Opinion dynamics 1: Social contagion moves hearts and minds

How to win elections and influence people

Assumed audience:

Campaigners without background in modern viral marketing

A campaigner friend asked this question:

What’s the least dense, most accessible article to introduce people to the idea that opinions are an emergent property of social networks?

What a great question. I do not know a good introduction to social contagion theory of opinions… So I’ll write one real quick.

Here is a draft. This post covers the following main points:

  1. We should start by assuming that social contagion is the main way we humans get our political opinions.
  2. We should lean in to point #1 because we humans are wired to underestimate the importance of #1.
  3. We should lean in to point #1 because other things that looks like they are important often end up being about point #1.
  4. We should really lean in to #1 because even if it were to turn out that points 1, 2 and 3 were all nonsense, social contagion would still where campaigners can achieve leverage most cheaply.

Too many points? We can summarise the summary thusly:

tl;dr If you want a good strategy to move hearts and minds, start by treating people’s beliefs as contagions transmitted through their social network.

This idea is what we call “obvious (once you know the answer)”, in that once you start thinking this way it seems obvious. But it is not, for most people, obvious before you start thinking that way. Certainly it did not always seem obvious for me I think that this is common, that most people who have experience with social contagion of ideas forget that not everyone thinks in these terms, and thus we do not spend time making the case that this is a good way of thinking about things to laypeople.1 And that might be why I cannot find a good introduction to this topic; no one who knows it seems to feel it needs mentioning.

Anyway! without further ado…

Background: The science of disease before contagion

You might recognise this famous map, which marked a turning point in our understanding of contagious diseases.

John Snow’s iconic map of the Broad St Cholera outbreak, from wikipedia

This map was made by London physician John Snow, of the 1854 Broad Street cholera outbreak. Before his slam-dunk demonstration of the importance of germ theory, prevailing opinion was that cholera was induced by “miasma”, some kind of noxious gas associated with corpses. These corpsey miasmas were supposed to transmit cholera to people. And indeed, cholera was associated with corpses but not quite in the way they supposed. If you are around the stinking unsanitary conditions that cholera epidemics frequently arose in, of course you might feel that you are at dangerous risk of disease. But there is a massive difference between how you can make yourself safe from germs that arise from people, and noxious miasmas that come from the ground. Until Snow’s decisive evidence of this new-fangled person-to-person germ theory, we did not have the tools we needed to manage this disease; authorities could not untangle correlation between cholera cases from the common underlying cause of those cases.

With the understanding that cholera was transmitted by the bacteria, which came ultimately from other people, came the power to be vastly more effective in cutting off disease. In particular, in identifying that affected people around Broad Street were all getting their water from a single pump,which ended up being contaminated by a sick baby, they could cut off the water supply at that pump.2

Of course, this was not to say that there was nothing more to learn about cholera. They had yet to identify the relevant bacterium, invent antibiotics and so on. But until they understood that the cause of disease was something transmitted between people rather than something oozing out of a place, they could not even ask the right questions to understand that disease better. Scientists could spend all day shoving flowers up their noses to identify the one that seems to do the best at “stopping miasmas” but they will do a terrible job at identifying what is really going on. On the other hand, once they know about the transmissibility of the disease they can ask all sorts of useful questions: Is keeping food and drink free from contamination helpful? Is it airborne? Which bacterium is responsible? Does heating kill it? Soap? And so on.

For some reason, difficulty in understanding contagion is a recurring feature of human history. Time and again we have been reluctant to see diseases as contagions we catch from other people, preferring to imagine they are caused by something in the environment. Consider Ignaz Semmelweis’s battle to introduce hand-washing amongst doctors, or the arguments on Twitter that 5G towers cause cause COVID-19 and so on.

And of course some illnesses (say, scurvy3, or leukaemia) are indeed caused by the environmental factors.4 So it’s not crazy to ask whether something is contagious or not, or even whether contagiousness is only one of several factors contributing to a disease. What does not make sense it to assume it it is not contagious.

In many organisations and industries, we are in a “pre-germ theory” stage when it comes to when it comes to understanding opinions. Often, we restrict our analyses of human behaviour implicitly to “miasma”-type theories, never leveraging the modern understanding of how ideas are themselves are contagious things, transmitted person-to-person. And yet, the data is there to suggest that opinions are very contagious indeed. We have an embarrassment of data that suggests that opinions people hold are strongly determined by the opinions and behaviours of their friends, and transmitted through interpersonal relationships. So, we have reason to be skeptical of theories of belief which omit social contagion.

