Whether you live in the USA or have just been watching the circus from afar, chances are that you agree: “polarization” dominates descriptions of the social landscape. Judging from the news alone, one might think the States have never been so painfully divided…yet nuanced public polls, and new behavioral models, suggest another narrative: the United States is largely moderate, and people have much more in common with each other than they think. There’s no denying our predicament: cognitive biases lead us to “out-group” one another even when we might be allies, and the game of politics drives a two-party system into ever-more-intense division, until something has to give. But the same evidence from social science offers hope, that we might find a way to harness our collective thinking processes for the sake of everyone and row together toward a future big enough to hold our disagreements.
Welcome to COMPLEXITY, the official podcast of the Santa Fe Institute. I’m your host, Michael Garfield, and every other week we’ll bring you with us for far-ranging conversations with our worldwide network of rigorous researchers developing new frameworks to explain the deepest mysteries of the universe.
In this episode we talk to SFI External Professor Henrik Olsson and SFI Complexity Postdoctoral Fellow, Omidyar Fellow, and Baird Hurst Scholar Vicky C. Yang about their work on social cognition and political identity. In a conversation that couldn’t be more timely, we ask: How can we leverage an understanding of networks for better political polling and prediction? What are the meaningful differences between one’s values and one’s affiliations? And is the American two-party system working for or against a cohesive republic?
If you value our research and communication efforts, please consider making a donation at santafe.edu/give — and/or rating and reviewing us at Apple Podcasts. You can find numerous other ways to engage with us at santafe.edu/engage. Thank you for listening!
Research we discuss in this episode:
Falling Through the Cracks: A Dynamical Model for the Formation of In-Groups and Out-Groups
A Sampling Model of Social Judgment
Harvesting the wisdom of crowds for election predictions using the Bayesian Truth Serum
Why are U.S. Parties So Polarized? A "Satisficing" Dynamical Model
Do two parties represent the US? Clustering analysis of US public ideology survey
Project Page for the SFI-USC Dornslife Polling Research Collaboration
For more on social cognition and collective decision-making, listen to COMPLEXITY episodes 9 with Mirta Galesic and 20 with Albert Kao.
Follow us on social media:
Twitter • YouTube • Facebook • Instagram • LinkedIn
Join our Facebook discussion group to meet like minds and talk about each episode
Podcast Theme Music by Mitch Mignano
(Time code starts after the introduction. Transcript edited by Rayyan Zahid.)
Michael Garfield (0s): Vicky, Henrik. It's a pleasure to have you on Complexity Podcast.
Vicky Yang (5s): Thank you so much, Michael.
Michael Garfield (7s): I would love to have a kind of roots, trunk and branches type of structure for this conversation. Starting with your backstories as researchers, like how you got into doing the kind of work that you're doing at SFI and what motivates and inspires you to take on this kind of research.
Vicky Yang (29s): How far back are we going?
Michael Garfield (33s): Wherever you feel like the story starts for you.
Vicky Yang (37s): Okay, I'll go first. I have a shorter story to tell than Henrik, probably being around for less. I grew up in Wuhan, China. I was born there, sadly, my hometown will always be known for COVID-19 while it does have many other good things. I came to the States when I was 19 years old to go to college. I started with studying physics and applied mathematics because I liked the clarity and the rigor of the quantitative methods and knowing things for sure.
At the same time, I also volunteered in a social psychology lab, and I was running human experiments to study implicit biases about gender and people with foreign accents. I was fascinated by that line of work because I feel like there are ways we can talk precisely about human biases and things close to your everyday life. Later in grad school, it occurred to me that these two lines of interests merge.
I started looking at human behavior through mathematical modeling and the data driven way. I was applying the methods that I was in love with to solve the problems that I was passionate about. So that continues till today.
Henrik Olsson (2 m 6s): Okay. It's my turn. I grew up in Sweden and all the parts of Sweden were close to the Arctic circle. Very dark in the winter, very, very light in the summer. I actually didn't know what to do after I finished high school. And then all of a sudden I found myself applying for Uppsala university. I went to Uppsala university and there I got interested in political science.
I started to think about applying to the doctoral program, political science, after I finished my undergraduate thesis. The thing was that I also had an interest in psychology and also did undergrad studies in psychology. I have a double major in political science and psychology. And the thing is that the doctoral program then was too difficult to get into as compared to political science. That's your psychology, that's the way it is. And then I started to become interested in how people make decisions, in political science and psychology, because I've been part of a strong movement in both psychology and political science, economics about using a kind of rational actor model to try to understand people that way.
That would become my background in the rational choice tradition and game theory that started my interest in decision-making. When I started my graduate studies in psychology, I sort of became interested in human judgment, human decision-making more broadly, and also specifically how people make judgments about subjective probabilities attached to future events and actions and so on. That led my interest in computation modeling.
In the early days, judgment and decision-making was not so much on cognitive plausible models. It was mostly based on simple verbal theories, important from economics and so on. We try to apply like a more collective psychology or cognitive science perspective to judgment and decision-making. That started my interest in computational modeling.
After that I got the opportunity to start as a researcher/scientist at a Max Planck Institute in Berlin, Max Planck Institute of human development. And there, I really got into the importance of interdisciplinary or transdisciplinary research, which is of course a hallmark of SFI. We had a team of biologists, economists, psychologists that worked on questions and started to solve them together. Not trenched into different disciplinary boundaries, but actually working together on trying to understand decision making.
