1 00:00:00,000 --> 00:00:04,330 I think in European politics there's also a fair amount of work tracing out this idea that 2 00:00:04,330 --> 00:00:07,590 people tend to get reinforcing information. 3 00:00:17,630 --> 00:00:24,530 Hello and welcome everybody to Game Changer, a podcast by TWS Partners. My name's Florian. 4 00:00:24,770 --> 00:00:25,610 This is Simon. 5 00:00:25,830 --> 00:00:26,990 And this is Miriam. 6 00:00:26,990 --> 00:00:33,110 And in this podcast, we explore applications of game theory in business and everyday life. 7 00:00:38,400 --> 00:00:41,000 Hello and welcome to today's episode. 8 00:00:41,460 --> 00:00:48,380 Maybe if you had this experience before, you're scrolling through social media or reading the news 9 00:00:48,380 --> 00:00:55,140 and at some point you start to wonder, am I really getting the full picture or am I just 10 00:00:55,140 --> 00:00:59,480 seeing the same perspective, same information over and over again? 11 00:00:59,820 --> 00:01:05,040 So our guest today, Adam Meirowitz, professor of political science with a courtesy appointment 12 00:01:05,040 --> 00:01:10,200 in economics at Yale University, will shed some light on this from a political perspective. 13 00:01:10,580 --> 00:01:11,780 So please enjoy the episode. 14 00:01:13,380 --> 00:01:15,420 Adam, welcome to the podcast. 15 00:01:15,720 --> 00:01:16,380 Thank you for having me. 16 00:01:16,520 --> 00:01:20,840 I'm excited that there's a place where people that think about strategy and incentives and 17 00:01:21,400 --> 00:01:23,940 aren't boring academics, engage in this stuff. 18 00:01:24,100 --> 00:01:25,520 So hopefully this is fun for people. 19 00:01:25,980 --> 00:01:26,600 Yeah, absolutely. 20 00:01:26,820 --> 00:01:27,560 I'm sure of it. 21 00:01:27,880 --> 00:01:33,120 Before we dive into everything, because we are going to talk a lot about elections and 22 00:01:33,120 --> 00:01:34,900 voters and all of these things today. 23 00:01:35,640 --> 00:01:40,600 And since your focus of research, or at least what we will be discussing today, is the US, 24 00:01:41,060 --> 00:01:47,460 I think it makes sense for our listeners based in Europe and also myself, if you could briefly 25 00:01:47,460 --> 00:01:53,120 give us the core features or the core aspects of the US election system, if you remind us 26 00:01:53,120 --> 00:01:57,620 the two-party system, how do the candidates compete and all of that, so we have the right 27 00:01:57,620 --> 00:01:58,000 setting. 28 00:01:58,420 --> 00:01:58,620 Definitely. 29 00:01:58,860 --> 00:02:01,720 And this is probably even a caricature of American elections. 30 00:02:01,840 --> 00:02:05,260 And maybe it seems very abstract for Europeans where things seem more complicated. 31 00:02:05,800 --> 00:02:10,480 But for us, the idea is that elections are largely about two parties competing for office, 32 00:02:10,600 --> 00:02:15,460 that the winner is chosen by majority rule, sort of winner take all is sometimes our term. 33 00:02:15,460 --> 00:02:20,660 And it's also a tradition where we think of candidates for the most part stake out some 34 00:02:20,660 --> 00:02:24,480 notion of policy that they would implement or at least try to implement if they're in 35 00:02:24,480 --> 00:02:24,860 office. 36 00:02:25,100 --> 00:02:29,800 And voters are evaluating these platforms almost as commitments and then making a choice. 37 00:02:30,180 --> 00:02:34,980 And it's largely a tradition where we think of things on a one-dimensional spatial spectrum. 38 00:02:35,320 --> 00:02:39,020 So thinking about there are platforms that could be more leftist or more rightist. 39 00:02:39,220 --> 00:02:42,060 In Europe, the interpretation may be a little different, maybe not that different. 