To pick an extreme example, people will copy the behaviour of their friends even unto death: 134 people burned to death in the notorious Beverly Hills Supper Club fire after failing to evacuate until much too late even when someone repeatedly raised the alarm.5 Now, did those people fail to evacuate because they had an innate preference for burning to death? Did a miasma of fire ignorance repress their better instincts? Was there no feasible way they could fact-check the existence of a fire?

A better explanation for this tragedy seems to be that, as far as the patrons were concerned, an epidemiological explanation for the belief. Their friends at the same table were not behaving as if that fire was real or important, and so why should they be the first to step out of line? There are many lessons there, but crucially for the current purposes it is that even at the risk of their lives people would prefer to act and think like their neighbours than fact-check their neighbours.

It might be tempting to imagine this example is abnormal but this kind of thing is famously easy to reproduce in the lab. This is well studied in the form of the bystander effect (Darley and Latane 1968; Latane and Darley 1969). And we all know many examples from history of collective delusions, false beliefs, millenarian cults, panics and other madnesses of crowds. The thing is, we are often reluctant to imagine that this madness-of-crowds dynamic might be ubiquitous in human life. Even if we do accept that idea though, that crowds are crazy, that beliefs are infections, we usually imagine it is that crowd of those people over there. Specifically, we, you and I, typically resist imagining crowd madness in ourselves, personally, right now. “Having a reasoned opinion,” however, is also what catching a contagious opinion feels like from the inside.

Being in thrall to ideas that we thoughtlessly catch from people around us is not necessarily a bad thing. Just because we caught a belief from a friend does not mean it is false. The belief that aeroplanes can fly would have seemed like collective madness a couple of centuries ago. These days, that belief is normal, and it is probably just as well for the efficiency of the airline industry that each passenger does not stop during the boarding process to check the equations of the Boeing engineers for themselves. On the other hand, just because a belief is contagious does not mean it is reflects the truth of the world. And just because it feels true does not mean we did not thoughtlessly acquire it from our friends or family.

Both true and untrue beliefs have a deeply social element. Perhaps because our antennae are so tuned to receive contagious beliefs and adopt them as our own, we find it easy to overlook that element in ourselves and in others. Whatever the basis, we can observe it is hard to think about beliefs as contagious. If we want to influence what people out there in the world believe, then we need to make that social element explicit, and work with it, or we are doomed to misunderstand the dynamics of influencing people, and do it badly.

And in fact, in many industries and fields, it is common to analyse beliefs with the same tools we use to analyse contagious diseases. In viral marketing, in internet modeling, in economics and in political science, we create and use epidemiological models of the spread of beliefs all the time. So, let us examine what these kinds of models are exactly, and how they differ from non-network models.

Very good. So I have argued that we need to consider contagion-explanations when understanding beliefs, and that we should not expect that those explanations should be instinctually obvious. What next?

It’s not what you know it’s who you know

Having argued that it is important to consider the contagiousness of beliefs, we need to get more specific so that we can make stronger statements. I do not want to try to come up with a complete epidemiological theory of why anyone believes any kind of thing ever, and the details of which beliefs have which degree of contagion and so on. Here, we are specialising our model, considering political opinions in particular.

Political opinions
Political opinions and beliefs_ are opinions and beliefs about how our society does work and how it should work, at a large scale.

There are two things that are special about this class of opinions:

  1. mustering political opinions in your support is essential, at least in a democracy, for getting things done
  2. political opinions are likely to be especially socially contagious for reasons that I will expand on in a moment.

What is special about “political” opinions/beliefs is that most of them are formed far from evidence. Political opinions are ones where we, as private individual citizens, will not get to see all the evidence first hand, because there is just too much of it. “Fire is hot” is not a political opinion, because I have all the evidence I need to support that and scars to prove it. “We need to put more funding into preparing to fight bushfires”6 is a political opinion. “Bushfires are made worse by the climate change” is a political opinion. “Covid vaccines are more (or less) dangerous than COVID-19” is a political opinion. And yes, “I should vote for [your choice of] party” is a political opinion.

All these opinions might in actuality correspond to true facts about the world (or not) but whether we hold those opinions is only remotely connected to whether those opinions are true.

