The real revenge of The Tipping Point

How simple stories can give us a misguided view of contagion

A few years ago, while researching The Rules of Contagion, I re-read The Tipping Point, Malcolm Gladwell’s wildly popular 2000 book about how things become wildly popular. I’d originally read it shortly after it came out, and remember there being some great stories. But how would the science hold up now I knew a lot more about the field?

Early in the book, Gladwell describes the investigation of a 1981 gonorrhea outbreak in Colorado Springs, in which epidemiologist John Potterat and colleagues had interviewed 769 cases. The investigating team found that 168 individuals had at least two infected contacts, suggesting these individuals played a disproportionate role in the spread.

Gladwell proposed an explanation for this:

Who were those 168 people? They aren’t like you or me. They are people who go out every night, people who have vastly more sexual partners than the norm, people whose lives and behaviour are well outside of the ordinary.

Which made me pause. Because I’d previously read that paper on the Colorado Springs outbreak, and I didn’t recall it concluding that.

So I re-read the paper, and re-read it again. Then I contacted the authors to make sure I wasn’t missing anything. I would later summarise my findings in The Rules of Contagion itself:

Were these people really so promiscuous and unusual? Not particularly, in my view: the researchers found that, on average, these cases reported sexual encounters with 2.3 other infected people. This implies they caught the infection from one person and typically gave it to one or two others. Cases tended to be black or Hispanic, young, and associated with the military; almost half had known their sexual partners for more than two months.

During the 1970s, Potterat had begun to notice that promiscuity wasn’t a good explanation for gonorrhea outbreaks in Colorado Springs. ‘Especially striking was the difference in gonorrhea test outcome between sexually adventurous white women from a local upper middle class college and similarly aged black women with modest sexual histories and educational backgrounds,’ he noted. ‘The former were seldom diagnosed with gonorrhea, unlike the latter.’ A closer look at the Colorado Springs data suggested that transmission was likely to be the result of delays in getting treatment among certain social groups, rather than an unusually high level of sexual activity.

It’s difficult to understand how a journalist could have read that Colorado Springs analysis and concluded that promiscuity was to blame. How it was unusual people who were responsible. How they aren’t like you or me.

A key thread that runs through the original version of The Tipping Point – and, as we shall see, its 2024 update – is that there are certain special people hiding in plain sight, and it is these people that drive epidemics. For example, the book recounts Stanley Milgram’s 1967 ‘small-world’ experiment. In that experiment, Milgram got three hundred people try to send a message to a specific stockbroker in Sharon, near Boston. It turned out that a quarter of successful messages reached the stockbroker via a local clothing merchant. This suggested that certain influential ‘connectors’ might play disproportionate role in spreading information.

In 2003, sociologists Peter Dodds, Roby Muhamad and Duncan Watts re-tested the theory with a large-scale email experiment. They selected eighteen target individuals across thirteen countries, and started almost 25,000 e-mail chains, asking each participant to reach one of the targets. Unlike in Milgram’s much smaller experiment, the e-mails were forwarded through a range of people rather than a few predictable influencers. Participants tended to pick people to forward messages to based on practical reasons like location or occupation, rather than perceptions of social connectivity.

The idea that connectivity can matter for information sharing isn’t wrong overall. Some high profile individuals undoubtedly have much larger reach than others. But the idea that there are little-known individuals who can deliver celebrity-like impact without celebrity-size budgets doesn’t hold up to scrutiny. When I spoke to Duncan Watts during my research for The Rules of Contagion, he nicely summed the problem of the different versions of the influencer theory: ‘There’s an interesting but not true version, and then there’s a true but not interesting version.’

(If you want a detailed, well-researched exploration of popularity and social dynamics, I’d recommend Watts’ excellent book Everything Is Obvious, Once You Know the Answer.)

Broken windows and broken theories

Another theory that cropped up in The Tipping Point was the ‘broken windows’ outlook on crime. Coined by James Wilson and George Kelling in 1982, the idea is that visible signs of disorder, like broken windows, can lead to more serious crime. The theory would be most famously adopted by police in New York City in the 1990s. They introduced intensive measures targeting minor offenses like fare dodging, and this coincided with a significant drop in crime.

Several studies have corroborated the idea that minor disorder can spark more minor disorder. For example, the presence of graffiti and stray shopping trolleys can make passers-by more likely to litter or use an out-of-bounds thoroughfare. Likewise, efforts to restore order – like clearing up litter – can prompt others to clean up as well. But there’s a chasm between results like these and a conclusion that heavy-handed arrests for misdemeanours can lead to a massive drop in serious violence.

Researchers have since challenged the idea that ‘broken windows’ policing was the main driver of New York’s crime reduction, given many U.S. cities saw similar declines with other strategies. Over two dozen theories have been proposed for the 1990s crime drop, including explanations like expanded access to abortion or reduced exposure to lead in childhood. Despite journalists’ ongoing search for a simple reason, most researchers acknowledge it’s a complicated problem, and there are likely to be multiple interlinked causes involved.