All of a sudden, a former self, or actually my wife had found herself getting offered a job at SFI. And back in 1993 when I started my doctoral studies, I read this book about SFI and I said, at one point in time, I will come there and I will do research at SFI. That dream came true. Five years ago, we moved there and came to SFI. It broadened the horizons even more because before I haven't been so much into contact with physicists and mathematicians to learn about all those tools that have been developed and used by the system mathematicians, especially when it came to belief dynamics.
So that started the interest in beliefs and how beliefs can change or not change. What makes people change their beliefs and what makes belief more resistance to change? We have now started several projects where we try to investigate the mechanism behind that and try to mold it based on insights from a wide range of disciplines.
Michael Garfield (6m 16s): Excellent. So just to call the shot before I take it, I would like to weave back and forth between work that you've done with other researchers, each of you, and then work that you've done together. And I would like to start with Vicky, your work that you advised on for a paper with Louisa Lee and Siyu Zhang on whether two parties represent the United States, then I'd like to proceed to the manuscript that the two of you are working on right now with Tamara van der Does about falling through the cracks, modeling the formation of social category boundaries, as a way of understanding the results of that first study.
And then Vicky to talk about work that you've done with Daniel M Abrams, Georgia Kernell, and Adilson E Motter recently on why it is that US political parties are as polarized as they are. And then Henrik to turn to a paper that you wrote on harvesting the wisdom of crowds for election predictions using the Bayesian truth serum with Wandi Bruine de Bruin Mirta Galesic and Drazen Prelec. But yeah, so I think that the structure here is the lay of the land, what does the American political system really look like quantitatively when we analyze it?
And then why do we see a difference between the position of us political parties and people's actual political positions, and then how we relate to one another and how that bounds up in identity and as you were saying, the sort of cognitive dimensions of this. And then how we can start to understand what the political results of this deepening polarization are likely to be like how we study these changes in the future of the elections and so on in this country.
So just to start, yeah, Vicky, I'd love to talk about this paper that you advised on. Do the two parties actually represent the United States. And I would just love to hear you talk a little bit about the way that this study analyzed the data and the results that it came to.
Vicky Yang (8m 32s): Yeah. This project started as a summer undergraduate research project. The two authors on the paper were undergraduate students at Northwestern and Louisa who led this paper just finished her freshman year. She is an incredible, impressive person who did all the heavy lifting in this study. In this study, we took a national opinion survey that surveys the people on concrete issues over policy, for example, should we have government health insurance or private ones?
And there are about a dozen of these issues. And we had to do the survey over time for, I believe back to the seventies or eighties. This paper is using machine learning technique of supervised clustering that says using the public opinion data that, if I ask a computer to characterize it by two clusters, where are the clusters? Using this method, we find what’s the best description of the data that we have in the political landscape and what is best described with a big centrist cluster and a small right-wing cluster.
We also asked a question where additional parties were representing a system. We looked if introducing additional clusters would allow models to have more clusters to describe the data, and we use a model selection method that asks how well this describes the data while penalizing for the additional parameter that we introduced. And we find a yes! Indeed, including additional parties will help us represent the data better.
The current two class two description isn't the best and the precise number of best parties I think is arguable because the variation in how well it matches the data. The benefit of introducing a new class eventually starts to decay. But we know two isn't the best. We guess it is somewhere like between three and six, but we don't have a precise answer at the moment. And because we picked, these dozen issues may have different weights between them, which we haven't been able to figure out.
We are at a ballpark of where the estimate is.
Michael Garfield (11m 8s): I think that's pretty well established, at least in the court of public opinion that we don't have enough political parties in the United States. So this brings me to the latest work that I have from the two of you with Tamara on how social category boundaries are formed in the first place and how it is that, one of the more disturbing things that your model formalizes in this paper that is like very, very clear, I think on the ground is that, anytime you have a position that's sort of in between two poles of a duality, whether or not that duality is a fair representation of what's actually going, on each side will regard the people in between zone as the other.
Why is this, why are so many people falling through the cracks, not just in terms of political parties, but in terms of, you know, other axes of identity, like gender or race, why is it that at the individual level and at the social level, these kinds of unfair otherings are going on.
Vicky Yang (12m 20s): I'll take a stab at this and then let Henrik fill in. So social categorization is the process by which we want to motto “where do people draw the boundary between us and them?”. People fall on a continuum. They're not discrete boxes like Democrats and Republicans, but they have a continuous distribution from liberal to conservative policy positions. But out of this continuum, somehow boundaries are drawn where groups are formed and people into groups think oh these people are us and these are them.
And we wonder, the question we’re wanting to answer is where and how do boundaries form? The way we have approached this question is we first started with a model that describes the combination of two processes. For each individual they're influenced by two things. One is to tell people a part accurately, that is they want to put people far away from them in the “them” bracket and put some of them in the “us” bracket.
And then the second thing they're influenced by is they want to agree with other people in their group where the boundary is. They don't want the group to have very different opinions. It's a combination of cognitive and social processes. And if we allow each group to draw their own boundary, we find that these two processes lead to the results such that the boundary is more exclusive than in the middle of the spectrum. It's the combination of both things that leads to this asymmetry.
