© Daniel Nettle, CC BY http://dx.doi.org/10.11647/OBP.0084.04
Broken glass, everywhere.
People pissing on the stairs; you know they just don’t care.
Where the previous chapter was about helping one another, this chapter is about the harms we can do. That is, we are concerned in this chapter with neighbourhood patterns of antisocial acts, acts that are to the detriment of another person or group. Prosocial and antisocial behaviour are logically distinct but nonetheless connected. It is not logically necessary that when people are less inclined to help each other, they will be more inclined to do each other harm, but in practice this often seems to be true. We have two sets of reasons for expecting there to be a greater prevalence of antisocial behaviour in Neighbourhood B than A.
First, there is the edge principle discussed in the previous chapter. I claimed that people in Neighbourhood B tend on average, as a consequence of deprivation, to be closer to the existential edge than those in A, and hence to be more likely to be pushed to the edge itself. When you are on the edge, risky options for solving your immediate problem start to look attractive, because their payoff is often immediate, and their best possible payoff is often enormously advantageous to the actor (though since they are risky, their best possible payoff is not their most likely payoff). The kinds of risky options that might present themselves to people in difficulty in a modern city include lying, robbery, stealing from individuals, or coercing others to change their behaviour. These are all antisocial acts. Indeed, to act antisocially can be considered as putting a higher valuation on the short-term maximal payoff of the act than on the possible negative consequences, some of which might be subtle (loss of reputation or friendship, for example), and many of which are deferred in time.
The second reason for expecting greater antisocial behaviour in B is empirical. Previous research has established very clearly that neighbourhoods where social capital is low are also those where crime tends to be high, and crime is the archetypal antisocial behaviour (Sampson et al., 1997). Crucially, social capital in these studies is measured in much the way we measured it in Social Survey 1, via self-report surveys. Given the neighbourhood difference in self-reported social capital reviewed in chapter 3, the clear prediction should be that antisocial behaviour in all its forms will be more prevalent in B than A. This is also what the Mountain People narrative leads us to expect.
The spreading of disorder and the maintenance
of antisocial behaviour
This chapter is about more than just whether there is neighbourhood difference in antisocial behaviour; that would be relatively quick to demonstrate. It is about using our neighbourhood study to try to understand the forces though which patterns of antisocial behaviour are maintained and transmitted. For what is absolutely clear is that antisocial behaviour is a transmissible condition. I mentioned Keizer, Lindenberg and Steg’s experiments in chapter 1. Those experiments showed that antisocial rule-breaking behaviours can be increased in a field setting by sowing the environment with small cues that other people are already being antisocial, such as graffiti or litter (Keizer et al., 2008). This ‘spreading of disorder’ principle is a powerful one. It lies behind the ‘broken windows’ theory of crime, which asserts that allowing relatively small antisocial acts to remain visible in the environment leads to increases in much more serious crime. The outputs of the small antisocial acts—the broken windows and graffitied walls—come to serve as informational inputs to other potential perpetrators, whose behaviour in turn serves as the input to others, and so on in an escalating cycle.
What interests me most about the spreading of disorder principle is its potential to make antisocial behaviour inflationary and self-perpetuating. Let us say that there is a small initial difference between two neighbourhoods; maybe there is one desperate person who goes around despoiling the public street in one of them but not the other. His acts serve as inputs to other decision-makers, who then are more prepared to commit small antisocial acts of their own; those in turn infect people contemplating more serious destruction, and before we know it one neighbourhood is obviously disordered—and has a high crime rate—and the other not. Even if we now remove the person who was the initial source of the disorder, the difference may persist. The experiences of the people in the two neighbourhoods have become different enough to lead them to choose different behaviours, and those different behaviours in turn feed into the experiences of their peers. Thus, the result could be a kind of pluralistic ignorance, where even if everyone in the neighbourhood has a preference not to behave antisocially, they all make the inductive bet—because of the overwhelming evidence around them—that other inhabitants are antisocial, and so they may as well follow suit. Thus, a small initial difference in neighbourhood conditions could lead to a rather large difference in outcomes, one with the potential to perpetuate itself indefinitely without counteracting forces.
