 Badi, je tudi pravda začniti, ne? Včetnje noga sečenja. To je tudi o jerarki. Všeč je Enrico Olsson z Vienna in Santa Fe. Kako? Tukaj sem sem. Vse je dobro. Stereo. Dobro, sreči, sreči. Odem, da se počekaj, da smo vse vzelo kot, da je vzelo, da se vzelo. And we can effortfully change our network connections, the way we integrate the information in order to solve the problems we are facing. So this is a complex puzzle, with lots of moving parts, lots of interactive parts, but today I'm going to focus on one specific part of this big, big, big puzzle. Od taj bi bilo, če mi je tukaj še začilo, da mi je z bole thicko. islands želje skróli, če izgleda obrvnih sítor. Prostim, kako je ovo, če je noga še izgleda, kako je zelo zelo vendiv. Svisno, če je do vseh značila, daAAk je, če je dne, kako je je zelo, kako je tačno zelo. Začala tako, če bo zelo, če je je zelo, če je poč鼻, about broader populations to the research on perceptions of inequality. So the last, let me say, 15 years or so, there's been a remarkable kind of uptick in research about perceptions of inequality. And actually this indicators of perceptions of inequality sometimes are better predictors of policy attitudes, public health metrics, and overall well-being than other objective measures. But in this literature, what stands out most when you read it, and you also see it in lots of reviews of this literature, is that there are lots and lots and lots of conflicting results there. Do people underestimate inequality? Lots and lots of studies showing that. Do they overestimate inequality? Lots and lots and lots of studies showing that also. So what is going on there? Like, what is this? So there is one example. So this is two studies published in 2014, 2016, by Seim Bezertal and Pejchen Goldstein. So here they use actually the same kind of the same measure. So they asked people about, so what is the percentage of people that have the annual income in the U.S., under 35,000, between 35,000, 75,000 and so on and so forth. And Pejchen Goldstein also asked this, but a little bit more sophisticated with more categories and so on. But what we see here now, to the left, the Chamber study, is that if you look at the lowest income category, there is an overestimation. And there is an underestimation of the highest income category. Ok, so what about Pejchen Goldstein? Well, surprisingly, I will show you conflicting results today. So we have the opposite area, underestimation of the lowest income category, and we have overestimation of the highest income category. So how can this be? There is pretty much a similar measure, similar things, but we still get very, very different results. We go through all and all and all of this literature of subjective perceptions using different measures. And I say different measures. And this is one of the things that also kind of stands out in this literature, is that there are so many different measures. And there has been, of course, in people discussing this, like, God, we have so many different measures. What should we do? So, for example, we have three different types of measures, which are used a lot in international service and so on. We have these distribution measures. We ask about distributions directly. We also have ratio measures. So you take questions such as, how much do you think the share amount of large natural cooperation earns? And then you also ask, how much do you think an unskilled worker in the factory earns, and you take the ratio of those two when they are correlated with different things? Another thing, like, ok, people do this research, oh, but people might not be so good in action numbers, so let's do pictures instead, distributional pictures. So here is more like an ordinal measure, where people have to match different shapes and distributions to what they think their country's distribution of income is. There's also a measure that's used a lot in lots of different international service. Ok, but there are also other problems here that have been pointed out in the literature. It's like, exactly what are we asking about here? So sometimes in, when you summarize this literature, you say that, oh, underestimation, overestimation, but of what? Is it annual income? Is it house of wealth? Is it before and after taxes? There's lots of different ways, different wealth measures that we used also. All these things that we'll come back to and discuss all these different measures and the actual measures for wealth. So, how can people do this? How can people make judgments of broader populations? Well, obvious first thing is, as we recall, what is in the distribution, you know it. There seems less likely, because most people don't know. They need to make some inference about this, what the distribution is and how much different people's incomes are. Other things that's also been used in literature, like, okay, let's not have a model, we have some uniform prior, maybe updated with something and some of that. That might also be something. Of course, all of this can be then kind of colored by our political preferences and other beliefs that kind of bias in judgments there and there and so forth. But what I want to focus on today is something that has been in focus actually in the last few years in literature on perception inequality and that is the characteristics of your immediate social environment. And it's been shown now in actually quite a lot of studies that people's perception of inequality and other populations actually is influenced by their immediate social environment. And that can be done in different ways. So, for example, in this article, they took the actual income in the zip code where their participants were located so how that related to their distributional preferences but also, in their case, their different preference for