The Rules of Contagion

by Adam Kucharski

  There were around 400,000 people in the full SSL database, with almost 290,000 of them deemed high risk. Although the algorithm didn’t explicitly include race as an input, there was a noticeable difference between groups: over half of black twenty-something men in Chicago had an SSL score, compared with 6 per cent of white men. There were also a lot of people who had no clear link to violent crime, with around 90,000 ‘high-risk’ individuals having never been arrested or a victim of crime.[87]

  This raises the question of what to do with such scores. Should police monitor people who don’t have any obvious connection to violence? Recall that Papachristos’s network studies in Chicago focused on victims of gun violence, not perpetrators; the aim of such analysis was to help save lives. ‘One of the inherent dangers of police-led initiatives is that, at some level, any such efforts will become offender-focused,’ Papachristos wrote in 2016. He argued that there is a role for data in crime prevention, but it doesn’t have to be solely a police matter. ‘The real promise of using data analytics to identify those at risk of gunshot victimization lies not with policing, but within a broader public health approach.’ He suggested that predicted victims could benefit from the support of people like social workers, psychologists, and violence interrupters.

  Successful crime reduction can come in a variety of forms. In 1980, for example, West Germany made it mandatory for motorcyclists to wear helmets. Over the next six years, motorcycle thefts fell by two thirds. The reason was simple: inconvenience. Thieves could no longer decide to steal a motorcycle on the spur of the moment. Instead, they’d have to plan ahead and carry a helmet around. A few years earlier, the Netherlands and Great Britain had introduced similar helmet laws. Both had also seen a massive drop in thefts, showing how social norms can influence crime rates.[88]

  One of the best-known ideas about how our surroundings shape crime is the ‘broken windows’ theory. Proposed by James Wilson and George Kelling in 1982, the idea was that small amounts of disorder – like broken windows – could spread and grow into more severe crimes. The solution, therefore, was to restore and maintain public order. The broken windows theory would become popular among police forces, most notably in New York City during the 1990s, where it inspired a heavy crackdown on minor crimes like subway fare dodging. These measures coincided with the massive drop in crime in the city, leading to claims that arrests for misdemeanours had stopped the larger offences.[89]

  Not everyone was comfortable with the way that the broken windows theory was adopted. One of them was Kelling himself. He has pointed out that the original notion of broken windows was about social order rather than arrests. But the definition of public disorder can be a matter of perspective. Are those people loitering or waiting for a friend? Is that wall covered in graffiti or street art? Kelling suggested that it’s not as simple as just telling police officers to restore order in an area. ‘Any officer who really wants to do order maintenance has to be able to answer satisfactorily the question, “Why do you decide to arrest one person who’s urinating in public and not arrest another?”’ he said in 2016. ‘If you can’t answer that question, if you just say “Well, it’s common sense,” you get very, very worried.’[90]

  What’s more, it’s not clear that aggressively punishing minor offences was the main reason for New York’s decline in crime in the 90s. There’s little evidence that New York’s reduction was a direct result of broken windows policing. Many other US cities saw a drop in crime during that period, despite using different policing strategies. Of course, this doesn’t mean broken windows policing has no effect. There’s evidence that the presence of things like graffiti and stray shopping trolleys can make people far more likely to litter or use an out-of-bounds thoroughfare.[91] This suggests that minor disorder will spark other minor offences. The effect seems to work the other way too: attempts to restore order – like picking up litter – can prompt others to tidy up as well.[92] But it’s quite a leap to go from such results to the conclusion that arrests for misdemeanours can explain a massive drop in violence.

  So what caused the decline? Economist Steven Levitt has argued that expanded access to abortion after 1973 played a role. His theory goes that this meant there were fewer unwanted children, who would have been more likely to be involved in crime when they grew up. Others blame childhood exposure to leaded petrol and lead paint in the mid-twentieth century, which caused behavioural problems later on; when the level of exposure declined, so did crime. In fact, a recent review found that, in total, academics have proposed twenty-four different explanations for the decline in US crime during the 1990s.[93] These theories have attracted plenty of attention – as well as criticism – but the researchers involved all acknowledge that it’s a complicated question. In reality, the drop in crime was likely the result of a combination of factors.[94]

  This is a common problem with outbreaks that occur on long timescales. If we intervene in some way, we might have to wait a long time to see if it has an effect. In the meantime, there might be lots of other changes going on too, making it hard to measure exactly how well our intervention works. Similarly, it can be easier to focus on the immediate effects of a violent event, rather than investigate longer-term harm. Charlotte Watts has pointed out that domestic violence can be transmitted across generations, with affected children becoming involved in violence as adults. However, these children can often be forgotten when discussing interventions. ‘We need to think about support for children growing up in households where there is domestic violence,’ she said.