You will notice that I am using belief and opinion as synonyms here, which might feel uncomfortable. In my opinion, there is no useful difference between belief and opinion, at least not in our current purposes.

Most of us are not bushfire survivors, but if our society has to develop a shared policy to deal with bushfires then we might all be asked to have an opinion on bushfires. These issues, where we all have an opinion but few of us have much information, are the ones where we expect the behaviour to be especially contagion-like. These are the circumstances in which contagion-type behaviour becomes important. We are less likely to see a huge outbreak of fire-denialism, than we are to an outbreak of climate-change denialism. When the connection to personal experience is remote, the dominant factor driving our belief is presumably social, which is to say, something best understood as a transmissible thing like a disease

In contagious beliefs like this we usually to consider the social network as a unit of analysis.

Social network
Social network is now the language we use to discuss online social media apps like TikTok or Facebook or Instagram. I am using social networks in a broader, older sense here. Sure, it might include facebook connections or whatever, but really I mean any repeated, interpersonal connection. If you talk to the barista at your local coffee shop, that barista is in your social network. You co-workers are in a social network with you.

In social network models we start from the perspective of the social relationships between people if we want to understand why they do a thing. If we want to understand why someone has a belief, we look at who might have transmitted it to them.

We can also imagine other factors, miasmas if you like, which might influence their belief, but a priori we suspect these factors are usually secondary influences upon political beliefs. No one is born with an innate opinion on the dictatorship of the proletariat or an appropriate maintenance schedule for the local government’s public gardens.

Because this is political science now, not epidemiology, we don’t talk about miasmas but refer to these as factors, or features. A sampling of such factors might be e.g. people are homophobic because of a particular brain structure (e.g. Ahn et al. 2014) or pro-environmental because of their genes (e.g. Chang et al. 2021), authoritarian because of economic stress (e.g. Ballard-Rosa et al. 2021), or conservative because of intrinsic values (which are presumable fixed in childhood) (e.g. Sheldon and Nichols 2009)… These various factor theories all suggest different dynamics that we expect observe in popular opinion, and different ways we might influence popular opinion to each other and to the network-type theories. In these factor theories, we lump people together based on factors such as race, age, gender, what-have you, and try to analyse what people who who possess similar factors are like.

So why should we favour network models? One reason: because contagion-type behaviour is important in describing the massive shifts in political that we observe in the world around us, right now. Here is one piece, (via Tanner Greer) that shows a particular opinion about free trade ripping through voters like COVID-19 through an unvaccinated population:

Credit CNBC

It is hard to find explanation for this kind of behaviour in terms of miasmas. The genes, brain structures and economic circumstances, fundamental values etc of American Republicans did not chance drastically from 2015 to 2016 but their support for particular policies did. What is going on? Well, that one, I assert, might well correspond to the rise of Trumpism, with Donald Trump as patient zero in that outbreak.7 People are attending rallies, talking to their friends and otherwise doing all the business of meeting people and taking on new opinions.

For another example, we could consider the deep canvassing model, devised the Los Angeles Lesbian, Gay, Bisexual, and Transgender Center in California.8 The research is great at showing how to talk to people to change their minds, but also, crucially, it demonstrates the effectiveness of changing minds by person-to-person contact.

Contagion-type, network-based models can easily support the kind of dynamical changes like this contagion, whereas the implicitly static models cannot. Just as we remained COVID-free for a very long time, then some of us who came into contact with COVID-patients acquired COVID, so can a public opinion remain dominant for a very long time then be overturned rapidly by an well-connected spreader. This also brings us to another important insight about social contagion: It suggests that we can, sometimes, produce massive changes in opinions, even apparently deeply held ones, by influencing people through their peers.

There are other models, of course. We might imagine a high-budget advertising campaign, changed opinions in this case. And indeed, advertising probably plays a role here too. But advertising is at the best of times an uncertain way of producing change, and we are generally suspicious that it can do anything like exponential spread through a population. Moreover, it is not incompatible — might advertising accelerate social network changes rather than cause it? How do we know, without investigating both these in convert?

How to do social-network-informed research

For a sampling of labs working on understanding opinion as network contagion, see, some of the following

There are many, many more.

How about this contagion idea in particular?

