 Welcome again for the second day. So before I leave the floor to the chairs and the speakers. So for the ice breaking session, I mean, there were a couple of people that wanted to present themselves. So we are going to do that after the coffee break, even that we have a long session. Should I say something about the? So hello, good morning. We will start the session with the talk by Professor Andreas Flage from the Faculty of Behavioral and Social Sciences of the University of Groningen. So please. OK, now I'm unmuted. Thank you very much. Good morning. Thanks for being here this early. Yeah, I'm pleased to have the opportunity to discuss and share some of our work on these topics here. So mainly it's on social influence and polarization. Inequality is not really my main research field. But I think, and we have discussed this, of course, a lot in recent days already, that it's clearly related to polarization and social influence, the topic of inequality. I would like to start with a few thoughts on how it could be related. So arguably, our society has a bunch of potential fault lines. I mean, here you see a real fault line. Let's see here's a pointer. Doesn't work anyway. There's a fault line, which doesn't mean that the Earth will split at that place. But there's a potential for it to happen. And in society, arguably, different groups who are clearly distinct or see them perceive each other as clearly distinct, maybe having distinct different interests, different worldviews, different values could begin to see each other very negatively, have prejudice, have negative attitudes about each other. And some of these potential fault lines are the higher versus the lower educated, something we seem to see increasingly becoming a fault line in some societies. Natives versus migrants, longstanding debate. Rich versus poor, another potential fault line. Or conservative values versus progressive values. Think only about the debate about woke versus conservative. If these dimensions of differentiation are correlated with each other, sometimes called consolidation, some call it demographic fault lines, then the potential for this to become a societal problem increases. So say if the same people who happen to be higher educated also happen to be rich and happen to be, say, politically progressive, that makes these distinctions and these differences even more visible and stronger. And arguably, of course, this can increase if inequality between these groups in things like income and so on, power perceived influence in society increases. Sorry, can I ask a question? I'm here. You can. You can. So is there any evidence that this correlation across time, I mean, is there any data showing how the correlation across dimensions is changing over time? Yeah, well, as far as I know, there is data showing that educational level and income differences are increasingly correlated, for example. For, say, the Netherlands, I'm not sure about the US. OK, some people are concerned about this, these potential distinctions, fault lines are being related to people retreating in their own bubbles. Here I have one voice from the president of the German Federal Republic. But of course, you can invoke many others here to say, OK, if people perceive each other as increasingly different, don't see common ground, they may retreat in their own bubbles, only echo their views they agree with, and maybe by that become increasingly feel increasingly negative about each other. I think that doesn't need a lot of additional explanation. We have also discussed some of these things already here yesterday. At least a bit of evidence that this is playing an increasing role in the US, but it seems also in some other societies, is here from a study of, well, this was published by Gansko. It's based on various data sources, including Yuguf data. So these are ratings of Democrats versus Republicans in the US of how they perceive the political outgroup in 1960 and in 2008. And then if you look at positive traits like intelligent in 1960, you would still think that political outgroup members could be somewhat intelligent. And you would not see much such a big difference between the political ingroup and the political outgroup. But in 2008, this has become a huge difference. And the same on the other side with negative traits, like being selfish. So these are fairly strong shifts in how negative different groups perceive each other. This is not own, well, this, let's say, mutually negative view of each other is not only present in the US. There are, for example, studies. Hey, why does it not move? No, it's got stuck here. Oops, now, oh, something happened. I didn't want to be here already. I want to be here. So this is from the Netherlands, a so-called feeling thermometer studies, survey study done by a Dutch political scientist, where people were asked to rate members of or adherents of other political groups on a scale from negative to positive, 0 to 100 degree, where 50 is neutral. The pattern you see overall is one where, say, in the middle, and that is, of course, not surprising, these parties are ranked roughly from their left to right position on a political scale. Roughly in the middle, we see, well, not too low temperatures, but if you compare the outer ends, how the outer ends feel about each other, so the most leftist about the most rightist and vice versa, then actually the temperatures are lower than in the US for the Democrats and the Republicans. But on the whole, if one would calculate measures of, for example, the degree to which this falls apart into two distinct camps, disliking each other, liking themselves, this would be clearly less polarized than the US because, for some part, there are so many more groups here, and they are so much more spread over a left to right dimension. But these distinctions are not just showing up in how people perceive each other and how they feel about each other, but for example, also in structural ways. Here's residential segregation, and we will hear much more about it, I guess, later. So, well, this is in the city of Chicago, but again, we can draw many pictures from many different places in the world which would look like that. This is so-called racial segregation, and it shows clearly similar people tend to sit