There may be counteracting forces, of course. They can come from prosocial behaviour, explaining the intimate link between absence of prosociality and presence of antisociality. Where people are prosocially inclined, they will be willing to invest in small acts of order restoration, like clearing up litter even if they did not cause it, or repairing a window that they did not break, and they will be seen doing so. Thus, visible prosocial acts can neutralize or reverse the self-reinforcing cycle (Keizer, Lindenberg, & Steg, 2013). Another way that prosocial behaviour can be a brake on the spreading of disorder is through third-party sanctioning. Third parties will often intervene, at cost to themselves, where they see antisocial acts occur. This can take various forms, the simplest of which is the social embarrassment of the perpetrator, but it provides some kind of deterrent. Third-party sanctioning can be thought of as a kind of prosociality, since it provides a specific benefit to the victim of the act and, through deterrence, a more diffuse benefit to the community. Where social capital is higher, people are more willing to sanction, and this is an important brake on the spreading of disorder.
In light of the foregoing discussion, we can therefore make several predictions concerning antisocial behaviour in Neighbourhoods A and B. First, there will be more of it in B, and it will span both minor acts such as littering and more serious criminal acts; according to the ‘broken windows’ theory of crime, minor and more serious types of antisocial behaviour should go together. Second, the spreading of disorder principle says that increased antisocial behaviour in B should be substantially mediated by a perception that other people in the neighbourhood are being antisocial; take this perception away, and the neighbourhood difference should be much smaller. Finally, given the lower social capital, we should expect less willingness to prosocially sanction wrongdoers in B. This should heighten the difference in antisocial behaviour. We will now scrutinize the data from the Tyneside Neighbourhoods Project to examine whether these predictions are supported.
The simplest place to start is littering in Observational Dataset 1. I coded a number of things that could be considered markers of antisocial behaviour. There was littering (dropping refuse to the ground) and its mirror image, bin (disposing of refuse in a designated on-street refuse container). A category of damaging recorded anyone apparently trying to break with hands or stones, or set fire to, the street furniture, a vehicle, or a building. Finally, spitting could perhaps be considered a kind of antisocial behaviour, since it is presumably convenient to the spitter but would be considered by many to negatively affect the hygiene of the environment.
As Figure 4.1 shows, there were large neighbourhood differences in all of the markers of antisocial behaviour, though the absolute numbers of acts observed were small in some cases. I saw littering six times as often in B than A; about every 31 minutes of observation, as opposed to every 3 hours. Relatedly, I never saw anyone put anything into a bin in B, though I saw it 4 times in A, and street bins are widely available in both neighbourhoods. These figures refer to acts of dropping litter. I am sure that a survey of litter already on the ground would reveal a huge neighbourhood difference too. Acts of damaging were fortunately rare, but stacked up 4-1 for Neighbourhood B. Finally, spitting was over four times as frequent in B as A. Note that the total number of people observed was reasonably similar in the two neighbourhoods, so these differences are not just a product of different opportunities for observation.
That tells us a lot about minor antisocial acts; is there a neighbourhood difference in crime too? Agathe’s police dataset recorded all incidents notified to the police over the period December 2010 to March 2011, 585 incidents in all. There were 385 in B to 200 in A, a ratio of 1.93. Figure 4.2 categorizes these by type of incident, showing also the ratio of the number of incidents in B to the number in A.
The only category of crime for which Neighbourhood B does not show an elevated rate is vehicle crime. This is unsurprising since there are fewer and less valuable vehicles in B. Other categories show at least 70% more incidents in B than A. This includes a greater incidence of what the police describe as antisocial behaviour, a category that refers to disputes, littering, and graffiti, among others. This is consistent with the evidence from Observational Dataset 1. Of particular note in Figure 4.2 are two things: burglaries are much more common in B than A, even though the monetary value of the households is surely much lower. Naively, you might predict that A would be the more attractive place for burglars to ply their trade. Secondly, the biggest neighbourhood difference of all is in violence. Crime overall is slightly less than twice as common in B than A; violence is nearly six times as common. The pattern we observed in Neighbourhood B is apparently typical of extremely deprived neighbourhoods: an excess of crime overall, and a particularly large excess of violent crime (Krivo & Peterson, 1996).