In modern social research we now have much terminology to categorise and understand the crucial social component of human behaviour. One technical distinction that is useful for the current purpose to us is homophily versus contagion. When we observe a group of people and we wonder why they exhibit a certain behaviour we might explain that in terms of causes.

This group of people behaves in such-and-such a way because similar people will form a group together
This group of people behaves in such-and-such a way because people who are in the same group tend to know each other.

Do political parties form policies that suit their members innate preferences (homophily)? Or do political parties exist and persuade their members to prefer certain policies (contagion)?

If your goal is to get certain policies accepted then the difference between these two ideas is very important.

Social network theories are ones that start from the premise that we want to study people in the context of the factors they are possess, or their connections within their social network. for the sake of exposition, we might characterise theories into “factor” and “network” type theories.

[Here we can imagine that that each circle represents a person, and they are classified by their support for two different parties, the red party and the yellow party.]

How exactly this happens is a massive research area. A lot of academics and industrial researchers have spent a lot of time and money trying to pin down the details of how belief is socially transmitted.

One ripper example of how to do this is the social penumbra paper (Gelman and Margalit 2021).9 This paper examines one particular quality of social graphs: how engaged the members of a certain group are with people outside that group.

Our study makes three contributions. First, it introduces the concept of penumbra as a relevant political characteristic of social groups, shifting the focus from the groups’ members to features of the group’s social circle (i.e., its outward contacts with nongroup members). Second, we provide a set of estimates of the penumbras of an array of different groups in society and highlight the significant variation in the penumbras’ size and characteristics (e.g., in terms of income, education, political leanings). Finally, we provide preliminary evidence regarding the political significance of penumbras, demonstrating that a change in penumbra status is in some cases also associated with shifting views on policy matters related to the group in question.

So, this is about groups that have learned to adopt a certain strategy of visibility and engagement with the society around them, and thereby succeeded in shifting social support for policies of interest to that group.

For example, about 3% of the US population identifies as gay or lesbian, about half the share of people with a mortgage “under water” at the time of our survey. Yet, the sizes of penumbras of the two groups go in the opposite direction, with 74% of respondents saying they knew at least one gay person and only 35% reporting that they knew someone whose mortgage is under water.… if a group such as underwater mortgage holders is hidden to much of the general population, this will affect the extent to which their problem becomes salient and how it is perceived by the public. These differences thus suggest why specific policies such as same-sex marriage and debt forgiveness to mortgage owners, while directly affecting similarly sized groups, could resonate very differently across the broader population.

This paper suggests that a group’s success in changing the beliefs and policies of broader society is strongly influenced by their success in making themselves both visible and engaged beyond their own group. This is true whether you’re talking about gun owners who’d like unrestricted access semi-automatic assault weapons, or LGTBQI+ people who’d like get on with their private lives without discrimination. In short, repping matters - take the pride out of gay pride & it might be a less powerful campaign. Pride is intrinsically a network phenomenon, about accepted within your network and amongst your peers, and learning how to propagate that acceptance to the mainstream has led to massive strides in legal support for this community.

There is much more material examining the social network structure of contagions; We see it in everything from exercise habits (Aral and Nicolaides 2017) to alt-right mobilising (Ribeiro et al. 2019). For a link dump on that theme, see memetics. For now, let us take it for granted that a lot of people think this is an important concept, and look at it from a different angle.

We are wired to underestimate the importance of social contagion

If none of this contagious-ideas stuff sounds very plausible to you, you should be aware that the research suggests that this is completely normal. Most people will find this kind of idea implausible even if it is true. Opinions do not work the way we think they do.

That is not because we are dumb per se, but because no matter how smart we are, humans are find it hard to reason about our own reasoning. 10 We are terrible are understanding how we ourselves think, and thus we are even worse and understanding how other people think, let alone whole societies full of other people. Recommended for a fun general introduction to the general topic of self delusion, see David McRaney’s You are not so smart (McRaney 2012), which has an excellent accompanying Podcast with some updated material.

St Anthony and the cognitive biases

tl;dr A sincerely-held personal conviction is what a socially transmitted belief feels like from the inside.