together in similar parts in the same regions of the city. And there is a little mixing, at least only in some areas, there's a bit of mixing. This doesn't have to be a problem. There are many structural reasons for that, like income differences and so on. But it could become a problem if these groups think increasingly negative about each other. So some questions that could arise here are these. Homophily, we have already talked yesterday about homophily. So this tendency of people to interact more with others who are more similar to them, also like those people more, maybe trust them, trust more into them, accept their views more than those of others, or reject those views less likely than those of others. Together with the structural segregation can arguably foster polarization in society and can foster it more if these differences become more clearly visible. For example, due to increasing inequality, it becomes more clearly visible that, say, the higher educated people also earn much more than the lower educated people and have all these progressive views that lead to, say, policies or support policies that may be perceived as having negative consequences for other groups in society. So one can ask how resilient social integration and consensus in this society on fundamental values then can be, or, say, if we would have polarization, how depolarization would be possible. In what follows, I will go through a set of, say, very simple models of fundamental social mechanisms of social interaction and discuss, well, which kind of answers they could suggest if we look at these kind of questions. So let's start from a very fundamental social regularity, if you want, in interactions between people's social influence. This is a quote here inspired more or less by, or a definition inspired by, Robert Axelrod, who says, well, I'm very open and generic about what social influence actually is. So it's basically everything that becomes more similar to this trait in another person if you interact with that person. And you can think of opinions. You can think of behaviors. You can think of music preferences or whatnot beliefs. But there's a lot of evidence from experiments, from different kinds of field research, suggesting that on the whole, we have this tendency. If we interact with other people, they are different than we are. We tend to become more similar. This is probably not the whole story. But it is an important part of the story. So one could argue, if this is the main driving mechanism in social interactions, then why don't we all become the same after some time? Well, models of this, I call it here, assimilative social influence suggests, indeed, like that. So there are analytical results on this. There is computational modeling work on this. Basically, if you make assumptions like people have opinions on a continuous spectrum, and if they interact, they move towards each other, then in the long run, as long as the interaction networks are connected in the sense that as a path from every node to every other node, we will see consensus according to those models, which is not what we seem to see in reality. So that raises some questions. Actually, for example, these questions. In one paper, we dubbed them XOROT's puzzle and Abelson's puzzle. So two people are thinking about this computation. The modelers, both of them also, or mathematical modelers. So if we have these tendencies, why don't we all become the same? Basically, XOROT's question, Abelson's question, even asked a bit earlier, how can we understand that so often, and this is in the US in the 1960s, in communities, you see these cleavages, groups falling apart, having very different views on lots of important issues. OK. So what could we do here? So one model we proposed some time ago in work with Michio Meis, Kara Tarkatch, was, I call it now here, the model of the bubble intuition. So this intuition, if you retreat more in your own group and you interact with people who are similar to you, in the end, what we will get is polarization. Broadly, the idea is people have a vector of arguments pro-con certain issues. That the balance of pro and con represents their opinion on a certain issue. The more pro you have, the higher the number, so to say, representing your opinion. If you interact with someone, you will adopt an argument from that person. It could be a pro or con, but of course, the more pro the person is, the more likely you get a pro argument and the other way around. Then there is homophily. So not all people who influence you are equal. This is different from the classical models of social influence, where everyone has the same. Everyone who is connected to you basically, no, I should say, everyone who is connected to you has influence on you. The degree of influence can differ. But here it is also that you may just not interact with some people after some time. So the more similar you are, the more likely you are to interact. But if you're sufficiently dissimilar, you may not interact at all anymore. Well, this is basically the main idea of this model. So it implies that if similar people interact, they jointly move towards a more extreme position because they get confirmation for their views from the other side. It is different from the axarot model in the axarot model if you interact with somebody else. You become more similar, but the axarot model doesn't have a continuous opinion space. It's a nominal opinion space. OK, well, here just a little illustration of, sorry? That's not here in this model. It's only the more similar and the more likely I am to interact. But in some other models, we also have that. So here we have a situation where, say, in terms of demographic traits, two groups are totally different from each other. So this is this demographic fault line. There comes, say, a new issue on which arguments initially are randomly distributed in the group. And now people start to interact and exchange arguments based on these rules I was just discussing. And roughly here, the gray lines you see mean there is still a positive probability of interaction between nodes. And their location in space represents how close or far they are from each other in this interaction network, which is aligned, of course, after some time with the network