These findings relate to several claims I have already made. I claimed that people in Neighbourhood B are more likely to be close to the edge than those in A. The fact that there are many more burglaries and robberies in B than A reinforces this point. Most burglaries are performed close to the perpetrator’s home (Bernasco & Nieuwbeerta, 2005). Thus, most of these crimes are probably committed by people from in and around the neighbourhood. As well as being very risky, they are probably not very lucrative. These are acts of people who are close to the edge. We have already seen the example of the men who robbed their own bookmaker in chapter 3. To take another example, an off-licence close to Neighbourhood B was robbed on May 8th, 2011 by a man with a large knife. He made off with two packets of cigarettes and a bar of chocolate, a total value of less than £20. He was arrested outside the shop. Only to a person at or over the edge could the possibility of £20 worth of chocolate and cigarettes outweigh the reasonably large chance of a prison term.
The finding that violence is so much more common in B relates to the claim I made in chapter 3 about social relationships in B being more volatile over time than those in A. We don’t have much detail about these violent incidents or why they happened. However, we know that most violence happens between people who know each other, and so at the very least we can say that its greater prevalence indicates that social relationships in B more often take the most extreme swing possible to the negative. Even if this is rare in absolute terms, it is much more likely to happen in Neighbourhood B than A.
The observational and police datasets are useful for telling us what people do, but they give us no real window onto why they do them. That is, they provide no way of investigating the cognitive processes underlying the decision to behave antisocially, since they do not furnish the opportunity to asking the litterers or criminals anything. Thus, a different method was required. We needed to recruit participants from Neighbourhoods A and B, offer them the opportunity to behave antisocially, and explore the cognitive and situational determinants of their decisions. Economic games were going to be the way to do this.
This part of the project was Kari Britt Schroeder’s work. I described in chapter 1 the economic game she designed that was attached to Social Survey 2. Briefly, players were formed into mutually anonymous triads. There was an initial allocation of £10 to each player. Player 1 could decide to ‘steal’ up to £10 from player 2, thus increasing his earnings to a maximum of £20. The unfortunate player 2 could not do anything about this. Player 3, however, could decide to sanction player 1 for his behaviour, by paying £2 to reduce player 1’s take-home amount by £6. At the time player 3 was filling in her survey, she did not yet know what player 1 had decided to do, so what we obtained from player 3 was a series of choices: if player 1 takes £0, would you fine him? If player 1 takes £1, would you fine him? etc. Thus, we ended up, for each triad, with a look-up table of what to do in terms of final payouts for every possible player 1 decision about how much to steal.
The player 1 decisions are an assay of willingness to commit antisocial behaviour. From the rest of Social Survey 2, we had data from those same player 1s about their perceptions of norms concerning social cheating. The questions Kari asked to probe perceived norms were about three domains of real-world antisocial behaviour: cheating on benefits, cheating on taxes, and cheating on public transport fares. We asked both about injunctive norms or acceptability (to what extent is it acceptable to cheat in this way?) and descriptive norms or prevalence (how much do people in this neighbourhood actually cheat in this way?). These norms ratings were on an effectively arbitrary continuous scale. The spreading of disorder principle predicts that the greater player 1’s perception of the prevalence of cheating in the neighbourhood, the more she will steal from player 2. It does not predict any association between the amount player 1 steals and her rating of the acceptability of social cheating, though common sense would make this prediction. So many findings in the social sciences are intuitively obvious: it is nice in this case to have a theoretical prediction that does not completely coincide with intuition. Unreflective intuition would say that people who think social cheating is less acceptable will steal less, whereas the spreading of disorder principle suggests that people who think others around them are cheating less will steal less. These are not mutually exclusive of course, but it was interesting to investigate which one panned out.
Let us look at the results for the player 1s (Figure 4.3; I have plotted these in a different way from Kari’s more sophisticated analysis in Schroeder, Pepper, and Nettle [2014]). First, player 1s behaved differently in the two neighbourhoods. The top left panel of Figure 4.3 is what is called a violin plot. It shows how the data (the amounts stolen) are distributed across their possible range of £0 to £10. Where the violin is wide, there are many observations, and where it is narrow, there are few. As you can see, in Neighbourhood A the violin is wide at £0. This means that most people stole £0; this is reinforced by the fact that the median amount stolen (black dot) is £0. There is a very thin neck in the middle of the range, meaning very few people stole an amount like £5, and a little bulge at the top, representing a small number of people who stole everything. The violin for Neighbourhood B is quite different. The bulge at zero is not so pronounced; there were many fewer people who stole nothing. There are more marked bulges in the middle and at the top, meaning that many more people stole half or everything. This is reinforced by the fact that the median theft for Neighbourhood B was £5. In other words, the middle person of the people we sampled in A took nothing from player 2; the middle person of the people we sampled in B took half of what there was to take. The neighbourhood difference was robust to controlling for the subjective value of a few pounds, as well as obvious covariates such as age and sex.