And then from that, we see that the middle becomes then seeing from both sides.
Henrik Olsson (14m 4s): The psychological background to this is that in many ways there's a lot of research in psychology about categorization. There's lots of different models, mathematical models that try to understand individual categorization of objects. Is this dangerous or not? Is this a cup, or is it a glass, stuff like? But when it comes to social domains, these formal models are not so prevalent, and they're not many models that try to integrate both this kind of internal code to process that we have and the social influence from other people.
And what we're doing here is try to do exactly that and which we're also doing in my other research. We're trying to integrate the individual decision-making processes; I believe updating processes with the social influence and the influence that your immediate or the larger social network has on you. I think that is one main contribution. In order to understand social categorization, we need to really look at both the individual and the social processes this goes on.
Michael Garfield (15m 16s): Yeah. One of the things that I think will feel kind of intuitively true, that you are able to validate with this model is that Democrats and Republicans both sort of perceive independent voters with mistrust. And that this feels to me like, again, this is not merely a political issue, that this is something that we see going back, you mentioned in this, in this paper, for example, people of mixed race, you know, or I think this is also true of people, of mixed cultural standings, people who are non-binary and gender.
Vicky Yang (15m 55s): Let me give you more information on the kind of data analysis we also did with a paper that also showed this behavior in the observations. We took political survey data, and we asked “how do Democrats and Republicans”, which are measured by their party registration of date they report, “how did people of these two groups see three different groups of people, members of their own party members of the other party and the political independents?”.
The data we're able to get was to question political independents. They only asked 80 and 84 and after 84, they threw that question out, which I'll come back to later. In that data, we find that Democrats see themselves very favorably, Republicans less favorably, that's predictable. And then the question is, how do they see political independents? If I didn't see that model, I wouldn't easily guess maybe they're in the middle of Democrats and Republicans.
However, in the data we find independents seen as sort of the same in terms of positivity with Republicans. And then the same thing happens with the other party. For Democrats or Republicans, independents are seen as unfavorably as members of the outer party, which means if you are independent, you got the worst of both worlds. That you don't feel you belong to either group. And you're not even in the middle of between my party and their party.
You're sort of considered as theirs on both sides. And then when I saw the results, this data is political attitudes, and I also learned by searching for these kinds of data, I realize there are so fewer people who actually learned about political independents compared to what they know about people on the two extremes. There is a very little data gathering about how independents are perceived and a more detailed picture of what goes on in their lives is needed.
And the second thing I'll say is that when I saw this result, I kind of can relate anecdotally to many of these things. For example, I grew up in China, but I lived in the US for a long time. In the US I'm considered Chinese. When I go back to China, many people tell me that I'm too Americanized. I'm sort of neither in a way. You mentioned, the study of, of multiracial people. One anecdotal example is Obama about how in the beginning of his political career he struggled between people who thought he was too black.
And people who thought he was not black enough. And that kind of behavior was also replicated in a number of experiments where they asked participants how they perceive an individual of mixed race, and why people think they're not white and black people think they're not black. They fall through the cracks. We think it is possible that this model can be extended to other attributes. And it will be great if we can gather more data on that. Henrik and I, with Tamra, we're working on experiments.
One is we hope to get a more recent picture of how political independence is perceived currently, which we don't get from this national survey anymore. And then we also hope to see if we can extend it to other attributes.
Michael Garfield (19m 38s): Yeah. Even in Santa Fe, I don't know how widely true this is in the United States, but even in Santa Fe, if you're registered as an independent, you're not allowed to vote at the local level. You know? There's lots of ways that this damage, like, again, you're talking about in that last paper, that enormous third cluster that just gets sort of punted back and forth.
Vicky Yang (20m 1s): Yeah. What I'd like to highlight here is there's a distinction between a group identity or identification with a party versus what you believe in terms of liberal versus conservative beliefs. These are two different things, but recently they're increasingly mixed up to feel like one thing, especially when we talk here in debates in the media about politics, but really party identification or group identification in general sits on top of this continuum of liberal conservative belief.
So how does group identification arise from this distribution is a non-trivial question, while we often think is one thing and where the boundaries are drawn are also affected by other forces of social and cognitive influences. It's not necessarily we're presenting the underlying distribution of what people believe on concrete issues.
Michael Garfield (21m 4s): Yeah. You know, you mentioned in this paper, one speculative example that you offer in terms of how people fall through the cracks is that the disconnect between issue polarization and social polarization and previous empirical research has found that identification with political parties and antipathy toward the opposing party increased disproportionately compared to opinion on issues. So it's like we're, you know, to sort of toss a line out to later in this conversation, Henrik, when we're talking about how do you pull people, how do you gather information, you know, in a way that accurately represents the landscape of opinion, there's this, this question of, I think both of you are familiar with the evidence that issues polling makes Americans look a whole lot more in agreement than asking the question of who are you going to vote for?
And that brings us to this piece on satisficing, right? And, and the polarization of American political parties that there's something going on here. Even if you assume, as you do in this, in this paper, Vicki, that the public itself does not change in their opinions over time, where they stand on the issues that the parties can still become more and more polarized. And this is set aside from issues like the ones we discussed with Carl Bergstrom and Jevon West on the show earlier about disinformation campaigns and search filter bubbles, and the influence of the media generally on the polarization of public opinion, that there's something else going on here.
