed from the last one, and still cannot point to the moment the whole thing started.
🎲 The Bonus
The researcher who parked the car became the first vandal in his own study, and the part of the story where strangers join him may have unfolded on a college campus, with his own students going first. Meanwhile the essay that reshaped American policing leaned its central claim on that single, informally documented car. One metaphor carried thirty years of policy.
An exercise: find the window you have not fixed. The email you keep meaning to answer, the small bug you have rescheduled three times, the folder you will sort out later. It is probably not sitting there because it is hard. It is sitting there because it keeps saying nobody is in charge.