The upper right panel of Figure 4.3 shows how rated acceptability and prevalence of social cheating compare across the neighbourhoods. There is no real neighbourhood difference in the ratings of acceptability. Respondents from both places thought that social cheating was pretty unacceptable. There was, however, a huge difference in the perceived prevalence: respondents from B thought that social cheating was much more common in their neighbourhood. This matters, because of what the lower two panels of Figure 4.3 show. There was no relationship between how much player 1s took in the game and how acceptable they regarded social cheating as being (lower left panel). This is very striking and, as mentioned above, unintuitive: people who a few minutes later went on to take all £10 rated social cheating as just as unacceptable as people who went on to take nothing. By contrast, there was a relationship between perceived prevalence of social cheating and amount taken: the more common you regard social cheating as being in your neighbourhood, the more you take from poor old player 2. The neighbourhood difference in the perceived prevalence of social cheating, it turns out, largely explains the neighbourhood difference in how much of the £10 player 1 took.
These results resonate with our expectations. For one thing, they support the general claim, backed up by much empirical research, that normative influence is very important (Cialdini, Reno, & Kallgren, 1990). Under a range of circumstances, people are prone to doing as much or as little as they perceive others in the surrounding population to be doing. Normative influences are often much stronger than people realize; respondents protest that such influences are unimportant and that they base their decisions on more elaborate and independent criteria, such as values or extensive reasoning. The behavioural evidence tends to suggest otherwise (Nolan, Schultz, Cialdini, Goldstein, & Griskevicius, 2008). The results also confirm the more specific prediction of the spreading of disorder principle: the more the participants felt they saw evidence of cheating going on around them, the more they stole. This is really just the importance of normative influence applied to antisocial behaviour in particular.
We now turn to player 3. What is she doing in all this? Recall that we predicted people in Neighbourhood A might be more willing than those in B to sanction antisocial behaviours, even when they were not the injured party. This did indeed turn out to be the case (Figure 4.4). The figure shows the proportion of player 3s from each neighbourhood saying that they would impose the fine, for each of the possible amounts that player 1 might steal from player 2. In both neighbourhoods, there was a sense of graduated sanction. That is, most people would let pass a small theft of a pound or two, but as the theft became bigger, more and more of them said that they would intercede. The proportion opting to sanction never approached 1 in either neighbourhood; there were plenty of people who were simply not going to get involved. However, the proportion not willing to get involved was much higher in Neighbourhood B than A, and as a consequence, the probability of getting fined increased less as the amount player 1 stole got larger. In Neighbourhood B, even a player 1 who stole the whole £10 was more likely to get away with it than not.
What explains the greater willingness to sanction in Neighbourhood A than B? People in Neighbourhood A reported greater trust and did not rate so highly the value of the £2 the fine cost them. Both trust and the subjective value of £2 were important in predicting people’s willingness to sanction (specifically, willingness to sanction was greater where the subjective value of £2 was low, and amongst those for whom the subjective value of £2 was low, willingness to sanction increased with trust of the neighbours).
The combination of variation in the subjective value of £2 and variation in trust was sufficient to explain the neighbourhood variation in sanctioning behaviour. These results suggest that people will intercede on behalf of third parties when it is not too costly for them to do so, and when they trust that they are part of a viable prosocial network. Interceding is after all a kind of prosociality, and so, like other kinds of prosociality, we are more willing to embark on it when we have confidence in the social entity within which it will occur.
An experiment with information
The data from Kari’s study that I have presented so far shows a clear association between willingness to behave antisocially and the perception that others behave antisocially (that is, the perceived descriptive norm). However, the claim we wanted to make was stronger than just an association: we wanted to argue that perceived descriptive norms play a causal role in maintaining greater antisocial behaviour in Neighbourhood B than A. Demonstrating causality is a challenge. Surveys are informative, but the general consensus is that only experiments (randomised control trials, if you prefer) can really tell you anything very firm about cause and effect. An experiment in this context would mean a study where participants were randomly assigned to experience different descriptive norms about cheating in their neighbourhoods. If the association is causal, their antisocial behaviour should then vary according to the norm they were assigned to. Of course, it is rather difficult without vast resources to think of a way of experimentally assigning different participants to experience different descriptive norms in their neighbourhoods. Nonetheless, Kari came up with an ingenious plan that went at least part of the way towards a true experiment.