Could you speak to that?
Vicky Yang (22m 37s): Yeah. I will first try to clarify a few different definitions of polarization that people tend to confuse and make this discussion really confusing. When people talk about the parties polarizing, they often refer to members of Congress or, or political leaders. And you can measure polarization in multiple ways, but for political parties, in multiple ways that we can measure polarization, it shows they have been polarized.
For example, people have looked at the co-voting network in Congress, how often congressional members are voting on the same side of the bill. And we find that congressional members of different parties are voting less and less across party lines. So that is one measure of polarization. People have also measured that in other terms. Some people can, they're just a quantitative method through which you kind of assign an ideological position, so we put every congressional member on a line based on how they voted in the past. And you kind of see how much this distribution is spreading out.
And that indeed also is occurring. However, we ask, is this polarized Congress, an outcome of a polarized public? That question becomes more complicated because there are two popular measures of polarization for the public, and they show different results. The first is sorting that is asking how much do you people's party identification corresponds to their liberal conservative sort of ideology to spectrum.
It used to be, you can have a liberal Republican and the conservative Democrats, so people can be on the left side of the space, but identify with the Republican Party that is happening increasingly less. If we measure polarization by sorting, yes, it's happening. They are also many scholars that argue sorting is happening to a lesser extent in the public that in the Congress. Also an outcome of sorting is that if you take out people who already identify as Democrat and Republicans, and you try how people who identify with these two parties change over time, you'll see the center of the two parts drifting apart in the public.
So that is the outcome of sorting. However, it gets more complicated when you look at the second measure of polarization, which is called dispersion. That is if I'm blind about who identifies with what party, I just look at everybody in the country and look at how they're distributed in this liberal to conservative space. And there is very little signal about systematic changes over time. It's really not clear if this distribution changes at all.
There was data that showed a fluctuation, but no systematic trend. The most likely picture of what's going on in the public is that the people have somewhat a one humped distribution with more people in the middle. However, the way people associate political parties with their liberal comes over that positions has been changing.
Michael Garfield (25m 56s): I wonder in thinking back to episode nine with, to, and talking about how people project a local information bias based on their their personal networks and then onto the global situation. I wonder how much of the difference in the distribution between the American public at large and Congress has to do with the fact that in Congress, you're sitting in the same room with people that, you know, or have a different opinion and sort of like, you know, calling back to this paper on, you know, the in between category that like, there's something about people having to define themselves against someone else sitting across the hall from you that is driving this phenomenon.
And in Congress, just noting that we haven't actually gotten the satisficing model yet, but I'd love to speculate and see what y'all think is going on there. Why is there this difference at that level or between those two levels?
Vicky Yang (26m 52s): Yeah. Does a satisfying model propose one way this kind of thing can happen, The way you said, like, if politicians are motivated to distinguish themselves from those over on the other party sounds very plausible too. I haven't studied that mechanism in itself. Somebody else probably has. I'll just give a brief summary of what the satisficing model is doing. Sort of the backstory of that.
There was a classic model called the down ??Zia model for parties. And it gives a very paradoxical result if everybody votes and if everybody votes for the party closest to them, and if the parties want to get the most votes, each party wants to be in the middle of the spectrum. That is not what's happening currently in politics. And one question is why would that happen? And also, what can explain this disconnect between the voters and the party. It seems like the voters themselves, if we take out their party label, they haven't changed very much.
And how can you have voters with mostly, somewhat moderate policy positions and you get parties that are totally not representative of them. The model assumes the satisficing behavior. Okay, let me back up one second. In most mathematical models that study voting on, they assume people have some objective function they want to maximize. They pick the party that maximizes the objective. However, in psychology, there's a lot of evidence for a satisficing behavior that is when people are faced with a complex situation, they don't optimize.
However, they settle for what is good enough. The biggest novelty of this model is that it gives a quantitative translation of this bounded irrational behavior. We are able to put that into mathematical forms, and then we latch the parties, optimizing their position in response to satisficing voters. In this model most parameters we actually predict the parties will want to stabilize at some finite separation.
They won't want to collide together. Like the ??Downsville model said the intuition about where they stabilize is entirely can be explained as two things. One is so basically when you, when people do satisficing, if you're too far from them, the voter may feel not represented at all by either party. And they just don't turn out to vote at all. If you are too much on the tail, you lose the people in the middle.
However, you go to the middle you, one loses the people on the tail, two you need to compete with the other party for the middle of voters. There is a balance between where you are, and then the distance between the parties is determined by one parameter. That is sort of how much ideological purity there is in the party. Whether your party is appealing broadly or appealing narrowly.
And the way we measure that parameter by the ideological spread of the congressional members of that party. We took that method that put every congressional member on the line, and we looked at how wisely they're being spread out. We infer the parameter from data that way. And we know, both the model and the data say that if you have narrower parties, then the parties would want to be placed further apart to win more votes.
So that the winning strategy becomes a more polarized political landscape If ideological purity is increasing.
Michael Garfield (30m 53s): So to Henrik, I'm thinking about the conversation I had with Rajiv Sethi very early on like episode seven of this show, when he was talking about stereotypes and how we lean on stereotypes more when our decision-making is under some kind of spatiotemporal pressure, you know, the dark alley in which you're meeting someone and you have to make a snap decision and we lean on, you know, our conditioning and heuristics rather than getting to sit down.
