Clustering is common with other types of violence too. In 2015, a quarter of US gun murders were concentrated in neighborhoods that made up less than 2 per cent of the country’s overall population.[17] When Gary Slutkin and his colleagues set out to tackle violence as if it were an outbreak, it was neighbourhoods like these that they planned to target. They called the initial programme ‘CeaseFire’; this would later evolve into a larger organisation called Cure Violence. In those early days, it took a while to work out precisely what approach they should use. ‘We took five years of strategy development before we put a single thing on the street,’ Slutkin said. The Cure Violence method would end up having three parts. First, the team hires ‘violence interrupters’ who can spot potential conflicts and intervene to stop the transmission of violence. Someone might end up in hospital with a gunshot wound, for example, and an interrupter will step in to talk their friends out of a retaliatory attack. Second, Cure Violence identifies who is at greatest risk of violence, using outreach workers to encourage a change in attitudes and behaviour. This can include help with things like job hunting or drug treatment. Finally, the team works to change social norms about guns in the wider community. The idea is to have a range of voices speaking out against a culture of violence.
Interrupters and outreach workers are recruited directly from the affected communities; some are former criminals or gang members. ‘We hire workers who are credible with that population,’ said Charlie Ransford, Cure Violence’s Director of Science and Policy. ‘To change people’s behaviour and talk them out of doing something it helps if you have an understanding of where they’re coming from, and they feel like you have an understanding and maybe even know you or know someone who knows you.’[18] This is another idea familiar in the world of infectious diseases: hiv programs will often recruit former sex workers to help change behaviour among workers who are still at high risk.[19]
The first Cure Violence project started in 2000, in West Garfield Park in Chicago. Why did they pick that location? ‘It was the most violent police district in the country at the time,’ Slutkin said. ‘It has always been my bias – as it is for many epidemiologists – to head for the middle of the epidemic, because it’s your best test and you can affect the greatest impact.’ One year after the programme started, shootings in West Garfield Park had dropped by about two thirds. The change had been rapid, with interrupters breaking the chains of violence from one person to another. So what is it about these transmission chains that makes interruption possible?
Late on a sunday afternoon in May 2017, two gang members emerged from an alleyway in Chicago’s Brighton Park neighbourhood. They were carrying assault rifles. The pair would end up shooting ten people, killing two of them. It was retaliation for a gang-related murder earlier in the day.[20]
Shootings in Chicago are often linked like this. Andrew Papachristos, a sociologist at Yale University, has spent several years studying patterns of gun violence in the city. A native of Chicago, he’d noticed that shootings were frequently tied to social contacts. Victims would often know each other, having previously been arrested together. Of course, just because two people are connected and share a characteristic – like involvement in a shooting – it doesn’t necessarily mean that contagion is involved. It might be down to the environment they share, or because people tend to associate with those who have similar characteristics (i.e. homophily).[21]
To investigate further, Papachristos and his collaborators obtained data from the Chicago Police Department on everyone who’d been arrested between 2006 and 2014.[22] In total, there were over 462,000 people in the dataset. Using this information, they plotted a ‘co-offending network’ of people who’d previously been arrested at the same time. Many of the individuals hadn’t ever been arrested with someone else, but there was a large group who could be linked together through a series of co-offending events. Overall this group included 138,000 people, or about a third of the dataset.
Papachristos’s team started by checking whether homophily or environmental factors could explain the observed patterns of gun violence. They found that it was unlikely: many shootings occurred in a linked way that couldn’t be explained by homophily or environment, suggesting contagion was responsible. Having identified the shootings that were likely due to contagion, the team carefully reconstructed the chains of transmission between one shooting and the next. They estimated that for every 100 people who were shot, contagion would result in 63 follow-up attacks. In other words, gun violence in Chicago had a reproduction number of about 0.63.
Fifty simulated outbreaks of shootings, based on the dynamics of violence contagion in Chicago. Dots show shootings, with (grey) arrows indicating follow-up attacks. Although there are some superspreading events, most outbreaks involve a single shooting and no onward transmission.
If the reproduction number is below one, it means that an outbreak might spark but it rarely lasts very long. The Yale team identified over four thousand outbreaks of gun violence in Chicago, but most were small. The vast majority consisted of a single shooting, with no additional contagion. However, occasionally the outbreaks were much larger; one included almost five hundred linked shootings. When we see these highly variable outbreak sizes, it suggests that transmission is driven by superspreading events. Analysing the outbreak data from Chicago in more detail, I estimated that transmission of gun violence was highly concentrated. It’s likely that fewer than 10 per cent of shootings led to 80 per cent of follow-up attacks.[23] Just like disease transmission – which can be similarly influenced by superspreading – most shootings didn’t lead to any additional contagion.
The chains of transmission in Chicago also revealed the speed of transmission. On average, the generation time between one shooting and another was 125 days. Despite the attention given to
