Experimental evidence for the Broken Window Theory

In the late 80's, New York experienced a high rate of violence and crack was everywhere. In 1985 when George L. Kelling, coauthor of the article "Broken Windows", was hired as a consultant to the New York City Transit Authority, the subway was awfull. Kelling implemented new measures. He made every graffiti disappear and cleaned every station. Day after day after day, new graffitti would be made in the night and removed during the day, until oneday the new policy started to be successful and graffiti progressively disappeared. Mayor and police department of New York also employed the same method, they implemented a zero tolerance policing with easier arrestee procedure. Police started enforcing the law very strictly, against subway fare evasion, public drinkers, urinators, and the like. The rates of both petty and serious crime fell suddenly and significantly.

New York crime and drug decline is one of the best example of a successful implementation of the Broken Window Theory (BWT). BWT states that signs of disorder, like graffiti, dirty streets, broken windows… induce more disorder. Not only more graffitti and other petty crimes, but also more serious crimes like murder, robbery, etc. Consequently, removing the minor signs of disorder is thought to induce a decrease in the amount of more serious crimes.

Figure: Minor signs of disorder in Paris, France (author: Jean-noël Lafargue).

The BWT has been implemented in many cities around the world, with some success, but until now, the causal arrow leading from minor crime to more serious ones has remained highly speculative.

In a recent paper, Kees Keizer, Siegwart Lindenberg and Linda Steg conduct insightful and delightful field experiments to assess the BWT. I'll detail just one example to give you the flavour of the six experiments. In one setting they looked at whether individuals would steal an envelope visibly containing a five euro note. "The white (addressed) window envelope sticking out of a mailbox (situated in Groningen) was very noticeable for everyone approaching the mailbox, and it was clearly visible that the envelope contained a €5 note". In the baseline condition the mail box and the ground surrounding it were clean. In one test condition the mail box was covered with graffitti and in another the ground was covered with litter.

The results were quite dramatic, the rate of robbery doubled between the baseline and the "disorder" conditions! In the baseline condition, 13% of passer-bys stole the envelope, with graffitti this rate raised to 27% and with litter to 25%.

The authors conclude: "There is a clear message for policymakers and police officers: Early disorder diagnosis and intervention are of vital importance when fighting the spread of disorder. Signs of inappropriate behavior like graffiti or broken windows lead to other inappropriate behavior (e.g., litter or stealing), which in turn results in the inhibition of other norms (i.e., a general weakening of the goal to act appropriately). So once disorder has spread, merely fixing the broken windows or removing the graffiti may not be sufficient anymore. An effective intervention should now address the goal to act appropriately on all fronts."

Abstract of the paper:

Imagine that the neighborhood you are living in is covered with graffiti, litter, and unreturned shopping carts. Would this reality cause you to litter more, trespass, or even steal? A thesis known as the broken windows theory suggests that signs of disorderly and petty criminal behavior trigger more disorderly and petty criminal behavior, thus causing the behavior to spread. This may cause neighborhoods to decay and the quality of life of its inhabitants to deteriorate. For a city government, this may be a vital policy issue. But does disorder really spread in neighborhoods? So far there has not been strong empirical support, and it is not clear what constitutes disorder and what may make it spread. We generated hypotheses about the spread of disorder and tested them in six field experiments. We found that, when people observe that others violated a certain social norm or legitimate rule, they are more likely to violate other norms or rules, which causes disorder to spread


"Broken Windows" by James Q. Wilson and George L. Kelling, Atlantic Monthly, March 1982

"The Spreading of Disorder" Kees Keizer, Siegwart Lindenberg, and Linda Steg, Science, December 2008 It can also be found here.


  • comment-avatar
    Kate Devitt 17 December 2008 (03:23)

    The experimental results show that petty crime can be increased. But, I worry about extending such results to the more substantive thesis that increased petty crime increases serious crime, such as murder and rape. I remember reading in Freakanomics (Levitt & Dubner, 2005) that the more likely cause of a reduction in serious crime in New York was the introduction of legalized abortion in the 70s. I don’t know if they’re right, but it is another perspective to take, and another reminder not to conflate correlation with causation. Intuitively there is a large psychological difference between petty crimes and substantive crimes, a difference that is reflected in our legal system. I wonder if it is a wowse-rish response to believe that scribbling on transport connects meaningfully with violent crime? Of course, we cannot test violent crime experimentally, which means we’re stuck observing petty theft, littering and other minor infractions in our research centres. Reference: Levitt, S & Dubner, S. (2005) Freakonomics. William Morrow.

  • comment-avatar
    guest guest 17 December 2008 (09:55)

    has anyone discussed using this as a diffuse, low intensity urban guerrilla tactic, i.e. causing somebody’s territory to become unstable by destabilising the behavior of its citizen with minor signs of disorder? what about delegitimizing an incumbent on the eve of a political election?

  • comment-avatar
    Olivier Morin 17 December 2008 (10:43)

    I share in Kate’s skepticism, although Levitt and Dubner’s debunking of the BWT is itself controversial. But if you don’t take Levitt and Dubner’s argument, consider the fact that another team considers that the downward trend in serious crime that was attributed to broken window policies could be predicted by a rudimentary regression to the mean model. And of course there is the crack trade, the fact that crime also abated in cities that did not adopt a broken window policy, etc.