And this is like a classic anecdotally in terms of, "Oh, I'm not a racist. I have one black friend", you know, like I know this one person, you know, I have dinner with them, and they're great, but the rest of this out-group is not. And I'm curious what your thoughts are in terms of what Vicky was just saying, you know, as the decision becomes more complex, people tend to break down. And I think this would be a good spot actually, just to dig in a little bit more about the details of what you're talking about in this area.
When, when we're talking about bounded rationality, what is bounding this rationality, because this is a big thing also with the work that SFI does and economics, and like the way that we model economic systems. I don't know if you care to be our tour guide down this particular lazy river, but I think, you know, laying out some basic stuff about bounded rationality, and then how it's determining sort of where people start leaning on cognitive props in these kinds of decisions would be really interesting.
Henrik Olsson (32m 28s): I want to push back a little bit about this bounded rationality, because I think the general kind of view of or in many senses when people hear about the rationality that they think about something less than rational. I don't think that it's true because we talk about optimality. We talked about optimizing, but in real life, we cannot optimize. There is no objective function that can be optimized.
There is no way we can, we can ever come to the normative correct in many situations. We take actions, we make decisions, and we do not know what the future will bring. We do not know that when we make the decision, but we can define other aspects of UN other norms, other rationale norms. For example, how well do different decisions, strategies work in different environments, try to figure out when is it good to use satisficing. When there is a good reason to use decision-making.
Sometimes they work. Sometimes they don't work. And the thing is to try to understand when they work and when they do not work. And in many situations, there is no way we can calculate what the maximizing decision would be because there is no way we can know it. There is a, there is a non-quantifiable uncertainty there, even though we would like to have quantifiable as are there, but we cannot do it. that's one thing.
When I was at the Max Planck Institute in Berlin, one of the main foundations there was bounded rationality by Herbert Simon's ideas. But the specific interpretation of Herbert Simon's ideas is that it's not less rational. If it says good or even better, it can be a better guide to rational decision-making in real life than actually maximizing decision-making. I also want to add one thing with polarization and, beliefs and, actions and decisions.
I think also there is this distinction. We need to think about those beliefs and actions. And of course it could be that you can have a belief, an ingrained belief, but in strategic circumstances at the point of when you are making a decision in parliament, you must take an action, you must make a decision and you must vote for something because your party colleagues must see you in a certain way, because you haven't actually changed your belief. This has a strategic thing that you do in that circumstance.
And it can also be the same here. I am a Republican for my whole life, but I don't like Trump, so I will vote for Biden. But as soon as there is a reasonable Republican again, I'm going to vote Republican again. We have this kind of thing that you have the beliefs and you have the action decisions which need to be both whole. You need to consider both at the time because of different measures. If you ask people about beliefs, you can get one, and if you see their actions and voting behavior then they can look totally incompatible.
But you need to look down to the circumstances of the specific action of voting and the circumstances that took place and why that person might've been voted that way.
Michael Garfield (36m 1s): Yeah, that reminds me when I was talking with David Krakauer for the transmission series about David Kinney's article for that series where he was talking about how scientists want to offer to policymakers, the spectrum. You know, that it is like, this is how much uncertainty we have about the situation. This is the rigorous way of talking about it. But a spectrum of possibilities is less useful when you're making a policy decision. Then this is the most likely outcome.
And there's this sort of philosophical problem of how scientific advice even happens. I mean and like you said, this is true, not just in the advising policymakers, this is true in selecting them that, you know, the contextuality of these decisions, you know, forces a decision in certain ways. That seems, I think, you know, to call back to the beginning of this conversation very much like what's going on socially in terms of the excluded middle in these, at the group level, which is a seemingly inescapable tragedy in some respects that maybe, like you said earlier, Vicky, that what we're looking at is we just need better models.
I know that to instantiate them in society, you know, like I know a lot of people that are pushing for rank choice voting as like one way of getting around this kind of problem. But that brings us Henrik to the last piece I wanted to talk about with the two of you, which is the piece you led authored on harvesting the wisdom of crowds for election predictions using the Bayesian Truth Serum. And I think for those who didn't hear your conversation with Mirta, I think it makes sense to give a little bit of background on this paper, you know, to talk about how different kinds of polling happens and sort of the relative strengths of different kinds of election predictions, polling, and how that reveals through those different methods, the ways that people are taking into consideration, all of these different contextual restraints, you know, it shows how people are making these decisions.
Henrik Olsson (38m 10s): The last few years, it has actually been an ongoing conversation or discussion about polling is that it's got a lot of problems with it. And the last few years you've seen the response frequency of different surveys that contacts people are low. Less than 10% is going to answer your survey and what type of results can we get from 10% when you try to sample some part of people?
It is going to be a very biased sample. There has been lots of discussion. And also in the last few years, there has been some rather mis-prediction in terms of elections in Brexit and so on. And also, the discussion about the last presidential election about how polling actually didn't work well. On a national level, it worked well, but in important States in the US the polling didn't work at all.