Self delusion leads us astray in many parts of life, and having a hard time seeing through our own flimsy opinions is one of those parts. If I acquire an opinion from my social network, I am likely to wilfully forget that I learned it from somewhere else. We call this Belief Change Blindness (Wolfe and Williams 2018), which is a tendency to believe that we have always believed what we now believe. Another is that we are all naive realists, which is to say, we tend to overestimate how much facts and reasoning have influenced our beliefs (but at the same time, we are likely to believe that that guy over there is full of superstition). Putting those together, I am unlikely to notice acquiring this belief from someone else. Instead, I am likely to think I have always felt this way, and I am likely to think it is based on facts, even if it is some random crap I heard a co-worker say off-handedly. If taking on a belief through gossip and hearsay feels like adopting smart, evidence-led beliefs, is it any wonder that we underestimate how big an effect gossip and hearsay has? The truth is, we are all sheeple most of the time, me, you and the person at the bus stop.

It gets worse when we are required to reason at a big scale. People are demonstrably terrible at thinking through the how opinions behave at large scales (Watts 2014). There are many experiments examining this — I quite like Salganik and Watts (2008), which is a large experiment where people were persuaded to like bands because experimenters manipulated the apparent popularity of their songs; At least in the short term, what is cool amongst your peers overwhelms what is good. When I put it that way, it sounds like I am saying something very obvious. But we behave as if making things cool was not how the world works, all the time, in political campaigning.

Writ large, Everything is Obvious (once you know the answer) is Duncan Watts’ laundry list of examples of how all our intuitions about how society works are self-justifying guesses divorced from evidence, except for his, because Yahoo let him build his own experimental online social networks.

So, however much you feel like social networks determine belief, add a little more on top of that, because you are likely underestimating it.

Why is in important to understand this? If we do not we might fail to understand how and why social media is weaponized and how to do it effectively ourselves. Whe might ascribe more importance to factors that we can find in the census —for example we might want to discuss “the Vietnamese community vote” and “the Boomer vote”. But this assumes away the possibility that those labels are attached to a more complicated — and possibly more influencable — social network of humans that we can reach in a way not captured by their census categories.

If we do not think this through properly, we can easily start a dangerously facile analysis. Consider the question repeatedly asked in a recent US election: Why did so many Latinos back Trump? The convenience of a single factor label like “Latino” hides a diversity of Hispanic communities.

“I literally had somebody ask me in 2016 what’s the bumper sticker that is going to mobilize Latinos?” García Bedolla says. “What we actually needed were mobilization strategies that would talk to the third, fourth, fifth, and sixth generation English monolingual Mexican Americans in San Antonio… That’s the kind of specificity that we need.”

Social network analysis invites us to think of the network of social connection that citizens are in, and whether we can reach them like normal people, through their communities, rather than the checkbox they ticked on a form somewhere.

Social contagion sets us up to ask useful questions about how we can make a difference

Meanwhile in Australia, Jakubowicz and Ho, in Was there an ‘ethnic vote’ in the 2019 election and did it make a difference?, correlate conservative policy preferences to ethnic community prevalence. There is not necessarily anything wrong with that per se. But consider the types of interventions that are suggested by considering this as a factor rather than a network problem. In the network version, a person’s Chinese ethnicity is an indicator that they are likely to have a social network including an above-average number of other ethnically Chinese people. If the people in this indivudal’s network were disproportionately likely not to vote for my preferred candidate, then I am encouraged in a social network model to imagine ways in which I being this individual and my candidate’s social networks together to chance this preference.

On the other hand, if I imagine Chineseness is an immutable fact leading to intransigent opposition to my preferred candidate then I am left with little to do to further my candidate’s interest. At least, little that I can do in a modern social democracy, where disenfranchising or expelling ethic groups or other categories based on the presumed political alignment is something we leave to authoritarian states.

Static factor models invite us to consider people as obstacles. Immutable factor models, that is, tend to suggest despair or despotism. Network models invite us to consider other human beings as precisely that. Social network models, that is, tend to suggest democratic engagement.

Other things that looks like they are important often end up being about social contagion

If you have a alternative factor-based theory to social contagion — say, that people of a certain ethnic group tend to share an opinion — then that also might turn out to be, partially or entirely, a theory about social networks in disguise. Let’s say you notice a particular belief that has an ethnic association, such as the famously common belief in Japan that blood type is implicated in personality type. Is there something about the ancient cultural heritage of Japan? No, it only dates back to the 1970s. This belief might be readily transmitted between people of Japanese background because, like most humans, Japanese are more likely to know people of the same ethnicity as themselves. Maybe there are other beliefs that the Japanese community would be open to transmitting, even?