40 00:02:42,060 --> 00:02:46,700 And then there's sort of another feature, which is not always involved when game theorists 41 00:02:46,700 --> 00:02:51,800 do American politics, but it's becoming increasingly important, which is in addition to this spatial 42 00:02:51,800 --> 00:02:53,200 ideological aspect. 43 00:02:53,620 --> 00:02:55,400 We also think there's a vertical dimension. 44 00:02:55,680 --> 00:02:59,700 So candidates could generally be better and more or less everyone agrees about people 45 00:02:59,700 --> 00:03:05,580 that are less corrupt, more organised, more reputable, have money that they can use in 46 00:03:05,580 --> 00:03:07,060 productive ways for campaigning. 47 00:03:07,060 --> 00:03:12,720 And so we really think about two-party competition when platforms are commitments on a spatial 48 00:03:12,720 --> 00:03:18,500 sense and voters look at both this horizontal spatial sense and also some vertical components. 49 00:03:18,980 --> 00:03:20,100 Okay, that's a great introduction. 50 00:03:20,940 --> 00:03:25,480 And to continue this a little bit from the game theoretic side, or at least the economic 51 00:03:25,480 --> 00:03:31,200 side, guessing some of our voters remember from their studies, maybe the median voter. 52 00:03:31,740 --> 00:03:35,560 So there's a classical models that go in that direction. 53 00:03:35,560 --> 00:03:41,080 And in these models, candidates are often predicted to converge to the centre in the 54 00:03:41,080 --> 00:03:41,940 political sense. 55 00:03:42,600 --> 00:03:45,100 Where does this come from and why? 56 00:03:45,500 --> 00:03:49,160 Is it a benchmark for your research or is it a benchmark for your research at all? 57 00:03:49,840 --> 00:03:51,300 Yeah, no, definitely a research. 58 00:03:51,440 --> 00:03:56,420 And if I'm honest, in political economy, this is a tradition that probably dates back to 59 00:03:56,420 --> 00:04:02,000 like the 1920s when industrial organisation economists were thinking about, here's a 60 00:04:02,000 --> 00:04:02,600 simple story. 61 00:04:02,600 --> 00:04:07,180 You're trying to locate on a boardwalk or on a single street and where the firms that 62 00:04:07,180 --> 00:04:09,440 are trying to maximise their customers locate. 63 00:04:10,020 --> 00:04:11,180 And there's a centrist force. 64 00:04:11,280 --> 00:04:13,940 You sort of locate in the middle of the boardwalk or in the middle of town. 65 00:04:14,060 --> 00:04:19,959 In the 50s, this intuition jumped into political science where Anthony Downs sort of talked 66 00:04:19,959 --> 00:04:22,880 about competition on an ideological spectrum. 67 00:04:23,380 --> 00:04:28,220 And there's this really strong force for locating at the preferences of the median voter. 68 00:04:28,520 --> 00:04:30,900 And so our idea is that we have a bunch of voters. 69 00:04:30,900 --> 00:04:33,200 They each have some spatial ideal point. 70 00:04:33,340 --> 00:04:36,320 They look for candidates that are closest to their ideal point. 71 00:04:36,800 --> 00:04:41,300 And the idea is that if one of the candidates is located at the point where the median voter 72 00:04:41,300 --> 00:04:45,580 is located, a candidate that's to the right of her is going to get less than half of the 73 00:04:45,580 --> 00:04:46,620 people voting for her. 74 00:04:46,760 --> 00:04:48,840 And the candidate at the median gets more than half. 75 00:04:49,180 --> 00:04:52,700 And similarly, if a candidate's to the left of this median, she'll get less than half. 76 00:04:53,220 --> 00:04:57,980 And so when voters are looking at candidates and the candidates care about winning, we have 77 00:04:57,980 --> 00:04:59,220 this incredibly strong result. 78 00:04:59,220 --> 00:05:00,520 You locate at the median voter. 79 00:05:00,960 --> 00:05:05,540 This logic then jumped into work where we thought about parties actually having policy 80 00:05:05,540 --> 00:05:05,980 motivation. 