So traditionally polling questions ask about intentions. Who are you going to vote for? That's the traditional way of doing it, but we can also try to leverage the wisdom of crowds of people and who they think are going to win the election. This type of question has actually been asked for a long time. They started so far back as the 1930s in different surveys, but haven't been widely adopted by the polling industry and all this aggregated result, all the polls you see now, are mostly based on onboarding intention.
Sometimes they report these election expectation questions, but mostly as a curiosity not as that real prediction. But the thing is that this winner's expectation questions, they are shown to actually predict election outcomes very well, but they're not foolproof. And one thing that is usually leveraged against the question is that there is a risk that you have a bias that you project on your own onboarding intention.
You get the bias that way. A few years ago, we started to think about how your immediate social environment kind of affects you and how you can use information you have about your immediate social environment to make predictions about different characteristics in the population. That led to a theory that I'm not going to talk about here, and that led us to developing questions that ask not about who you're going to win the election, but you ask who do you think your friends are going to vote for?
That's an interesting question, because the idea is that you can reach people that might not answer the poll question, the survey questions. You basically enhance the sample. You reach one person, but you get the sample to become larger and you can reach other people that you may not be able to reach. We try these questions now in several elections and show that they outperform consistently the voting intentions, and they all can also help you perform reputation questions.
So that's the background to the work we're doing now together with researchers from MIT and University of Southern California, where now have a project where we try to use this different question together with other more sophisticated ways of integrating information from these different sources of over information, present with these questions, to be able to make these predictions more accurate.
But in addition to that, we also try to understand what type of factors are behind problems with polling. There's lots of talk about the shy voter effect, for example. I don't want to admit to who I am going to vote for if there is some shameful candidate. But if someone asks me, I may say, I don't want to do that. But if I was asked about how others will vote.
Your social circle might be less, less susceptible to these kinds of biases or reluctance to answer questions. But then on the other hand, there have been arguments that this winner's expectation questions can have a large bias in them. It can also affect the social circle question. For example, there's talk about, especially now in the election here, but also in other elections that it could be that some of the candidates or the government in the country will do something before the election that will change the outcome.
They think something like the Magical Trump effect, for example. He will lose, so he will change something. He will propose something that makes the landscape change totally, and he will win. And then these kinds of expectations can then creep into these winner expectation questions, which might bias them because it might not happen, which might make them less reliable in, for example, in this election, in the US and there's been arguments in the media about that.
We try to understand now, try to ask other survey questions. We are asking a big national representative sample of respondents of up to 5,000 people around the US about these three different questions on voting questions, winner and reputation questions, and what we call social circle questions. When you ask them about how you think your immediate social environment will work. But in addition to that, you also try to understand there that if people think that something's going to happen just before the election, that was changed to assault somehow, or how much discomfort they feel by divulging this information.
About who they would vote for and how much discomfort do they feel when they are asked about their friends and family, or they are going to vote. We try to understand what can affect these different polling questions. And then in addition to that, we have done more or less explicitly in the beginning, more or less work with sophisticated ways of aggregating this information. And we are actually now working on other ways to integrate information, all these three questions.
we're actually now writing a new version of this paper that you mentioned that have a new and exciting way or doing this, but it's not prime time yet, but we're working on it
Michael Garfield (45m 31s): While we're awaiting the new paper. I would love to dig in a little bit, because I just think that this is, this is like a cool technique here. You know, one of your co-authors on this Drazen Prelec has written extensively about this, this technique, the Bayesian Truth Serum and how you help ensure that you're getting honest answers in polling when you don't have a way of independently verifying that the person is giving you an honest answer. I'd love to hear you provide an exegesis of this for listeners, if you will, how does this actually work in your study?
And then in what situations do it seem to improve the quality of the polling or where does it seem like it's not helping?
Henrik Olsson (46m 15s): The thing is, I mean, this Bayesian Truth Serum, as you said, that it's developed by and developed in the mid 2000 semester, I saw this article about that. As we use this in several areas, it's basically a scoring method, as I said for incentivizing truthfulness, in the sense of honest, informal, careful answers to any type of number of viable questions.
So basically, in a very simple way is that it uses the election winner expectations. You either use the question that you asked, the own intention question or the social circle question. The survey used the own intention question, but we are also using the winner expectation question and, basically overweight participants who provide more informed, more careful own intention questions. So that is basically what it's doing, and that is a technical thing about it that can be shown that it's Bayesian updating and so forth.
But what the battles are always participants who provide more informed careful own intention, and it uses its own intention question to do that in a clever way. And we see that it actually works, not in certain circumstances. For example, if we use the winner election expectations to overweight social circle expectations questions, then we see that it almost always improves over and above the social circle question.
We get an improvement. We get our own intention that predicts something, and then we have a winner's expectation, maybe a little bit better, but even better than that is the social circle question. But we can even above that, we can improve the social circle by applying the Bayesian Truths Serum.
Michael Garfield (48m 21s): But you mentioned in this paper that it doesn't really improve predictions for the margins of the winner.
Henrik Olsson (49m 0s): So that's something we try to figure out what it is, but the margin in the US is also particular for a two-party system that they use the margins in this way. And if you have a many party system and look at how it predicts overall all different parties in other countries, it would be a different view of it, but we're still trying to figure out exactly what is going on with margin predictions.