Kelling himself would later clarify that the broken windows theory focused on maintaining social order, not on arrests. Given the subjective nature of what counts as disorder, he also voiced regret about the policies that had come from misapplications of the theory:

You’re just asking for a whole lot of trouble, You don’t just say one day, ‘Go out and restore order.’ You train officers, you develop guidelines. 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]?’ … And if you can’t answer that question, if you just say ‘Well, it’s common sense,’ you get very, very worried.

In a recent interview about Revenge of the Tipping Point, the 2024 follow up the original, Gladwell also acknowledged the problems with ‘broken windows’ policing:

The idea that crime was an epidemic and that criminal behavior was contagious is correct. But the idea that broken windows and stop-and-frisk were the correct response to a contagion is completely false.

He also suggests that readers shouldn’t get too tied to these ideas anyway:

I hold ideas very loosely, and I think it’s important for people who write about ideas to remind their readers to hold their ideas loosely.

It’s good to see an author acknowledging that ideas sometimes need updating. But examples like the Colorado Springs outbreak illustrate that it’s not just about holding ideas loosely. There’s a big difference between: 1) researching a phenomenon thoroughly, presenting the current best explanation, and acknowledging this may be subject to change; and 2) apparently skimming a research paper, then writing up a pithy explanation that can be debunked simply by reading the original paper properly.

Which brings me to a recently publicised example from Revenge of the Tipping Point, in which special people hiding in plain sight are once again the topic of interest. In this case, those people who shed lots of airborne particles when the speak. Who are these ‘super-emitters’? Well, according to the new book, one analysis found ‘the biggest predictors of high production of aerosols were age and body-mass index (BMI)’. Which prompted more questions:

What if age and obesity really are the two biggest predictors of super-spreading? Does that mean in the middle of a pandemic passengers will refuse to sit beside an overweight person on a plane? What if the answer is viscous saliva, and a scientist comes up with a 10-second test to measure if someone is in the 99th percentile? Would a restaurant or a movie theatre or a church be justified in asking everyone to take a saliva test at the door?

If we’re going to ask questions about targeting certain age groups and body shapes, I thought it was worth looking at the study that motivated all this. Below is what appears to be the key figure from the original paper. The x-axis shows participants’ body mass index multiplied by their age (referred to as ‘BMI-years’), and the y-axis shows a measure of exhaled particle concentration (note this isn’t a measure of virus, just random aerosols that come out of people’s mouths in normal life).

Now, you don’t have to be a statistician to spot that there is not a very clear trend in the graph above. Some people with a high BMI-year value shed lots, and some don’t. Some people with a lower BMI-year value shed lots, and some don’t.

Even if there is, on average, a slight trend in one direction, that’s not what we need for a control measure that targets individuals. It’s a problem I’ve written about before: we shouldn’t focus on population averages if we’re interested in individual outcomes. For example, let’s do a quick sense check on the above plot for policy use. Suppose we set the cutoff for ‘super-emitters’ at 1000 BMI-years, just before the first couple of apparent ‘super-emitters’. How many people in the UK meet this criteria? Overall, just under 65% of UK adults have a BMI over 25. And about half of the population are over 40. So that’s around a third of the population who would have a BMI-years measurement of 1000 or more. Like the ‘Great Barrington Declaration’ in autumn 2020, it’s hardly a targeted intervention once you run the numbers.

If we’re interested in targeting variation in aerosols, ideally we’d need a good predictor of whether a specific person will emit lots of particles. But even if we could predict this, is it what we really want?

During an epidemic, the probability someone sheds lots of infectious virus depends on two things: 1) the probability they would shed lots of virus if infected; and 2) the probability they are infected in the first place. We can write this equation down as follows:

\(P(\text{shed lots of virus}) = P(\text{would shed lots if infected}) \times P(\text{infected})\)

An accurate ‘superemitter test’ would give us an insight into one of those terms (the probability they would shed lots of virus if infected) but not the full equation. Hence for a targeted epidemic response, what we also need is either a test of whether someone is infected (like a PCR test), or even better, a test of whether someone is shedding quite a lot of virus currently (like a rapid antigen test).

Even at the peak of the UK COVID epidemic in winter 2023/24, only 4-5% of the community was testing positive by rapid test. So if we ignore information on individual infection status when assessing individual risk, we’re missing a key part of the picture – and targeting a lot of people who currently have zero risk of transmitting to others. (Much like how COVID lockdowns were such a blunt, blanket measure.)

This potential to miss the real picture is a major problem, in my view. Whether we’re talking about ‘broken windows’ policing, little known ‘connectors’ or superspreaders who ‘aren’t like you and me’, easy conclusions about contagion can quickly lead us in the wrong direction. When it comes to understanding how things spread – and what to do about it – we need good science and careful reasoning, not just great stories.

Cover image: dandrew via Unsplash