redistributional policies. But what we have done in our studies is that we use perceived wealth. So what we do is we ask people in your social circle, where the social circle is defined as people you are being in contact with, your friends, your family, and so on, but what percentage of people in your social circle are having an income that is less than 25,000 so we get the distributional judgment of that. And then we use that as a basis for our models of how you use that information to make judgments of broader populations. So the main idea here is then the population estimates are based on perceived characteristics from people's social circles. So what about these social circles? How do they look? Can people actually do these judgments? So how accurate are they? So here are examples of individual social circle estimates. So these on the left here are six different participants that are located on different places in the distribution themselves. They make these social circle estimates. And we see here now for three different distributions, also wealth, work stress, interest in a model that can explain distributional judgments of characteristics in for many different target kind of variables. And we see here that let's see if I can manage this. Oh, yeah, it works. Whoa, this is difficult. Okay. So for example here we have these participants, it's located at the lower end here, at one. At the top here. And you see that the social circle estimate is like, okay, it's like mostly in the bottom part here and it tapers off here. And then if you look down to this one, the better off, mostly of estimator social circle is mostly where that person is located in the distribution. Which is then, well, obviously we interact with people that are similar to us, we live in neighborhoods that have similar wealth but we also see that it's not always the case. I mean, for example, this participant is actually here up here at 6. But it seems that the participant in this particular study has actually social circle is located way lower than others. So we get this also variation of these social circle estimates. Okay, so they can actually make social circle estimates because one could be that they just report where they are and actually do this. Okay, are these estimates accurate? Well, we have some for some indirect evidence that these are actually accurate. So, this is results from a study in the Netherlands, a national representative survey where we ask people about house of wealth, number of friends, work stress and so on. Different types of distributions. And we see if you take the average of these social circle estimates, they are remarkably corresponding to the population census for these different variables. So that's at least an indirect evidence that there is something there in the social circle that's actually accurate. Okay, but now you're thinking but the population estimates are biased. Or can that be? But before I go into that, I can also say that we have used other indirect indicators that show that these social circle estimates are actually accurate. So we have used these social circle questions in predicting, forecasting elections. And what we do then is basically ask people, okay, in your social circle define this, people you interacted with in the last 30 days that can vote. What percentage do you think will vote for Trump? What percentage do you think will vote for Biden? And others. And we show here that this social circle question can outperform the traditional question of asking us who will you vote for in six different national elections across four different countries. And we see here in the picture here in the figure the absolute error for own intentions is the traditional question and social circle question that the average error, the error here is lower for the social circle across these three. And then we see that same pattern across all these different elections. Which is a further indication that social circle estimates actually capture something that's real in your social circle. So how can this be, why? Why is it the case that this outperform the own intention? Well, one is simply a kind of a wisdom of crowds effect. Because basically what we're doing here, you ask one person, but it gives estimate of what other people work. So basically you increase the sample size. If this is accurate representations of the social circle we vote, then this will actually kind of increase the sample size and also increase the accuracy. But maybe most more importantly is that this way of asking can reach populations that you don't reach by other surveys. So it could be that many people that they don't answer it. No matter what that person would have answered. So that's also a contributing factor to why this works. Ok, back to the question that I asked. So how can it be that the population estimates are still biased if they base their judgments on the social circles? Yes. Oh, that's just sorry, that's just what will you vote for? Do you mean elections? So the question is ok, so let's take the election. So here there are three types of questions. One question is what will you vote for in the election? That's the only question. The other question is the social circle question. What percentage in your social circle do you think will vote for Biden, Trump and so on and so forth? And then we actually have the pop so in this case you see here the election winner question here also is called election winner. Then we ask basically who will win this is a population question the election winner. Who do you think will win the election? Which is basically an estimator of the population, what the population will do. So these are the three questions we have here. Yes, of course. So this is a forecast of the election results. Like the standard forecast in election results