Michael Garfield (49m 0s): So maybe it's just my privilege of being this close to this kind of research that makes me feel like a lot of these discoveries are formalizing intuitions about how politics works in this country. What do you feel has surprised you about the findings that we've discussed today or, you know what it reveals about the way that people make decisions or what it might suggest about the future political landscape of the United States?
Vicky Yang (49m 34s): Okay, I'll go first. I guess one thing that I've learned that felt surprising that I wouldn't have known before I started doing this, is we always talk about how the US is a divided country, how we're so polarized. And it makes it feel as if people who belong to these two groups are just very different people who really dislike the other side. However, there's a distinction between which group you identify with versus what you actually think about concrete things.
And I feel the group identity is making us feel that we are more different than we actually are when we really boil down to what we think on actual stuff. And the analogy I like to draw is with sports. You can have Yankees fans and the Red Sox fan, for example. There are two groups, and they have a strong identity of the group, and they don't think about members of the other group.
However, if you really look at the people and how they think about the sports, they may not be that different. There's like this sort of fandom I suspect in what we're seeing. What we're seeing in the group conflict is very real, but it doesn't mean people of the two groups are at the core of different people.
Michael Garfield (51m 5s): Yeah. That seems like an important takeaway while things are in some sort of social centrifuge right now, kind of, a bomb on a wounded country in that respect. Henrik, what about you?
Henrik Olsson (51m 17s): Oh, that's a good question. I think so, so I started out doing research on individual decision-making. Individuals who sit in the lab make decisions, but the last few years I've come to realize that is the most important thing that we have. And it's also been informed by all the social structure, the social environment, and how important the social environment is. And that we need to put even more effort to understand the influence of the larger social environment on your belief in individual decision-making.
And also, how well actually people know the social environment. And I think we come to realize that we cannot talk about individuals as an isolated Island by themselves and need to talk about their network. And of course, that's what we talk about all the time at SFI. We need to see things in their perspective and their network connections. And I think that is the way we need to go forward with, especially in psychology, which has not been so much interested in the actual social connections between people and investigating the actual reality of your social environment.
We have an interest in reading the nets of other people, Tom is this and that, Louise is like this and that and then you have some reaction to that. We haven't actually investigated the real social environment with people. And I think that's the major insight. And also, that in order to understand it, we need to even more focus on your immediate social environment and the impact you have on individuals.
Michael Garfield (53m 3s): Yeah. You know, to that point, just looking over the USC Dornsife election website, which we'll link to in the show notes, I love that there's, you can break down all the different polling graphs that you know you're taking and how the further you get from the independent party affiliation being, I guess, in the middle, you get the strong Republicans are saying, Trump's going to win by a lot. The lean Republicans are predicting less and the same is going on in the opposite way on the democratic side, but then you've got it right in the middle.
You've got that band of independence or no affiliation where the margin that they're predicting is so small, it's within the gray area of like the demilitarized zone of no statistically significant results. I don't know if you agree, but that seems to speak to what you were just saying. That it's sort of a, there is again to, you know, to cast back all the way to the first paper we discussed here, that there's a sort of like optical illusion in which it's like hard to see into, and maybe even like out of the middle in these situations.
Henrik Olsson (54m 8s): Yeah. And they're also on the website that we'll see this kind of quite big discrepancy between the different predictions from these different questions. If I look at my own intention, it looks like Biden has quite a comfortable lead, but then if you look at reputations, then that goes to a very basic known significant difference between the candidates. And then we have the social circle question we've done. So a little bit bigger leap, but still come on narrow lead for item that can have many, that could be the truth, but it can also be a reflection of what we talked about earlier about that people are going to expect something's going to happen before the election that happened 2016, everyone was surprised it's going to have to begin.
This winner's reputation question may be underestimating the Biden lead, but we think that the social circle question is less susceptible to that. If I will bet my money on something, I would bet on this then it is the social circle question, given that.
Michael Garfield (55m 8s): So just because we've reached the top of the tree, and we're touching the sky here, I'd like to end with a more speculative question for the two of you, which is, as you mentioned in Vicky and Henrik in this piece, falling through the cracks. That again, the boundaries of where we delineate our social groups have changed over time. I mean, if you look at, you know, LGBTQI, that whole thing has exploded over the last few years, you know, like the census is including so many different racial answers than it used to.
And I'm curious, you know, to me, this looks sort of like if I zoom way out, it looks like the tree of life following some sort of maximal entropy production algorithm, you know where it's like, you know, evolution is sort of through the you know, different adapted anatomy is trying to solve the puzzle of how to model its own facts. You know, that it's like the biosphere it's, you know, modeling itself the way that we model the economy from within it.
And so something like this is like, what do you imagine based on your work for the future of the American political system, do you think that we're going to continue to have a two party system, or do you think, do you think that we have a realistic bid at going more of the way of like the Australian parliament where, you know, there are like three major parties in seven minor parties. Do you think that we're going to order like call to Stephanie Crabtree's work or the essay that she wrote for the transmission series about the decline of Chacoan culture and how certain things in more sort of stable times are able to kind of homogenize and conglomerate because of economies of scale, but then if you disturb those systems, then they break apart?
And I'm, you know, I'm wondering, you know whether what we're seeing right now is actually sort of the foreshadowing of a fragmentation of American political parties as a way of addressing some of the problems that we're seeing right now. And what are your thoughts on all of that?