  Historically, it’s been difficult to analyse intergenerational transmission given the timescales involved.[95] This is where public health methods can help, suggests epidemiologist Melissa Tracy, because researchers have experience analysing long-term conditions. ‘That’s the strength of epidemiology, bringing that life course perspective.’

  Using public health approaches to prevent crime would be hugely cost-effective, both in the US and elsewhere. Adding together the social, economic and judicial consequences of the average US murder, one study put the cost of a single killing at over $10m.[96] The problem is that the most effective solutions may not be those that people are most comfortable with. Do we want to feel like we’re punishing bad people, or do we want less crime? ‘When it comes to behavior change, threats and punishment are just not that effective,’ said Charlie Ransford of Cure Violence. Although punishment might have some impact, Ransford suggests that other approaches generally work better. ‘What is ultimately most effective at changing a person’s behavior is when you try to sit down and try to listen to them and hear them out, let them air their grievances and really try to understand them,’ he said. ‘And then try to guide them to a healthier way of behaving.’

  Projects like Cure Violence have historically focused on in-person interactions, but online social contacts are increasingly influencing the spread of violence as well. ‘The environment has changed,’ Ransford said. ‘You need to make an adjustment. Now we’re hiring workers who specialise in combing through social media to look for conflicts that need to be responded to.’

  When dealing with crime and violence, it helps to understand how people are linked together. The same is true of outbreaks; we’ve seen how real-life contacts can drive contagion ranging from smoking and yawning to infectious diseases and innovation. But the strength of influence online won’t necessarily be the same as face-to-face encounters. ‘If you think about contagion of views about acceptability of violence,’ said Watts, ‘the reach may be much larger, but the number of people who act might be smaller.’

  It’s a problem that a lot of industries are interested in. However, they generally aren’t so interested in controlling contagion. When it comes to online outbreaks, people tend to care about transmission for the opposite reason. They want to make things spread.

  5

  Going viral

  ‘Your nike id order was cancelled,’ read the e-mail. It was January 2001, and Jonah Peretti was trying to
get some personalised trainers. The problem was the name he’d requested; as a challenge to the company, he’d asked for his trainers to be printed with the word ‘sweatshop’.[1]

  Peretti, then a graduate student in the MIT Media Lab, ended up exchanging a series of e-mails with Nike. The company reiterated that it wouldn’t place the order because of ‘inappropriate slang’. Unable to talk them round, Peretti decided to forward the e-mail thread to a few friends. Many of them forwarded it to their friends, who forwarded it on, and on, and on. Within days, the message had spread to thousands of people. Soon the media picked up on the story too. By the end of February, the e-mail chain had gained coverage in The Guardian and Wall Street Journal, while NBC invited Peretti on to the Today Show to debate the issue with a Nike spokesperson. In March, the story went international, eventually reaching several European newspapers. All from that single e-mail. ‘Although the press has presented my battle with Nike as a David versus Goliath parable,’ Peretti later wrote, ‘the real story is the battle between a company like Nike, with access to the mass media, and a network of citizens on the Internet who have only micromedia at their disposal.’[2]

  The e-mail had spread remarkably far, but perhaps it had all been just a fluke? Peretti’s friend and fellow PhD student Cameron Marlow seemed to think so. Marlow – who would later become head of data science at Facebook – didn’t believe a person could deliberately make something take off like that. But Peretti reckoned that he could do it again. Soon after the Nike e-mail, he got a job offer from a multimedia non-profit called Eyebeam in New York. Peretti would end up leading a ‘contagious media lab’ at Eyebeam, experimenting with online content. He wanted to see what made things contagious and what kept them spreading.

  Over the next few years, he would start to piece together features that were important for online popularity. Like how jumping on emerging news stories could drive traffic to websites. And how polarising topics got more exposure, while ever-changing content kept users coming back. His team even pioneered a ‘reblog’ feature that allowed people to share others’ posts, a concept that would later become fundamental to how things spread on social media (just imagine how different Twitter would be without a retweet option, or Facebook without a ‘share’ button). Peretti would eventually move into news, helping to develop the Huffington Post, but those early contagion experiments stuck in his mind. Eventually, he suggested to his old boss at Eyebeam that they create a new kind of media company. One that specialised in contagion, taking their insights about popularity and applying them on a massive scale. The idea was to compile a rolling stream of viral content. They called it BuzzFeed.

  Not long after duncan watts published his work on small-world networks, he joined the Department of Sociology at Columbia University. During this period he became increasingly interested in online content, eventually becoming an early advisor to BuzzFeed. Although Watts had started off studying links in networks like film casts and worm brains, the world wide web contained a wealth of new data. In the early 2000s, Watts and his colleagues began to explore these online connections. In the process, they would overturn some long-held beliefs about how information spreads.

