Opinion dynamics 2: Let’s go viral

1 What would a social-contagion-informed strategy look like?

TODO: describe how civil society has an advantage here over corporate campaigning:

Network interventions are cheap

• pay for themselves
• grassroots
• have attached business models
• I will literally transmit beliefs in my social graph if you ask me to
• You should ask me to transmit social beliefs in the right way and to the right people so that I do it productively

Declaration: I do not have all the answers here. In fact, no-one has all the answers here (although I suspect that some political organisations have more answers than they are sharing). Academically, this is a fresh research area, and it is not my own core research area.

But in any case, a tidy package with a bow on it is not the goal here. What we do gain by thinking about networks is a list of methods to try that just might be better if we stop assuming them away.

We are a long way into this piece, so I will just start from one. There is a sequel coming with some more stuff in it.

2 Social contagion is where campaigners can achieve leverage cheaply

One objection I hear from campaigner friends is that it is not helpful to know about social contagion in beliefs because there is no way to influence social contagion. This is false in the strict sense; we intervene in social contagion all the time in the course of our lives, as that is what human interaction is. This blog post has likely travelled to you through some social network or other.

We might, however, admit that in any given case we may not immediately see social-network-based strategies which are fast or cheap or scalable, or some other desirable quality. Thinking back to the analogy with epidemiology, we can imagine the same objections being raised to the early germ theorists. “What use is this germ theory? It is not like it gives us any way of controlling disease.” And likely the early germ theorists, too, despaired of the possibility of teaching people to wash their hands or boil their water. Back then, though, the answer was “So let us throw that in the bin and return to treating diseases as caused by miasmas.” Rather it was to use contagion theory to decide what to investigate next and devise strategies informed by contagion models. And that is how we were able to unlock new productive methods to control disease, from public education, through to modern antibiotics and vaccines, that have radically altered our ability as a society to cope with disease.

Moreover, to discount the importance of proactive social network methods is to discount the real and sophisticated organising methods. To discount that possibility is to ignore the social-network-based opinion changes that have in fact been stewarded by many and diverse groups, often in a sophisticated fashion. Social-network-based organising has been harnessed by campaigners for queer rights, by fascists, by communists, by unionists, by gun owners, by suffragists and anti-slavery campaigners, by civil rights campaigners and by churches, and now, by trolls, marketers, terrorists and recruiters.

TODO: example citation for each use?

From another angle again, Tokita, Guess, and Tarnita (2021) suggests that intervening in social networks is how radicalisation works. In studying Twitter users they discover that people choose to self-segregate by news source. TODO - that’s polarization. connect to terrorism radicalisation pathway?

Peer pressure has complicated dynamics depending on the structure of the communities (Damon Centola 2010).

For a case study which illustrates perfectly how this plays out, I recommend the podcast Absolutely Devious Lick.

Social brain hypothesis (Bretherton and Dunbar 2020; Dunbar 1993) pro-social group behaviour (Boyd and Richerson 1992; Richerson, Boyd, and Henrich 2003)

How good are we at forming good facty beliefs? Scott Alexander discusses the irritating case study of bodybuilders suggests…. tl;dr: We are not very good at facty beliefs?

• New research model reveals how contagion spreads among network of connected people
• Optimized Propaganda with Bayesian Networks: Comment on “Articulating Lay Theories Through Graphical Models” (Discussion of Powell and Weisman (2018); source)
• Contagion and self-fulfilling dynamics
• Minimum viable superorganism is a famed introduction to the prestige economy.
• Gun Control and Homophily in Social Networks
• Sergey Ivanov, GML Newsletter: Homophily, Heterophily, and Oversmoothing for GNNs

Via Byrne Hobart, Odlyzko and Tilly (2005):

This note presents several quantitative arguments that suggest the value of a general communication network of size $n$ grows like $n\log (n)$. This growth rate is faster than the linear growth, of order $n$, that, according to Sarnoff’s Law, governs the value of a broadcast network. On the other hand, it is much slower than the quadratic growth of Metcalfe’s Law, and helps explain the failure of the dot-com and telecom booms, as well as why network interconnection (such as peering on the Internet) remains a controversial issue

An interesting phenomenon happens with free speech;

Persuasion is a publication and community for everyone who shares three basic convictions:

• We seek to build a free society in which all individuals get to pursue a meaningful life irrespective of who they are.
• We believe in the importance of the social practice of persuasion, and are determined to defend free speech and free inquiry against all its enemies.
• We seek to persuade, rather than to mock or troll, those who disagree with us.