81 00:05:06,200 --> 00:05:09,980 So it wasn't about winning, but it was about winning with policies that they liked. 82 00:05:10,540 --> 00:05:14,740 And so the idea is that if you thought about one party that liked right-wing policies, another 83 00:05:14,740 --> 00:05:18,980 party that liked left-wing policies, they still in equilibrium would have to locate 84 00:05:18,980 --> 00:05:20,360 at the median voter's ideal point. 85 00:05:20,560 --> 00:05:24,340 If you're at the middle and I move to the right, it's great for me if I win, but I'm 86 00:05:24,340 --> 00:05:25,700 going to get less than half the voters. 87 00:05:25,860 --> 00:05:26,540 And so I lose. 88 00:05:26,540 --> 00:05:30,880 And so the only story that makes sense as an equilibrium is locating at the middle. 89 00:05:31,380 --> 00:05:34,320 And it wasn't until we thought about introducing some uncertainty. 90 00:05:34,660 --> 00:05:39,140 And so if you think about parties that don't know exactly where the median voter lies, 91 00:05:39,340 --> 00:05:42,180 maybe there's uncertainty about who's going to show up on election day. 92 00:05:42,580 --> 00:05:47,820 Maybe there's uncertainty about some national tide where all the voters decide they're a 93 00:05:47,820 --> 00:05:50,000 little bit more leftist than we thought they were. 94 00:05:50,440 --> 00:05:55,120 So then when you have uncertainty about where this median voter is, candidates that care 95 00:05:55,120 --> 00:05:57,420 about policy face a tradeoff. 96 00:05:57,820 --> 00:06:02,440 If they locate at where they think it's most likely that the median voter lies, they're 97 00:06:02,440 --> 00:06:03,520 more likely to win. 98 00:06:04,100 --> 00:06:09,200 If they locate at a platform closer to their end of the policy spectrum, so rightist parties 99 00:06:09,200 --> 00:06:13,740 locate to the right, they're happier if they win because when elected, they enact a policy 100 00:06:13,740 --> 00:06:15,600 that biases policy in their favour. 101 00:06:16,100 --> 00:06:20,560 So we got in the 80s, Calvert and Whitman is sort of the famous papers, this idea that 102 00:06:20,560 --> 00:06:21,700 there's a stark tradeoff. 103 00:06:21,700 --> 00:06:26,980 If I moderate my platform towards the place where it's likely that the median is, I win 104 00:06:26,980 --> 00:06:29,460 with higher probability, but I'm less happy when I win. 105 00:06:29,820 --> 00:06:34,020 If I choose a more extreme platform, I win with lower probability, but I'm happier when 106 00:06:34,020 --> 00:06:34,360 I win. 107 00:06:34,880 --> 00:06:37,920 And so equilibrium has both parties balancing this out. 108 00:06:38,400 --> 00:06:41,940 And what we get is finally a prediction for platforms that are not the same. 109 00:06:42,180 --> 00:06:44,860 Rightists compromise a little bit, but they're still right of centre. 110 00:06:45,400 --> 00:06:47,080 Leftists are left of centre. 111 00:06:47,080 --> 00:06:53,180 And then from the 80s till now, we've had lots of cool papers that add neat moving parts 112 00:06:53,180 --> 00:06:56,020 and abstractions and generalisations to this core logic. 113 00:06:56,320 --> 00:07:00,860 Yeah, very interesting that introducing the uncertainty will lead then to different equilibria 114 00:07:00,860 --> 00:07:03,460 and not the typical split the voters, so to say. 115 00:07:04,320 --> 00:07:04,640 Great. 116 00:07:04,740 --> 00:07:10,820 In your most recent paper, you also think about selective exposure in the context of 117 00:07:10,820 --> 00:07:12,160 this electoral competition. 118 00:07:12,160 --> 00:07:17,400 And you introduced something that is called echo chambers, or at least they play a role 119 00:07:17,400 --> 00:07:18,140 there in your model. 