so we compare that to the actual percentages that Trump, Biden and other got. Just by averaging the social circle estimate by taking the average of the older what you think that what would you will vote and also average of ok who will win the election. And all that we show this. But the idea here then when we go further here is that now we come to the second step is that we actually use this information in circles to make this population estimates. And that is the how can we go from how do people use this information in order to make population estimates. What? Oh sorry. Yes. So typically in this election studies you have several waves before the election. So wave one is further away than wave three. Wave three here is always just before the election while this other waves are several weeks or several months before the election. Yes. Ok, so more specifically we usually use a definition I mean this is also for the debate exactly how we can do this. But we say that we tell them that you are now going to estimate the percentages of different things in your social circle. Or your social, we call them social contacts because it's more easily for people to understand that. And we define that those that are your friends, family or acquaintances that you have interacted with frequently in the last 30 days. That's a standard thing that you use. And then in addition in this election studies we also need to have some things that need to be over 18 down to that. So that's our usual definition that we use throughout our different studies. If you suppose there is a social network then this is not possible. It's a bias example. Yes, it could be but then if you have a national representative sample then you are sampled in a representative way and then you can give information about others that are like you really like you. So it becomes maybe more like a two-step not snowball sample but almost like that. Yeah. Yes, true. That's all true. That's all true. That's true. Yes, that's true but it also brings this opportunity that you can get people that don't answer the survey. If I ask you and some of your friends they never answer the survey. But you can estimate what that person will say then we get information about that person when we ask you. So that increases the information value that you can use to predict. Okay. Any more questions? No, actually this is just pretty, I mean what you said, what will you vote for? Trump. What will you vote for? Biden. And then there's a standard question in the survey for forecasting. That's all we're doing. We're just asking standard questions about people. If it was election today, what will you vote for? Or actually what is the percentage chance that you will vote for Biden? And then you ask you what is the percentage or your social circle that will vote for Biden? The whole population in the U.S. will vote for Biden. These are the three questions. Yes. So these are three different questions. And I mean this election winner question has been used a lot actually. I mean who will win the election? That's been used in the U.S. in the early 2030s, but not so much in the polling industry. They're still using the question who will you vote for? It was the election today. What is the outcome of the election? What is the prediction if I take the average? The average. We just have to do exactly the same as any server polling thing. This is just a side point. I just wanted to show you this. OK. Is this something of us? No, so actually this is so the error lower is better. This average, absolute error. But I know you're on top. So this is actually quite a typical. These two things, the 2010 and 2020. If you look back throughout the polling history and if you compare this election winner question the error is usually lower than on attention. It's usually better than asking people about yourself. Not in these cases, but overall it's actually the case that asking about what people who think will win the election is actually usually better than just to ask everybody who they will vote for. OK, let's move on. I just wanted to show this. We can actually use the information in this situation. OK. Back to the biased distributional assessments. So we see here this. There is some bias when you ask people about characteristics in overall broader populations. So in order to understand why is this the case we need to go into exactly how people are doing these estimations. So the first thing we can notice population estimates which is on the bottom part of this figure estimated population distributions for health and wealth, number of friends and work stress. And we have separately plotted here these estimations or perceptions for worse off and better off people. And the same for estimated social circle distributions for worse off and better off. First we can notice, oh look the population estimates fall and we also see that they are a little bit more flattened. See there is more peaky the social circle distributions than the population distributions. So here is the the key to that is that that can have many different causes like statistical causes that error in the way you translate from your social circle distribution to population distributions that error is going to flatten things. What we think is happening here in our model is that people know, I mean I know in my social circle I know that my social circle is not representative of the population. So I try to kind of smooth that out a little bit because I know that this is not the case and one way of doing that is that you retrieve instances of people from your social circle but you try not to retrieve those are more similar to you but those are more dissimilar to you which also then lead to this flattening distribution. So that is mechanism that we implement in a formal model of population