The experiment was done using the player 2s from the Theft Game described in the preceding section. Recall that the unfortunate player 2 had no active decision to make in the game; he or she was the potential victim of player 1’s theft. However, we did ask player 2s whether or not they expected player 3 to come to their aid and punish player 1 in the event that player 1 stole £5 from them. Expecting player 3 to come to your aid is not the same thing as stealing, of course, but it is related to it. Presumably if people in a neighbourhood began to expect more sanctioning by third parties, their willingness to commit antisocial behaviour would be reduced. Thus, we in effect used the player 2 expectations about sanctioning as a marker for the perceived extent to which antisocial behaviour could be got away with.
Now, how could we experimentally manipulate people’s experience of descriptive norms in the neighbourhood? Other studies have done this by, for example, deliberately littering in some areas or on some days (Cialdini et al., 1990). However, this was difficult to do for a study on a whole-neighbourhood scale taking place over several months. Instead what we did is a really rather subtle manipulation of the social information available to the participants. We acquire our beliefs about descriptive norms at least partly through social communication; we are influenced by what others say the norms are. So we decided to present some player 2s in B with information that implied other residents thought the neighbourhood descriptive norms were not as bad as they really were; we also presented some participants in A with information implying that other residents thought the neighbourhood descriptive norms were worse than they actually were. We called this the norms treatment: in the norms treatment groups, you got information implying that your neighbours thought your neighbourhood was more like the other neighbourhood was in reality. These participants were to be compared to two control groups who received no information at all about what other residents thought about the local descriptive norms. The clear prediction was that perceptions of the probability of player 3 sanctioning would be shifted: in A, the norms treatment should reduce the perceived probability of sanctioning, whilst in B, the norms treatment should increase it.
The details of how to implement the norms treatment were quite involved. As mentioned in chapter 2, for ethical reasons, Kari did not want to give people false information, but she did have to manipulate their informational state. She thus decided on the following solution: in the norms treatment, instead of the questions asking for ratings of the prevalence of cheating on taxes, benefits, and public transport fares in the neighbourhood, player 2s would see a statement informing them that we had asked ten of their neighbours what they thought about these issues, and showing the average results. We really had asked ten people in each neighbourhood, of course, so what the participants saw were real data. In fact, we had asked many more than ten. The experimental manipulation consisted in the fact that the ten we chose to present were unrepresentative: ten of the most favourable for Neighbourhood B, and ten of the least favourable for Neighbourhood A. We hoped that receiving this biased social information would cause people to shift their own representations of what the descriptive norms concerning cheating were in the neighbourhood.
The results from the norms manipulation are shown in Figure 4.5. In the control conditions, respondents from B were slightly less likely than those from A to expect player 3 to intercede on their behalf. This difference does have a basis in reality, since player 3s from B really were less likely to intercede. However, in the norms condition for Neighbourhood B, the expectation of player 3 sanctioning was strikingly higher—around 60% as opposed to around 30% for the controls. One of our predictions was thus fulfilled. The other prediction was that the proportion expecting sanctioning should be reduced by the norms treatment in Neighbourhood A. There was no evidence for this. In fact, the proportion of respondents expecting sanctioning was slightly higher in the norms condition than the control for Neighbourhood A.
These results are really very striking. For Neighbourhood B at any rate, they confirm the causal importance of perceived norms in cognition concerning prosocial and antisocial behaviour. They also show how strong the social transmission of these perceived norms is. Just by telling people that their neighbours thought that the neighbourhood was a slightly less antisocial place than the behavioural evidence suggests it is, you could effectively and instantly double people’s expectations that a stranger would intercede to sanction antisocial behaviour. And if residents’ expectations that antisocial behaviour would get sanctioned could be doubled, you would instantly provide a massive deterrent effect against antisocial behaviour itself. Thus, our experiment suggested that the apparently entrenched antisocial culture of Neighbourhood B might actually be quite labile: you just needed to persuade everyone in B that everyone else in B was motivated to be prosocial and not antisocial. It would become a self-fulfilling prophecy.