  At the time, marketers were getting excited about the notion of ‘influencers’: everyday people who could spark social epidemics. Nowadays, the word ‘influencer’ has evolved to refer to everything from influential everyday people to celebrities and media personalities. But the original concept involved little-known individuals who can spark word-of-mouth outbreaks. The idea was that by targeting a few unexpectedly well-connected people, companies could get ideas to spread much further for much less cost. Rather than relying on a celebrity like Oprah Winfrey to promote their product, they could instead build enthusiasm from the ground up. ‘The whole thing that made it interesting to people in the marketing world was that they could get Oprah-like impact from small budgets,’ said Watts, who is now based at the University of Pennsylvania.[3]

  The idea of such influencers was inspired by psychologist Stanley Milgram’s famous ‘small-world’ experiment. In 1967, Milgram set three hundred people the task of getting a message to a specific stockbroker who lived in the town of Sharon, near Boston.[4] In the end, sixty-four of the messages would find their target. Of these, a quarter flowed through the same one person, who was a local clothing merchant. Milgram said it came as a shock to the stockbroker to find out that this merchant was apparently his biggest link to the wider world. If an innocuous merchant could be this important for the spread of a message, perhaps there were other, similarly influential people out there too?

  Watts has pointed out that there are actually multiple versions of the influencer hypothesis. ‘There’s an interesting but not true version,’ he said, ‘and then there’s a true but not interesting version.’ The interesting version is that there are specific people – like Milgram’s clothing merchant – who play a massively disproportionate role in social contagion. And if you can identify them, you can make things spread without huge marketing budgets and celebrity endorsements. It’s an appealing idea, but one that doesn’t hold up under scrutiny. In 2003, Watts and his colleagues at Columbia re-ran Milgram’s experiment, this time with e-mails and on a much larger scale.[5] Picking eighteen different target individuals across thirteen countries, the team started almost 25,000 e-mail chains, asking each participant to get their message to a specific target. In Milgram’s smaller study, the clothing merchant had appeared to be a vital link, but this wasn’t the case for the e-mail chains. The messages in each chain flowed through a range of different people, rather than the same ‘influencers’ cropping up again and again. What’s more, the Columbia researchers asked participants why they forwarded the e-mail to the people they did. Rather than sending the message to contacts who were especially popular or well connected, people tended to pick based on characteristics like location or occupation.

  The experiment showed that messages don’t need highly connected people to get to a specific destination. But what if we’re interested simply in making something spread as far as possible? Could people who are more connected in the network – like celebrities – help ensure it takes off? A few years after the e-mail analysis, Watts and his colleagues looked at how web links propagate on Twitter. The results suggested that content was more likely to spread widely if it was posted by a person with lots of followers or a history of making things take off. Yet it was no guarantee: most of the time these people weren’t successful at creating large outbreaks.[6]

  Which brings us to the more basic version of the influencer hypothesis. This is simply the idea that some people can be more influential than others. There is plenty of evidence to support this. For example, in 2012 Sinan Aral and Dylan Walker studied how a person’s friends influenced their choice of apps on Facebook. They found that within friendship pairings, women influenced men at a 45 per cent higher rate than they influenced other women, and over-30s were 50 per cent more influential than under-18s. They also showed that women were less susceptible to influence than men and married people were less susceptible than singles.[7]

  If we want an idea to spread, we ideally need people to be both highly susceptible and highly influential. But Aral and Walker found that such people were very rare. ‘Highly influential individuals tend not to be susceptible, highly susceptible individuals tend not to be influential, and almost no one is both highly influential and highly susceptible to influence,’ they noted. So what effect could targeting influential people have? In a follow-up study, Aral’s team simulated what would happen if the best possible people were chosen to spark a social outbreak. Compared with choosing randomly, the pair found that picking targets effectively could potentially help things spread up to twice as far. It’s an improvement, but it’s a long way from having a few little-known influencers who can spark a huge outbreak all by themselves.[8]

  Why is it so hard to get ideas to spread from person to person? One reason is that issue of people rarely being both susceptible an
d influential. If someone spreads an idea to lots of susceptible people, these individuals won’t necessarily pass it on much further. Then there’s the structure of our interactions. Whereas financial networks are ‘disassortative’ – with big banks connected to lots of small ones – human social networks tend to be the opposite. From village communities to Facebook friendships, there’s evidence that popular people often form social groups with other popular people.[9] It means that if we target a few popular individuals, we might get a word-of-mouth outbreak that spreads quickly, but it probably won’t reach much of the network. Sparking multiple outbreaks across a network may therefore be more effective than trying to identify high profile influencers within a community.[10]

  Watts has noticed that people tend to mix up the different influencer theories. They might claim to have found hidden influencers – like the merchant in Milgram’s experiment – and used them to make something spread. But in reality they may have just run a mass-media campaign or paid celebrities to promote the product online, in effect bypassing word-of-mouth transmission altogether. ‘People either carelessly or deliberately conflate them, to make the boring thing sound like the interesting thing,’ Watts said.