estimates. Ok. Then we can implement a simple statistical version of this. Extremely simple, as we call SSM there is basically a model that adjusts towards the average of all social circle estimates and makes it more smooth. As compared to what we call a regression account where simply that let's take the actual population distribution and at the same process of kind of smoothening it out, basically a regression account and we see here that our extremely simple account can then account for both worse of and better of people. That's just an example. Ok. So now ok, so now we are in the position to try to explain these conflicting results that we saw in the beginning. Ok. So we saw here that there is a difference in overestimation and underestimation of the lower and higher income category between these two. We see that there is basically the same kind of measure, distributional measure but to note here is that to the right, the page in Goldstein study, they used a national representative survey. It's actually what we will expect then due to this regression account and others that we wouldn't underestimate the lowest income category and overestimate the highest income category. But in the chambers at all they used a convenience sample for American mechanical Turk which then is not national representative. So what you can do here now, ok, what type of participants do we have here actually. So here we see the income distribution because they reported this in the supplemental materials. And we see that the participants incomes is more severely skewed than the actual population income. And then the simple idea that, ok, if people based the judgment of populations from their social circle we would find this that they owe. They will give a lot of they will give a lot, a lot more in the under 35,000 and they will give less in the 75,000 over. And they will give the smooth a little bit because this is quite strong effects in terms of their actual distribution. So even the smooth a little bit we will still expect that this part can hold. So it's important to not just take the studies that they are we need to know where do the participant comes from how do they solve the task. So there is at least one contributing factor to this conflicting results for this distribution of measures. Ok, that was distribution of measures. What about this ratio measures? That's been used a lot. And and the more, the consistently when you use this measure it's consistent to find that people are underestimated in the quality. So for example in 2012 the estimated ratio was 30 to 1 there was also a national representative survey when it was actual 154. And but then, what is this question? How do people solve this question when they see this question? So when I see this question, ok, how am I supposed to solve this question? Is it a typical Sherman? Is it any Sherman? Someone I know is it such a mean or median is unclear exactly how what should you base this on. And we don't know actually how people solve this and we need more research and exactly what's happening when people are getting this question. One little indication that it might not be so much about social circle in this question is that if you look at for example this study site there from 2014 it's actually similar ratios across socioeconomic status which might suggest that this is actually less influence on social circles for this particular question. Because people from different parts have similar ratios. So that's a little bit unclear of this ratio measure. We'll come back to ratio measures a little bit later. Ok. Then we have this ordinal measures. So this shows results from Nihouse 2014. So Nihouse did this aggregate measure. Because we can think of this is now from the US with different percentages. That also different things. But probably those that choose type A there is a high concentration in the lower part they are probably from the lower part of the distribution the income while those in type A probably based on what we discussed so far are probably from the higher part. But if we combine all this somehow then we should actually get a pretty good picture of the distribution correspondence with the actual distribution and the estimated distribution based on this measure. And that was also what Nihouse did there. So she combined all this and what we see here if we combine all this we actually see that we see a little bit of underestimation of those categories. Pretty good. Ok. That's all good and fine. But this is just one example. If we then look at the different countries of course we still see difference between countries as we pointed out again and again. But again what explains these differences? I mean, across countries how do they interpret these different categories? Do they do that the same way across countries across cultures, across different norms for different things? Still unclear if we can actually compare this across countries and we don't know how people different people and individual differences and culture differences how you interpret these different pictures. That's where we stand on this measure. And of course different samples of participants. Ok. Yes. True. We can still have differences. I mean, what do you perceptions of what do you believe is this between zero and $20,000 or is it between zero and $50,000 or is it between zero and $200? Exactly. We don't know the difference if I can take a look at the other slide. Yeah, and it's also a very crude measure. And then people so, I mean, they use this measure for different things also. They use it for ideal type of societies. They show that this measure actually that correlates pretty well with redistributional preferences how you want an ideal society to look like for example. Yeah, I'm still skeptical about this measure because