In the past years, the political and intellectual energy has been with illiberal movements. Too often, the advocates of free speech and free institutions have been passive, even fatalistic. It is high time for those of us who believe in these enduring ideals to stand up for our convictions.

Or, more exotic for me, Open Source Defense — “100% gun rights, 0% culture war”.

The most generous interpretation of these projects, which I personally incline to, is that here are some individuals who truly wish to engage across the political spectrum, burst filter bubbles and make society as a whole work as a collective project (even though they might disagree with me about some details). The least generous interpretation of those projects is that they are led by people pushing malign agendas who wish to make the other side look shrill and aggressive. Of course, we can have both. Whichever you believe is their intent, if you believe these projects have a chance of success, the rational response seems to be to meet their onslaught of openness, reason and curiosity with openness, reason and curiosity of your own.

In fact there are variant strains of social contagion hypotheses in circulation, depending on how broadly or narrowly we define the terms here, but I think we can gloss over that for now.

There is another concept that is very useful here:

Social graph

My social graph is who I talk to; my family, my friends, my co-workers, my co-religionists, the people I argue with on the internet, etc. We can also refer to this as my “social network” but if I say that, it sounds like I am talking only about Instagram or Facebook or something purely technological. Those internet-based social networks represent one important type of social graph, but not the only one.

Why do we want to think about social graphs? Because these are where we get opinions from.

For a small number of people, who study bushfires, or who live through them, or who suffer from a serious COVID-19 (or vaccine) side effect, their information is not so much a political opinion; they have information of different kinds. But for most of us, our information about these society-wide problems and responses is a political opinion.

For a smaller number of people, who study bushfires, or who live through them, or who suffer from a serious COVID-19 (or vaccine) side effect, their information is not so much a political opinion; they have information of different kinds. But for much of the public, our information about these society-wide problems and responses is a political opinion. And those political opinions, although we might think otherwise, are transmitted to us through our social graph. Let us look at some of the evidence about this.

This is how doing something to fit in with our peers can feel like taking a stand against a society gone mad.

We often get this wrong. I get this wrong every time I try to change a relative’s mind by arguing with them about something; Every time I do that I am falling prey to the fallacy that they (or I) get opinions from argument. I am not saying persuasion of this sort never happens, but certainly being persuaded by argument is an infrequent occurrence for me, if I am being honest with myself, and it is slow hard work for the poor soul who seeks to persuade me.¹

Outgroup biases mean that it is hard for us to have good models of how other people form their own beliefs; those people in that group all think alike. They are not diverse and independent-minded like us over here. Pro-tip: Everyone in every group thinks that, and few of us have data to back it up.

Starting Not Your Imagination— Society Is Going Insane and I Can Prove It

1. We underestimate how social our beliefs are

2. the social networks through which we get our opinions are hard to reason about

3. as a result, we operate on other models

   1. rhetorical persuasion
   2. opinion dynamics are a property of who you are (ethnic group, economic status…
   3. Personality “types”, intrinsic “values”

4. By focusing on those effects we tend to discount a more important effect, which is that social networks can explain most of what we believe

For bonus time we could consider persuadability, evidence from influencers, and data from the internet, covariates like in-group membership and status, etc. We can muddy the waters by talking about how homophily and contagion are indistinguishable by observation; mention the obesity question.

I do not notice social contagion. Social contagion feels like

(Goel, Mason, and Watts 2010):

It is often asserted that friends and acquaintances have more similar beliefs and attitudes than do strangers; yet empirical studies disagree over exactly how much diversity of opinion exists within local social networks and, relatedly, how much awareness individuals have of their neighbours’ views. This article reports results from a network survey, conducted on the Facebook social networking platform, in which participants were asked about their own political attitudes, as well as their beliefs about their friends’ attitudes. Although considerable attitude similarity exists among friends, the results show that friends disagree more than they think they do. In particular, friends are typically unaware of their disagreements, even when they say they discuss the topic, suggesting that discussion is not the primary means by which friends infer each other’s views on particular issues. Rather, it appears that respondents infer opinions in part by relying on stereotypes of their friends and in part by projecting their own views. The resulting gap between real and perceived agreement may have implications for the dynamics of political polarization

Some of our preferences clearly are pretty hard-wired in and come from some other place (god? genes? zodiac?). We don’t really need to worry about THAT here though, because opinions are more like roller skates than sexual orientation. That is, they are more like fashion items than hard-wired choices.