    The funny thing about the paper by Keizer et al. is that, while parading as a scientific rehabilitation of BWT (that’s how it has been received in, e.g., The Economist) it contributes almost nothing to the actual controversy. We still do not know whether repairing broken windows has a significant impact on serious crime. All Keizer et al. do is illustrate the theory with experiments that do not adress the controversial question, i.e. the impact on serious crime in the real world. Given the political importance of adopting or rejecting zero-tolerance policies, we might want to trust real social scientists on these issues.

  • comment-avatar
    Ophelia Deroy 17 December 2008 (11:17)

    I had just read this paper – wondering whether it would cast light on the sudden spreading of disorder in recent riots – (Greece, Paris, etc.). It is true that in the 2006 riots, there was no ” disorder ” in the clean, smart areas of Paris, but, as Kate Devitt points out, disorder is not just petty crime (let’s say that breaking shopwindows and setting cars on fire is more than littering, or making noise late). The paper is not entirely convincing though, and I am not sure of its conclusion. Keizer’s inspiration is what he calls the ” Cialdini effect ” (also known as ” bandwagon effect ”) i.e. people’s tendency to reason (his word) “if a lot of people are doing A, it’s probably a wise thing to A” and to do what they observe others are doing (R.B. Cialdini, Psychometrika 72 (2), 263 (2007)). At first, it seems that what is at stake is a sort of negative version of it, i.e. people’s tendency to reason : ” it a lot of people are not doing A, it’s probably okay not to do A either. ” But the paper doesn’t then say much about the fact that this involve an explicit inference or an implicit one (or what ?).

    Still, the idea seems to be that the violation of the tested norm (do not litter, for instance) results from such an explicit inference : it’s already quite a reflexive one, for the inference starts with the ” observed violation ” of a certain nom (the perception of a conflict between a sign ” no graffiti ” and a wall full of graffitis). What the authors are concerned with though is that this inference results in people’s thinking they are entitled to violate another norm (or feeling they are entitled ?- it isn’t clear here). They call it ” cross-normative ” effect.

    Let’s then agree that the inference does not concern the observed action A per se (doing graffitis) but the fact that the action is a violation of the norm ” don’t do A ”. But what the experiments show actually is ambivalent between : (1) ” If people don’t respect norm A, then it’s okay for me not to respect norm B either”. (their ”cross-normative ” conclusion ) (2) ” If people don’t respect social norms (let’s call this S), then it’s okay for me not to respect social norms either ” (a more direct process, with no ” crossing ” involved). The experiments provide them with a special occasion to test their respect of social norms – like littering or not littering. But there is no need to say that it was the breaking of the ” no littering norm ” that was suddenly made OK by the seeing of another violation, let’s say of the ” no graffiti ” norm.

    Nor do they really provide very good arguments against the fact that people are doing a risk-evaluation (not thinking about the breaking of the norm being suddenly okay, but thinking it’s not too risky in the circumstane not to comply with the rule, still seen as the appropriate thing to do). So the inference could go the following way : (3) ” If people haven’t been caught doing A, then i won’t certainly be caught doing B ” (another cross-normative inference) (4) or ” If people haven’t been caught breaking social norms (S), then i won’t be caught either if I do so ” (without crossing)

    They try and answer this point, true, but not in all their experiments. In the experiment 1, they say that ” littering is not generally punished by the police, as graffiti are ” – but what does generally mean ? It’s quite different whether there is no sanction, or there can be, even though it is rare. And doesn’t it make the inference all the more tempting : ” if people haven’t been caught doing something usually punished, i probably won’t be caught doing something even less usually punished ” ? The fact that people could reason from evaluating probabilities of being caught disappears from experiments 2 to 4 – where it is much wanted. And at the end, they argue for a complicated story, far from the inferential mechanism initially at stake : ” people don’t necessarily copy the inappropriate behavior they observe but let concerns other than appropriateness take center stage ”. The story seems now to go as a des-inhibition process : it ” lets ” other concerns take center stage. Or perhaps do they mean that the inference is something close to (2) : (2’) ” If people let other concerns other than complying to social norms govern their behaviour, then it’s okay to do the same thing ” (in which case, it’s more like a Cialdini effect).

    I don’t mean to argue for the explicit inferential model in such spreading of disorder, nor to put as a constraint that the behaviour has to be copied, and the same kind of action to occur in antecedent and the consequent of the inference, or in observation. I have nothing against the idea there can be cases of less straightforward imitations, or transmissions, and I am certainly sympathetic also to the idea it is less explicit (at least in the first experiment, the fact that the place is not clean can influence behaviour in several ways – see for instance recent experiments by Schnall et al. ”Disgust and Embodied Moral Judgement” , Pers.Psych. Bull. (2008) : here ). I also don’t share some of the intuitions behind the experimental setting : is putting the flyer on someone’s bike a form of ” littering ” ? Do you break a norm if you give it to someone else of the community to do the right thing instead of you ? That would mean that lots of our actions are actually breaking social norms.