of, I mean, we don't know how people actually interpret that and what they're doing there. Ok. And of course, then we have this question about different wealth measures. Sometimes this can be very difficult for participants or people in general to understand. You see sometimes in instructions to people explain different wealth measures like aggregated net worth across households. People know that. So there can be one problem. Another problem is of course if we now rely on our media social environment to make population distributional judgments of course we have different visibility in different parts of society. We see different wealth in different parts. We can have different social norms about sharing information about the income of wealth. We're also of course going to influence how we make judgments about inequality. Ok. Let me, what is the time actually? Ok. So, obvious conclusion is that results of preceding equality depend on the way it's measured. That's not necessarily bad because it depends on what we are interested in also. Are we interested in in the upper part and lower part of the distribution? Maybe the ratio measure could be good there. Are we interested in the whole distribution? Ask about the whole distribution. But don't expect them to coincide with the type of process people are using to come to these estimates and the actual shape of distributions. And also I hope you are convinced that estimates of inequality or in general estimates of population characteristics depend on immediate social environment. And I think it's important to before we start relating measures of inequality, we like redistribute the preferences. I think we need to kind of understand how people actually are doing these judgments. Like, what type of representations do they have of their immediate social environment? Or, what are the processes that works on these representations? And finally then, how do these processes then interact with the structure of the immediate and broader environment? And then of course on top of that we have different ways of asking which also interact with all these three other things. I mean, there's lots of open questions there, but let me just take a few things. So, when we ask about inequality, what are we exactly asking about? Are we asking about inequality, or are we asking more about political preferences? So, these ratio questions, for example, when I see these questions, I immediately start thinking about redistributive policies and so on. Like, oh, yeah. But if I ask at the distribution, I might not think about that so much. Percentage here, percentage there. Or maybe, this is just me speculating, maybe it's not surprising, it's actually shown that ratio questions are pretty good to pretty good predictors, or they actually at least correlate with political preference like redistributive preferences. And they seem to correlate more than distribution questions. Although distribution questions has not been used so much. But still, there's something to think about. And of course also, what part of distribution are we interested in here? I mean, different questions target different parts of the distribution. So, we're interested in the low part, high part, both of the low part, the difference between the low and the high part, the whole distribution of wealth, different groups of gender race, et cetera. Different concentrations of wealth in different parts there might trigger different beliefs about inequality. And then, of course, we have a bigger question. What are we asking about? Are we asking about the interest in the local inequalities? In the national, between groups in the country, are we interested in the global inequality? Because all of this is different and might trigger different processes, there are different distributions involved and so on, so forth. And on top of that we have intergenerational inequality and all of that. And that was what happened. Thank you. Thank you for very inspiring talk. Thank you. Thank you. So, you talked a lot about how people estimate income and so on. Have you also studied how people perceive the fairness like what they think, what distribution they would perceive as being fair or unequal? No, we have not. There is a lot of studies on that, actually. So, both in terms of, especially when it comes to this pictorial studies, so you ask both to match it to your nation or society or whatever you ask about. But also, how do you think the society should be? And I don't know that literature 100%, so I just know that there is a lot of studies about that. But we have not done that. So, I was just wondering, I mean, are there also studies that shows that how people perceive the dangers of climate change compared to what actually the specialists know, like how people are perceiving how bad things really are. And because I was also feeling that it might also depend which part of the planet it might be. Yes, definitely. So, I have not done that, but I know there are studies, but if you just extrapolate a little bit here. If it is important to meet a social environment, it seems reasonable to suggest that people, I don't know, in different parts of the world, in different weather, I mean usually the weather every day or you hear about different catastrophes and so on. So, this is my question. We need to address this while I am living in the northern part of Sweden and I am also concerned. But on average, how does it happen? On average, do people really perceive the dangers well or are they way of the mark? I think that I don't know for climate and estimates for the... I mean, there is also a lot of uncertainties there. I mean, there are no risks, but exactly if you take a model outcome of the predictions of what's going to happen with the average temperature in 200, 300 years, there's also