Antonio García Martínez, in The Holy Church of Christ Without Christ, belabours the point that faith-based engagement is how we predominantly engage with the world. Or, as Herbert Simon and Eliezer Yudkowsky could have co-authored, belief is how a heuristic feels from the inside.

When it comes to statistics, we understand bucket models very well and have been working with them for a long time. This makes sense when you look at all the early successes of statistics in agriculture and industry going back more than a century. If you are trying to learn which breed of cow is best to keep on your farm, then it makes sense to analyse cows in terms of which breed/bucket that cow comes from.²

Social network models are much newer and less well understood, but they are also very important. A big example from recent history is epidemiology, the study of the spread of a disease. If I try to understand my risk of catching a disease by a bucket model, then I am throwing out much useful information. To work out my risk of catching COVID-19, for example, there is some information in which demographic bucket you can assign me to, but really, the most informative bit of information about my risk of catching COVID-19 is whether I have been in contact with someone else who has the disease. That is, COVID-19 spreads along certain connections in my social graph.

And indeed we should expect the similarity to viruses to not be absolutely perfect for every possible idea. There is no literal physical bacterium or virus which transmits ideas around in little strands of RNA or DNA.

OK, what exactly do I mean here by social contagion of beliefs? Do I mean that ideas are literally like viruses, and that there is some tiny molecule that carries belief from one person to another? Absolutely not. The ways that contagious beliefs are like contagious diseases is in their broad-brush behaviour. Ideas can rapidly spread from person to person, and the greater the contact the greater the potential is for transmission. If two groups are very separated they will not transmit ideas, and so on. You get the idea. Opinions and viruses can both be described very well in terms of epidemiology. Just as we understand how disease spreads by looking at person-to-person contact, so do we with ideas.

The overarching theme is that to understand why people believe things you need to see who they hang out with.

I disagree. In fact, it is one thing that we can do. Walking down the street to talk to someone new is within the reach of most citizens.

Points to touch upon:

The problem with models of people’s intrinsic personality types and values are several:

1. We cannot ethically do brainwashing experiments or remove individuals from their communities, so our data is usually fairly observational; we do not know how hard it is to change a fundamental value.
2. Research about the stability of personalities is complicated and fraught (Big 5 traits are a mess, there are probably things that are stable about people over time but we don’t know how to disentangle that from their communities).
3. Even if we know what is stable over time, there is no God-given mapping from a small local personality trait to a nebulous massive confusing contested complicated thing like a mass-scale policy; Different policies get their values from how and by whom they are presented not by any intrinsic quality (If I ask what the preference for a given marginal tax rate for a given income bracket should be for a given vote conditional on their measured values of Agreeableness and Neuroticism, what will you tell me?)

Do we need to go into HPS and observational experiments here?

Return to earlier example.

It has been battle-tested and weaponised.

2.1 Bringing people together in common cause

Even if you have personal qualms about the stringency of the sampling methodology used in biblical tales, the Bible has served as an excellent blueprint for recruiting for many churches throughout history.

2.2 Let’s go viral

In a social-contagion informed policy we start by considering what it means to go viral. We care less about whether someone responds to a message per se than we care about how likely they are to pass on a message, and who they will pass it on to, and how.

Here we want to know if things are retweetable.

Hayward (2005) explicitly uses an epidemic model to study the growth of churches, where the growth or stagnation of the church is due to the contagiousness of recruits in their infectious phase.

Caveats: Duncan J. Watts and Dodds (2007) finds that influentials are defined relatively; if people are easily persuaded to take on a new idea it might be worth identifying the influencers in the network, but for entrenched ideas, might want to try other methods.

Weak ties and superspreaders (Bakshy, Rosenn, et al. 2012; D. Centola and Macy 2007; M. S. Granovetter 1973)

3 Correlation and causation and conditioning

Recycle some of the examples highlighted by Gwern correlation

3.1 Weak ties and connecting communities

Bridging communities is high value. This is well-studied in the context of church recruitment (Bretherton and Dunbar 2020; Jennifer M. McClure 2021; Hayward 1999, 2005).