  • comment-avatar
    Nicolas Claidière 19 December 2008 (09:33)

    I agree with many of the comments which have been raised so far and concern the general issue of the implementation of the Broken Window Theory and its policing consequences. I think it is quite clear that the paper by Keizer, Lindenberg and Steg is not meant to prove the Broken Window Theory and to put an end to the debate surrounding the policy we should adopt regarding petty crimes and disorder. The paper is really more modest than that. But still, I think it brings interesting material and raises new interesting questions. Let me clarify what I believe is the contribution of the paper by answering a few questions.

    Does the experiment tell us anything about:

    1) the link between the implementation of the BWT in New York and the following drop of crime and drug rates?

    As pointed out by Kate and Olivier, the matter is quite complicated and many factors beside the BWT implementation may be involved. Indeed, in such a complicated, real life, large time and space scale case, we expect any change to result from the conjunction of several, intimately mixed, factors. In such a case, even if correlational and main component analysis are always possible, they won’t prove anything, they’ll just point to some of the factors which may, or may not, be causally relevant. But then, this should prompt us to perform experiments in which the potential causally efficient factors are studied in isolation, with rigorous testing. I believe this is just what Keizer, Lindenberg and Steg have done. In a modest way, I agree, they haven’t proven the theory, no doubt, but they have brought evidence supporting it, which is great.

    2) the broken window theory?

    I certainly agree, as do the authors of the paper would probably do, that there is a huge gap between stealing a five euro note and stealing a bank. But there is also quite a difference between littering and stealing a five euro note. One may consider, for simplicity, because obviously we can make many more categories, that there are three different categories. Say: disorder (e.g. littering), petty crimes (e.g. stealing five euros) and serious crimes (e.g. stealing a bank). The BWT, as I understand it, assumes that there is a causal link throughout all three categories. The present experiments show that there is a causal influence of disorder on petty crimes. This may, or may not, extend to more serious crime, I agree. But as pointed out by Kate, it is quite difficult to experiment with serious crimes, we are therefore left with correlational analysis. Showing that disorder has an effect on petty crimes is not a definitive argument but lends support to the idea that petty crimes may influence more serious ones.

    3) the larger effects at the community level?

    The article definitely does not address this issue, but one way to go, as suggested by Stephano, would be to test whether minor signs of disorder destabilizes the behaviour of the local citizens. This was indeed the first idea that Kelling put forward in his original paper of 1982 (see link in the post).

    4) the psychology of disorder?

    How does disorder psychologically influence petty crimes? Ophelia raises interesting questions about the psychological processes that may be involved. Although I think the claim that there is a causal link between disorder and petty crimes is not questioned by the view we adopt on how this link is psychologically realized, this is an important issue following from this research.

    5) the policy we should implement?

    It is true that Broken Window Theory has been associated with zero tolerance policy and I agree that it would have been nice to insist on the independence of the two in the paper to prevent misreading. But the BWT is a claim about what people do, not what we ought to do about it. In fact, it is interesting to read the original paper of Kelling in which he militates against a zero tolerance policy. In that paper he explains that a community has its own relative rules defining what one can or can not do and that policies should be made so as to preserve these rules. It is very clear in the paper that the rules are only local and should be only locally preserved. If I remember well he also explains that BWT does not imply that we need to legislate about every minor theft. Quite the opposite, what we ought to do, he insists, is to restore foot patrols of policemen because those policemen progressively come to understand and enforce the local rules of the community to potential violators (often non aggressively he observes). Of course this could also bring other difficulties, you have to be extra careful when you give such responsibilities to someone. Anyhow, I think this shows the connections between BWT and policy are quite loose and the message ”take care of broken windows” a very reasonable one.

  • comment-avatar
    guest guest 19 December 2008 (18:16)

    The leap from low-level vandalism and graffiti to rape and murder is a large one. But, as Ophelia suggests, the leap to disorder, setting fire to cars, looting etc isn’t so great. And if these get out of hand, people get hurt and even die. Personal anecdotes aren’t reliable scientific indicators, but here’s something that happened to me. I was working late in west London. I locked up my office and went to my car (an old-style Mini), around 1.00 am. I found that all my car windows had been smashed. A Police car drove past, I caught the officer’s attention, and he came over to me. I felt angry, but also exhausted. Could I just go home and deal with the mess in the morning? Absolutely not, he told me. By morning, my car would probably have been set on fire. So I had to get towed, in the rain, sat in a car with no windows, to a garage that would take the car at 3.00 am. Nice. The Policeman suggested what had happened, based on his local experience. First, a thief broke one window. Typically a thief breaks a window without attempting to get into the car. He comes back five minutes later, to check if anyone has reacted to his initial broken window. No one had, but he found nothing in the car worth stealing. Later, others walked past the car, saw the broken window, and also noticed that no one had reacted to it. So they joined in, smashing more of my windows. The motive was not theft, but the sheer pleasure of destruction. The point about the Cialdini Effect is that people see evidence that other lawbreakers are “getting away with it” – that such destructive actions are not being deterred or punished. Therefore the inhibition on destructiveness is reduced.