Vicky Yang (57m 19s): Can I give you a somewhat rambling answer?
Michael Garfield (57m 22s): Yeah. I think that was the rambling question.
Vicky Yang (57m 25s): Okay. I'll give you two answers to like two aspects of a question. One, are we likely to see a multi-time party system? Unfortunately, I'm not optimistic because, with the role of election, as it is the plurality winner take all election is known to favor two big parties. While in most of Europe, they have proportional representation. You win 5% of the vote. You got 5% of the seats. The system means in the way it's set up is helping the two-party landscape persist.
And then on the local level, there are also a number of bills recently introduced that make third parties or, or independent candidates make their lives harder. Like you have to gather tens of thousands of signatures or, or things like that. I think it will be really helpful if we have more parties, but I'm not very optimistic that this will happen. I think the rank choice voting will be really helpful if that can be implemented on a bigger scale. I know there are many state and local elections using that.
And I think that will really help the representation problem. And then I'm going to give a rambling answer to a second question about what we see in the political future. Like currently one reason why like this, this world seems such a mess is because you have very strong group identities. And there's a very strong idea of like, they're like immigrants are seen as the personal representation or the scapegoats of many problems on the scene.
It goes back to reading reasonably read about like a social theory, which is still hypothesis is not confirmed. The other big thinking along the slide. It's about how, where does this kind of thing come from? And it's this strong group mentality, populism, and this basic want to be separated from, from other parts of the world. There's a hypothesis that this is a response to the increased globalization and urbanization.
We've seen over the years that there the world is increasingly connected. And then that's why we not only see Brexit, but we also see, you know, cause for Scottish independence and then Catalan. So that is really groups trying to draw a line between themselves and the interconnected world to preserve their identity. And some people also argue, there are also groups of people who are left not as better off as the people who really benefited from the increased globalization.
And when these people are unhappy and then you have a political candidate coming and saying, you are unhappy not because of this big systematic problem, nobody knows how to solve it. You are unhappy because of these people. And they give a face to the reason for your problem. And that becomes a very compelling narrative. I feel like addressing what we're seeing is a very big systematic question about this whole economic trajectory we've been on since the industrial revolution.
That's my wild speculation. I don't know exactly how to approach it and think that's something I'm thinking about right now, but I wish I have an answer. I don't, but I would love people to discuss more on that line.
Michael Garfield (1h 1m 6s): Yeah. Henrik, what about you? What are your thoughts on all of this?
Henrik Olsson (1h 1m 10s): Yeah, the first question about the multi-party...I think, as we get across the election system in the U.S. that is basically guaranteed to produce something like a two-party system. And also, in terms of the history and the context of the constitution, the amendments, all of that together, the history, it's very unlikely to change.
I mean, that's not going to happen for many, many years. Because there's so many things that need to change for that to change. You said about the connectedness, I think we feel this is a messy situation, but it could be that we feel it's messier because we get more information.
We may know more, I mean, in early times we had less information about what our leaders, our elected leaders, or other assistants in the country did. Now we get updates every second about everything. We get the information about their things. Everything looks messier. it's good to just be that it was equally messy, 50 years ago, 100 years ago, but we just didn't know it. And we got less information here.
We get so much information in every second that it just looks messy. It might be that under other administrations, some other times it was just as messy, but we just didn't know it.
Michael Garfield (1h 3m 15s): Right when I spoke to Carl Bergstrom and Jevin West said something similar last week when. Carl was saying he thinks that largely the accelerating news cycle is a manufactured phenomenon and that it doesn't actually impress upon us the need to make decisions at that much faster than we did before you know, web 2.0. Good information from two days ago is more valuable than unreliable information today.
And it sounds like, you know, he was kind of advising that we slow down. We exert patience. That we think that the more we allow the latencies inherent in our collective sense making and decision-making processes, we give them time for these things to percolate through the system, rather than try to zoom in and look at the coastline of Britain with our inch stick, or our centimeter ruler, you know that.
So certainly, you see a lot of fluctuation in the polling and when you're polling people as frequently as you are in the USC life study, I don't know. That's probably just cut that all out. Cause everything you just said was great. It's all super interesting stuff. I'm really glad that I got to talk to the two of you today. Do you have any final thoughts before we leave people, anything you want to sort of leave lingering in people’s minds when they sign out here?
Henrik Olsson (1h 5m 9s): Vote in November.
Vicky Yang (1h 5m 16s): I can tell people that Wuhan besides having COVID also has really good breakfast noodles, great lakes and cherry blossoms.
Henrik Olsson (1h 5m 52s): Actually, I cannot say that because I'm a foreign citizen. it will be an interference by a foreign entity, if I say something. That's actually true. A foreign citizen cannot tell someone else to go vote, that's not in the US I think so.
Michael Garfield (1h 5m 54s): Alright, let’s check, can a foreign citizen encourage voting?
Vicky Yang (1h 6m 2s): You can't put money into it, but you can do things, is what I remember. But I'm not a hundred percent sure.
Michael Garfield (1h 6m 11s): Yeah. I'll look into that. This has been fun, and I really appreciate the two of you taking the time to discuss your research. I hope that this has been as illuminating for our listeners as it has been for me. And I wish you the best with following up, furthering your processes of discovery. Thank you.