Famously Burt (R. S. Burt 2002; Ronald S. Burt 2004) argues that bridging holes in the network gives the bridging individuals who do so high power.

Also the types of belief and the types of network relations can interact (Damon Centola 2010; D. Centola and Macy 2007). It can mean several different things for an idea to transfer between two humans.

4 Incoming

Chaotic notes vying for a place in the body text.

• Charis Chang, Federal election 2019: The rising power of the Chinese-Australian vote
• Chart: How U.S. Latinos Voted in the 2016 Presidential Election
• The Whole Dysfunctional National Conversation About Guns In One Interactive Graph
• Most Tweeted Gun Politics Links, Feb 18-Feb 21

Segmentation faults: how machine learning trains us to appear insane to one another. Semi-counterpoint: Doctorow on Big Tech narratives.

Other factors:

Thinking things through

Of course people sometimes think things through and we humans do sometimes come to our own conclusions. However, it is hard for campaigners and policy nerds to appreciate how rarely people do this. It is not a never thing, it is just that the number of policies people Think through per week are low, because things are, for most people, competing with Netflix and

Rational discussion

You might believe that people are rational and won’t take on strictly ridiculous opinions. This is largely unimportant in the current setting. I cannot get people to believe that fire is not hot, but that kind of thing was already ruled out at the start because the hotness of fire is not a political opinion. More generally though, some policies are surely sillier than others and all things being equal, it is not crazy to suppose that people might prefer less silly policies. Having seen the kind of policies we will regularly vote for, and the policies people will argue for over dinner, I think that this effect is probably weak and easily overwhelmed by more powerful forces. For the purposes of campaigning, this effect is not relevant. Remember, we need to persuade a voting majority of the population, not just a small fraction of us that are nerds and wonks with enough time to discuss every issue.

Advertising

Yes, mass advertising probably has an effect on us, and because politics is a delicately balanced matter. However, as advertising bleeds into social influence more generally, this distinction gets less interesting.

Hip pocket

OK, there is evidence that people do evaluate the cost to themselves of various policies, so I am not going to argue that this is unimportant. On the other hand, how important people think any given policy is, and whether they spend the mental effort on “running the numbers” on any given policy against their own financial plan and how willing they are to be generous, whose modelling is considered legitimate, this is all modulated through what their peers bother to do, even here we might consider the modulating effect of social graphs. People vote for economic insanities all the time.

Some things are innate

Clearly some things are not massively socially contagious. Sexual orientation does not appear to be especially socially transmissible, for example. The number of people who claim to be turned gay (or straight) by a friend, or a particularly good podcast, might not be zero, but it is small.

In the world, other theories of what people believe and why are in play.

TBH I am not across all the varied theories that people have for why the population at large has the opinions they do. I have heard of theories that include intrinsic values, and personality types and demographic categories, which zodiac sign a person was born under, which Myers-Briggs personality type etc. There are surely many others. These all have differing degrees of evidence to support them, and differing degrees of usefulness in different contexts. I am not going to attempt to address all of these here; that is a whole project of its own.

Status, prestige, trust. Timing. Tribal identification.

Previously in opinion dynamics, we discussed the importance of social graphs in understanding belief, and the high likelihood that if we are trying to intuit our way through understanding society, we are likely to underestimate the importance of those social graphs to forming opinions in ourselves and in others.

Oncoming onslaught of fake news and autogenerated content. GPT4 etc, Microsoft is Playing with Fire GPT-4

Alternative models

• Targeted advertising
• Rational discussion
• values
• economic decisions, personal
• demographics

1. Data collection
2. homophilous networks
3. weak ties/small worlds
4. Targeting key nodes

Until we had the understanding of contagion, we did not even have the ability to recognise this distinction between the contagious and the non-contagious, and were cursed to be ineffectual at controlling contagious diseases.

The people going into labs to do social psychology experiments are not any stupider than me on average. They are having the completely normal human experience that I also have, which is thinking that I am (they are) clear-sighted and perceptive, and other folks are sheeple.

In Australia we have a version of this.

5 References