uncertainty of that. How do you compare it? To what are you going to compare the risks with, for example? But I don't know that literature, but I know there are studies that compare that, but I cannot also... Maybe I can just talk later So I have a general question. My impression is that most of what you are talking is about absolute perception or absolute wealth. Yes. Whereas my are at somewhere that we are better at perceiving relative wealth. So I can know whether say you earn, say, 20% more or 20% less so your wealth with respect to mine and somehow looks like this idea of social circles maybe goes in this direction in the sense that we have a better perception of how we stand in our... On the other hand, I mean... So my question is would the questioners improve if the questioner asked in terms of relative wealth rather than absolute wealth? So that's also a complicated question. So one thing is about in general the relative thing. I mean the ratio questions are a little bit like that. Like you compare that. But they have other problems then because it seems that they are way off in terms of the relative difference between the high earners and the low earners. And then we come to this question about the relative thing. The thing with the relative standing is going to involve so many other processes also. I am here in this position and compare yourself to this lawyer over there which might in effect trigger other process, more bias process. But as I said earlier we will ask about distributions. And we show here that distributions are actually quite accurate when you do the absolute percentages in different categories. But when you start to include comparison processes that might lead to different things. Having said that, we also have in our model we also have implemented this kind of relative and distribution judgments. And we can actually explain different we can say this we can explain different effects of asking relative or absolute judgments just because of the interaction between our processes and the distribution shapes of social circles and the actual distribution shape. But there is a bit too much to go into here. But yes, I can say that in our experience asking for absolute judgments or percentages in the distribution they are actually pretty good. Although because of that they use their social circle estimates you are going to see them a little bit flattened a little bit of a regression effect on it. But actually if you plotted it the bias, when you read the literature on inequality and any distribution you always see there is all the way off. There are no correlations with objective reality whatsoever that is not true. There is a correlation between them and if you ask them for the distribution shape they are pretty good. So when you say psychology what do you mean? No, but that's an interesting question. I mean, that's something that definitely worth pursuing because most in the social psychology literature about judgments of others is individual persons. And you say like what is the personality characteristics of FRIBA? Is it high on this? You're outgoing or a little bit neurotic maybe? And then you compare and then you ask the same questions and then we see how this corresponds. So there's been lots of work on that. We form impressions in the first 30 seconds and all this kind of stuff. But then as you said asking them about the general population in a similar sense that would be about the psychological processes more. So there's an interesting question. I mean, there's lots of maybe I'll try to hint at different things. So if I get this question what would I do? I mean, first obvious thing is oh, so there are many sources of information that you use to answer that question. Different process of that. One is obviously oh, yesterday I read about this CEO on $30,000 only okay. Then I base my judgment on that basically. Or I can't retrieve things from a memory about oh, I know this CEO or I don't know a CEO so I have to guess. So before I know that we need to know what do people know? What are the retriever processes or the inferential processes? So I think making a blanket statement of people do that is very much going to be context dependent in social environment. The influence that other sources of information like social media or regular media has. Unfortunately that is the only thing I can say at this point. There is a quick one I guess. So we talk about perception or misperception. So I was just wondering if you have any sides about changes in perception. Because I think sustainability problems so perhaps we understand how people change their perceptions. So in that sense we don't have the for this regular distribution. We have a little bit of that when we have the forecasting thing. Then we can see how do your own beliefs of what you will vote for and how that relates to your social circle. We can see that over time how that changes. We can then see okay why political events? How does it change the beliefs over time? We can see that in time series. We can also have ideas about how you influence by the social environment. Maybe the social environment is a better predictor of what you will do in the future. For example we can see some evidence that we ask people about the social circle how they will vote and how yourself will vote. We can see that in some circumstances when we say that your social circle will vote it's a better predictor of what you will vote for what you actually voted for. We seem to suggest then that there's this social influence that okay I'm not so sure about this maybe I go by them and then everybody around say like Trump Trump Trump Trump Trump okay I will vote for Trump. So the social circle can be used as a crystal ball in some circumstances to understand changes in your beliefs in your behavior.