of opinions. And then, of course, this takes a while. They exchange these arguments in the process gradually and slowly. These two groups sort out, so to say, if you want, it takes a while. But then if we wait long enough and are patient enough, then at some point, they apologize. Of course, this is a very simple model. So what we did in other work is trying to put this into a slightly more realistic setting. And now, actually, it's spatial setting. So I'm still in the right session. We varied here the degree of spatial segregation between two groups with a shelling type algorithm. Very similar. The model is slightly modified here, but essentially still the same assumptions. And then, yeah, this is what you see, not surprisingly, but confirming that also in this more complex spatial setting, we would have this process that strong segregation leads to drifting apart of the opinions in the two different groups. Of course, this doesn't happen always. It doesn't happen under all conditions. It requires, let's say, the right region and the right spot in the parameter space to happen. So here are two conditions we can draw from this kind of model for under which conditions polarization would arise. And these two conditions are sufficient homophily, must be sufficiently strong. So the difference in the likelihood to interact with someone who is similar to you as opposed to someone who is dissimilar to you must be sufficiently strong. And there must be some initial bias. It doesn't have to be very much, but the one group already must lean in the one direction in terms of its distribution of arguments and the other group a bit in the other direction. And then in the process, this will amplify. Still, with a simple model I showed you at the beginning, even for the most extreme combination of parameters, this polarization occurs only in about 50% of the runs. In the other 50%, after a while, consensus will emerge. OK. Nonetheless, from this so-called persuasive argument theory that we've modeled here, one can draw some tentative conclusions, which could be, OK, if we have this combination of strong homophily, initial opinion bias, maybe spatial or network segregation, polarization could be a likely outcome. And then, if this is the case, it would be a good idea to desegregate in the sense of, for example, bringing people from different groups together to talk to each other, to exchange arguments, because that breaks through the homophily. Tempting now to relate this to ideas like citizen councils bringing people from different societal groups together to debate important issues, maybe that's part of the reason why that could help to depolarize society. OK, now I have to check with the chairman how we are on time. We started a bit late, but I'm seeing that we technically I would still have five minutes, but I think I have a bit more. Sorry? I still have two 20, including discussion. That is? Yes. OK. So let's try to have time for discussion. OK, so far so good. One could say, but not everyone agrees with this idea. Desegregation, I think? Yeah, I mean, the simulation begins again. Yeah, that is what happened in the simulation, indeed, to could build in something like, OK, and now you will interact with this person from the other side, even if your probability is very low. And then you will get an argument pulling you in the other direction. In the real world, it could mean something like, OK, although now all the physicists talk with each other and all the others talk with each other, let's now sit some of them together at the same table and lock them up for half a day or so. Anyway, not everyone agrees with this. So this morning, we were already talking about social media, and of course, this is a famous representative of a big social media company. Well, this was, I think, in the situation around the congressional hearings that Zuckerberg made this statement here. Well, maybe if you force us to de-segregate, so to say, our network in Facebook, maybe it will backfire. Maybe, actually, if people are confronted with outside foreign views, it will make them more extremely rejecting those views. And actually, he could point to some research pointing in this direction. Many of you here know all this work probably much better on social media. But what I get from it is that there is some evidence pointing in this direction that if people are exposed to foreign political views, they may more extremely reject them, though the evidence is a bit shaky. And as I will say later, we cannot really replicate this strongly in experiments that we did. Anyway, the idea is not so new. Actually, one could derive it from classical behavioral theories, like, for example, social balance theory, cognitive dissonance theory, and so on. This general idea that if someone is an enemy negatively evaluated, then it's sort of psychologically costly to agree with that person. And it's more pleasant to disagree with that person. And it's pleasant to agree with my friends. So if we put this into a model and we did that, then you can get very different dynamics from what you have here for the persuasive argument model. Now, if similar people talk, they will come together. But if dissimilar people talk, they will move away from each other. Technically, that could look like this if you then also bring in differences between groups. So basically, this is derived from the very old social influence models. If two people interact, and we can generalize this to if I interact with a set of people in my network, then for every interaction partner, I'm pulled in the direction of the interaction partner with a force proportional to our disagreement. That's the classical models. But then what we introduced in these models at some point is the assumption that this weighting factor here can actually also become negative. And if it's negative, you're pushed away. And in some later work, we added that it's not only whether it's negative or positive, and how much does not only depend on the disagreement, but it also depends on whether we belong to the same group or not. And of course, the magnitude of this parameter here says how strong is this output projection. So this is how it could look like if this is the discrepancy, let's say the sum of this disagreement and same