But even if you think you really have found some hidden factor in people that makes them easy to classify and predict, you _still_might really want to think about the networks amongst people who have that factor. Imagine you are a marketer trying to sell a product, say, roller skates, and you have a factor-type theory of human personality. Maybe your theory says people have pro-roller-skate values, or fun-affinity or whatever you wish to call it. Now, to sell roller skates you need to find all the people in the world with pro-roller skate values. What on earth are pro-roller skate values? Which values are anti-roller skate values?

Now I do not work in marketing but I am pretty sure your roller skate strategy won’t do well compared to that of a colleague who thinks that there is no need to identify roller skate values. Probably a better marketing strategy is getting some people out there to have fun with roller skates, let other people see that roller skates are indeed fun, especially with their friends, and let viral marketing do the rest. In other words, let the social network do the marketing with spending time trying to chase the ineffable roller skate affinity.

But suppose that you are persuaded that roller-skate values are real and important.

  1. You can decide roller-skate values are fixed so there is nothing to be done to persuade them. Also no-one having roller skates is evidence that no one wants roller skates, so maybe selling roller skates is pointless.
  2. Or maybe there are enough people who have pro-roller-skate values — if only people could discover their true inner roller skate values. You need a way to spread the actualization of pro-roller-skate values amongst the latent roller-skate-positive population. Maybe I need to do that through a viral marketing strategy anyway.

When we reach point two we are left wondering if we have learned anything. We have maybe changed our estimate of how thoroughly we can succeed?

Majority illusions and visibility

TODO: explain how huge this possibly is.

In homophilic networks (0.5 ≤ h ≤ 1), the minority overestimates their own size (filter bubble) and the majority underestimates the size of the minority. The insets show the same information on log scale to make the amount of underestimation and overestimation comparable. As group sizes become more disproportionate, perception bias increases. (Lerman, Yan, and Wu 2016)

In the left social network, most people see the red party as the majority party, and in the right social network, most people see the yellow party as a minority.


Postscript: Did this help you?

NB: I am curious about the number of people who will gain something from this post. On one hand, hipster, up-to-the-minute viral marketers in tech startups will be aware of the outline of much content here, as it is their bread and butter and they have data to back it. On the other hand, you can get at many of the same conclusions by classic, intuitionistic old-school social organising, of the kind that is centuries or millenia old, and they have history to back that. In between these vintages, there seems to be a cohort of people who seem to make do with curiously little social network theory. This is my target: Social campaigning’s middle children (Hello if this describes you, middle child!) You might fit into this category if you acquired your ideas of human behaviour by osmosis from the google analytics dashboard, or maybe if you have never tried to measure your effectiveness in the world at all. Regardless, whoever you are, I would welcome feedback.


  • Damn it, Ignaz Semmelweis and Joseph Lister would have been better examples than John Snow.
  • follow-up about how to use social graphs in practice. Are all social relationships equal? In fact, that will be in opinion dynamics II
  • Mention that this miasma/network setup is over-simplified, but it is a passable intuition pump-prim as we move to more advanced stuff