This finding relates to the issue discussed at the outset of the chapter about how the spreading of disorder principle means that antisocial cultures can be self-sustaining. Once people start to believe that others in the surrounding community are on the lookout for themselves and not doing their bit for the common good, they will behave accordingly. Their behavioural output becomes perceptual input to their fellow community members, and so the loop is hermetically closed. The cultural tradition can persist even if whatever perturbation initially gave rise to it has been removed. Our Neighbourhood B results suggest that perhaps quite small and simple things might break the loop quite quickly: getting neighbours to talk to each other positively about the places where they live; volunteer days where people could see each other doing prosocial acts; deep community clean-up to remove the visible cues of disorder. Other experimental work with manipulations of perceived norms suggests that the effects of such interventions could be quite powerful (Cialdini et al., 1990; Keizer et al., 2008, 2013).
Before we conclude that our results show that community regeneration is going to be straightforward and require only superficial nudges, some caveats are in order. First, the shift in expectations of sanctioning that we achieved in Neighbourhood B in this experiment may well have been quite fleeting. We have no idea if there would still be an effect if we had asked respondents a week or even a day later. I suspect not; the thing about real-world experience is that it is happening all the time. One particular set of cues might be influential in the short term, but it will soon be overwritten by others. The second is a more general point. The chances are that what maintains the relatively antisocial culture of B is not just the social transmission effect whereby everyone does what they think everyone else is doing. There are also more people who are materially desperate, as I made clear in the discussion of the edge principle in chapter 3. People at or over an edge will sometimes do antisocial things regardless of what they perceive the local descriptive norm to be. Thus, breaking the cycle of social transmission might well ameliorate the neighbourhood difference in the short term, but it is always going to be in danger of breaking out again if the underlying socioeconomic issues are not addressed. I return to this issue in chapter 7.
The strange case of the norms effect that didn’t happen
There is an aspect of the results of Kari’s norms experiment that I have glossed over thus far: the complete failure of the norms treatment to produce the predicted result in Neighbourhood A. Being given information suggesting that the neighbours thought A was less prosocial than it actually is did not reduce the expectation of sanctioning. That expectation was if anything slightly higher in the norms treatment than in the control. In addition, a curious and unpredicted thing happened. Participants from A who received the norms treatment did not seem to want our money. Six of them spontaneously opted out of receiving the payment from the game: three suggested we donate it to charity, two told us to keep it for university funds, and one just said he did not want to receive it. Only one person in the Neighbourhood A control condition deflected payment in this way, so it seems like something may be going on.
One interpretation of these findings is as follows. People in Neighbourhood A know that it is a nice and prosocial place. They are not under any uncertainty about that, and so their beliefs are relatively immune from moment-to-moment influence. Strictly speaking, in information theory, the amount of information carried by a signal depends on the extent of the uncertainty in the receiver (Shannon, 1948). Our Neighbourhood A respondents were not uncertain, and so the cues we provided were not informative and did not lead to them updating their expectations. All the Neighbourhood A respondents did was to signal to us that we were wrong, demonstrating what a prosocial bunch the residents of A are by spontaneously transforming our economic game into an opportunity for charitable giving. Not only did they not heed the social information we provided in the norms treatment; six of them sought to actively counter it by a well-chosen prosocial gesture.
The corollary interpretation for Neighbourhood B is that people there are in considerable uncertainty about the state of the social world. Thus, even the rather subtle cue that we fed them carried considerable information, and their running representations shifted markedly. This relates to the finding reported in chapter 3 that residents of B feel they know their neighbours less well, even though they demonstrably interact with them more frequently, than residents of A do. That finding showed that residents of B feel that they need more information about their social world; the current one suggests that they are responsive to it when it comes. I suggested in chapter 3 that the difference in information-hunger is to do with people’s behaviour being more variable over time in B than A, and the same claim works for the information-reactivity we see here. When behaviour is variable, more data are more useful. Moreover, you need to give a strong weight to the most recent data, since that is what is most diagnostic about the current state of affairs. In an unchanging environment, you can give a lot of weight to your historical experience, but in a volatile environment, historical experience means little; it is the most recent data that are going to be of some help. Thus, residents of B may be the most tuned for any news or evidence going around of how social behaviour in the neighbourhoods is shifting. Interestingly, this suggests that it may be in disordered or deprived neighbourhoods where you could have the most dramatic short-term effects from interventions like clearing up litter or mending broken windows.