group or not factor. Oh, sorry, no, this should be the opinion distance because the same group or not is already in the two different curves here. So for the in-group, you need more disagreement to reject the view then for the in-group, you then for the output. So for the output, a relatively small degree of disagreement in this parameterization would be enough to actually reject their view and move away from them. Again, what these models produce is fairly intuitive, right, here's one example. So if we start from a slight initial opinion bias in both groups, not much. If the output projection is sufficiently low after a while they come together, even if they reject the output to some extent, but if it increases a certain critical level then we see polarization and we see the polarization aligned with the group boundaries, right? It doesn't, in principle, it doesn't have to be aligned with the group boundaries. The group could also polarize in other ways or the population if the initial variation in the opinions would be large enough. But if the output projection is strong enough then the group distinction will become the fault line along which this population falls apart. You can put this into, again, a spatial setting which we did in some papers. Yeah. In various ways. And again, and now the conclusion would be the other way around, right? And now it would be desegregation is not a good thing. At least if the output projection is sufficiently strong. So if we have a very segregated world in which the two groups interact rarely or not at all, maybe, we have a low potential for polarization. Again, it depends on lots of parameters but in reasonable regions of the parameter space. We did this in different ways in different settings but the result pretty much is the same. Actually, in one study, we moved to taking a real spatial setting. So this is a piece of Rotterdam. And on this piece of Rotterdam, we took from data from the Dutch statistical office. It's the 100 by 100 meter data, not the micro data. The local density of non-Western migrants. And this is now all theory. Please don't get me wrong, this is all theory. We thought, well, let's assume they disagree or agree on a certain opinion issue with each other. Everyone in the population and they have this initial distribution in which there's a slight bias of the one group to be on the one position, the other on the other position. And now we run the same dynamic on this map here. Then let me stop it here. Then we would see for a while that actually most adopt a fairly neutral position, that's great, but here in this area with a high degree of ethnic mixing, so a lot of interaction between people from different groups, we start to see people becoming extreme in both directions. If we continue from there, then after, what is it now running? Yes, it is running. After a while, these extreme views are spreading in space because people interact also with others who are similar to them and spread their extreme views in space. Now, please really memorize. This is theory, this is not reality, hopefully. But we wanted to see whether the mechanisms such as robust if we go into a more complex setting, yeah. How is the opinion related to non-Western immigrants? Well, it could be what we called an ethnically salient issue, like should we change migration policy? Should we impose harsher rules on immigrants to, I don't know, learn language or whatever? And you assume that the immigrants, non-Western immigrants and Western immigrants have too extreme opinion? Well, not in the beginning. In the beginning there's only a slight, the one group is a little bit more in favor, the other a little bit less in favor. But there is this out group rejection factor that we assume, impose theoretically, right? That's a theoretical assumption. And it could be something else, right? It could be high educated, low educated or whatever. Yeah. Yeah, so this network is connected in the sense that it can propagate through the network. So if in this, let's say, highly mixed region people become extreme, they are influencing others in more moderate regions and then these others tend to move in the direction of the extremists. Yeah, there's something built into this. If you look at the details, yeah. That is, yeah, because the non-immigrants are here in the majority. Yeah, you're right. Yeah, that's the point. Okay, let me skip the following slide. I think the point was clear. But this, I think, is something that should be mentioned. So, okay, this was theoretically playing around with this idea of repulsive influence or out group rejection. But actually we did a bunch of experiments trying to test this assumption. So for example, exposing people to opinions of others and then asking them how that would change their own opinion on something and giving them information suggesting that the other belongs to an out group they don't like, which we knew before from other questions we asked them before. We did this in offline lab experiments, colleagues of myself did this in online experiments. More recently we did it in online experiments with groups recruited from different ideological groups on Facebook. And currently we study co-evolution of networks and opinions in a school class. This is a paper about music tastes, the development about music tastes in a school class. And on the whole we find not much support for this negative repulsive influence. That's a bit in the more sophisticated studies we did here. But it's relatively weak compared to the overall tendency of people to move towards the opinions of others even if the others are relatively dissimilar to them. But if it exceeds a certain threshold it seems then there is some of this repulsion. But whether the repulsion is strong enough to actually generate polarization is another matter. And then for example this paper we show that what we find in the experiment if we stick it into a simulation model we don't get polarization. Even if we desegregate and debubble interacting a population. Yeah, sorry? Yeah. Yes, yes it will very much depend on the beta values. Yes, quite right. Yes, of course. Yeah, that's indeed a very important point. Given the time, should I stop? Including discussion. Or with? Then I stop because I prefer discussion. Thank you.