Ahn, Woo-Young, Kenneth T. Kishida, Xiaosi Gu, Terry Lohrenz, Ann Harvey, John R. Alford, Kevin B. Smith, et al. 2014. Nonpolitical Images Evoke Neural Predictors of Political Ideology.” Current Biology 24 (22): 2693–99.
Aral, Sinan, and Christos Nicolaides. 2017. Exercise Contagion in a Global Social Network.” Nature Communications 8 (1): 14753.
Ballard-Rosa, Cameron, Mashail A. Malik, Stephanie J. Rickard, and Kenneth Scheve. 2021. The Economic Origins of Authoritarian Values: Evidence From Local Trade Shocks in the United Kingdom.” Comparative Political Studies, July, 00104140211024296.
Broockman, David, and Joshua Kalla. 2016. Durably Reducing Transphobia: A Field Experiment on Door-to-Door Canvassing.” Science 352 (6282): 220–24.
Chang, Chia-chen, Thi Phuong Le Nghiem, Qiao Fan, Claudia L Y Tan, Rachel Rui Ying Oh, Brenda B Lin, Danielle F Shanahan, Richard A Fuller, Kevin J Gaston, and L Roman Carrasco. 2021. Genetic Contribution to Concern for Nature and Proenvironmental Behavior.” BioScience, no. biab103 (October).
Darley, John M., and Bibb Latane. 1968. Bystander Intervention in Emergencies: Diffusion of Responsibility.” Journal of Personality and Social Psychology 8 (4, Pt.1): 377–83.
Gelman, Andrew, and Yotam Margalit. 2021. Social Penumbras Predict Political Attitudes.” Proceedings of the National Academy of Sciences 118 (6).
Kong, Quyu, Marian-Andrei Rizoiu, and Lexing Xie. 2020. Modeling Information Cascades with Self-Exciting Processes via Generalized Epidemic Models.” In Proceedings of the 13th International Conference on Web Search and Data Mining, 286–94. WSDM ’20. New York, NY, USA: Association for Computing Machinery.
Latane, Bibb, and John M. Darley. 1969. Group Inhibition of Bystander Intervention in Emergencies. Journal of Personality and Social Psychology 10 (3): 215.
Lerman, Kristina, Xiaoran Yan, and Xin-Zeng Wu. 2016. The ‘Majority Illusion’ in Social Networks.” PLOS ONE 11 (2): e0147617.
McRaney, David. 2012. You Are Not So Smart: Why You Have Too Many Friends on Facebook, Why Your Memory Is Mostly Fiction, an d 46 Other Ways You’re Deluding Yourself. Reprint edition. New York: Avery.
Ribeiro, Manoel Horta, Raphael Ottoni, Robert West, Virgílio A. F. Almeida, and Wagner Meira. 2019. Auditing Radicalization Pathways on YouTube.” arXiv:1908.08313 [Cs], August.
Salganik, Matthew J., and Duncan J. Watts. 2008. Leading the Herd Astray: An Experimental Study of Self-Fulfilling Prophecies in an Artificial Cultural Market.” Social Psychology Quarterly 74 (4): 338.
Sheldon, Kennon M., and Charles P. Nichols. 2009. Comparing Democrats and Republicans on Intrinsic and Extrinsic Values.” Journal of Applied Social Psychology 39 (3): 589–623.
Shin, Minjeong, Alasdair Tran, Siqi Wu, Alexander Mathews, Rong Wang, Georgiana Lyall, and Lexing Xie. 2021. AttentionFlow: Visualising Influence in Networks of Time Series.” In Proceedings of the 14th ACM International Conference on Web Search and Data Mining, 1085–88.
Watts, Duncan J. 2014. Common Sense and Sociological Explanations.” American Journal of Sociology 120 (2): 313–51.
Wolfe, Michael B., and Todd J. Williams. 2018. Poor metacognitive awareness of belief change.” Quarterly Journal of Experimental Psychology (2006) 71 (9): 1898–1910.

  1. If you work in social media research or in viral marketing, or weaponised social media engagement, or troll farms, at this point you might be way past wondering why I am bothering explaining that your job is real. This essay is not for you! Maybe some writing on virality or the theory of social graphs.↩︎

  2. For more on this story see Simon Rogers, John Snow’s data journalism: the cholera map that changed the world↩︎

  3. Interestingly, Scurvy was erroneously believed to be caused by germs for a while↩︎

  4. genes are of course also a factor for many diseases, and in turn interact with the other factors in complicated ways, but that is a story for another textbook.↩︎

  5. The story of that fire is super interesting, by the way.Go and listen to Tim Harford’s podcast about that fire, if you’ve not heard about it.↩︎

  6. wildfires for Americans↩︎

  7. We might also imagine that political opinion happens through some other process, such as reasoned debate, or through media persuasion. We’ll come back to those options later.↩︎

  8. There is a fascinating scientific scandal in the history of this idea, what with the famous LaCour study, which was fake, and the Broockman and Kalla study (Broockman and Kalla 2016), which was real. There is a convenient Dave McRaney podcast on this theme.↩︎

  9. The experiment design in this paper, by the way, is extremely clever. Experiment design is not really our focus here, but another thing a campaigner could also learn from this paper is how to design surveys to indirectly extract high reliability information from average punters. Do your surveys identify the size of someone’s social circle by asking them how many Marias they know?↩︎

  10. I think of this as the Dunning-Kruger theory of mind.↩︎

No comments yet. Why not leave one?

GitHub-flavored Markdown & a sane subset of HTML is supported.