120 00:07:18,340 --> 00:07:21,620 Can you tell us what echo chambers are in this context? 121 00:07:22,140 --> 00:07:22,420 Yeah, yeah. 122 00:07:22,500 --> 00:07:27,840 For us, this was sort of an attempt to introduce to this research tradition, these models dating 123 00:07:27,840 --> 00:07:33,460 from Hotelling in the 20s to Calvert-Whitman in the 80s, the starkest description of what 124 00:07:33,460 --> 00:07:35,660 we think is going on with the media now. 125 00:07:35,920 --> 00:07:40,520 So we used to think that voters all roughly got the same information about elections. 126 00:07:40,520 --> 00:07:43,740 In the US, there used to be a world where there were three networks that would tune 127 00:07:43,740 --> 00:07:44,840 in and tell you about elections. 128 00:07:45,420 --> 00:07:50,480 But if you think about going from a small number of media outlets to cable, to the 129 00:07:50,480 --> 00:07:55,560 internet, to every other form of social media, the idea is now that we think for the most 130 00:07:55,560 --> 00:07:58,520 part, individuals have these curated media feeds. 131 00:07:58,720 --> 00:08:02,480 Like they get information that makes them happy, whatever it is that drives them to 132 00:08:02,480 --> 00:08:05,720 demand certain kinds of facts or certain kinds of discussions. 133 00:08:06,220 --> 00:08:10,320 And so we have this sense, definitely in American politics, there's a lot of empirical 134 00:08:10,320 --> 00:08:10,640 work. 135 00:08:10,720 --> 00:08:14,660 I think in European politics, there's also a fair amount of work tracing out this idea 136 00:08:14,660 --> 00:08:18,300 that people tend to get reinforcing information. 137 00:08:18,540 --> 00:08:23,240 So if I'm right of centre, I'm super happy to hear other people that are right of centre 138 00:08:23,240 --> 00:08:27,280 tell me how wonderful candidates that are right of centre are, tell me that right of 139 00:08:27,280 --> 00:08:31,200 centre candidates are going to win, tell me that left of centre candidates are incredibly 140 00:08:31,200 --> 00:08:33,740 corrupt and no one likes them and they're not going to win. 141 00:08:34,120 --> 00:08:37,799 And so there's this logic that maybe this is sort of this vicious cycle where everyone 142 00:08:37,799 --> 00:08:42,620 to the right gets very homogeneous, but biassed or skewed type of information. 143 00:08:42,880 --> 00:08:46,260 Everyone to the left gets some different biassed or skewed information. 144 00:08:46,760 --> 00:08:51,360 Of course, maybe things are more nuanced and we partition this ideological space into 145 00:08:51,360 --> 00:08:52,360 different categories. 146 00:08:52,800 --> 00:08:57,020 But we're trying to think about how this idea of selective exposure or echo chambers 147 00:08:57,020 --> 00:08:59,460 might impact the strategic incentives for candidates. 148 00:08:59,700 --> 00:09:01,020 Do they moderate more or less? 149 00:09:01,240 --> 00:09:06,540 So basically echo chamber means that my beliefs as a voter are reinforced because I only get 150 00:09:06,540 --> 00:09:08,760 selective information about the candidates. 151 00:09:09,000 --> 00:09:09,500 Yeah, exactly. 152 00:09:09,720 --> 00:09:13,160 And we're sort of agnostic on, I think, two different directions or mechanisms. 153 00:09:13,740 --> 00:09:18,300 It could be that the sources that are providing information and selling stuff on the internet 154 00:09:18,300 --> 00:09:20,300 are getting increasingly good at targeting people. 155 00:09:20,440 --> 00:09:22,800 So they sort of know what makes me happy and what I tune into. 156 00:09:23,340 --> 00:09:26,900 And so they send information that's biassed in favour of what I want to hear. 157 00:09:27,280 --> 00:09:29,840 Or it could be that it's just who do I choose to listen to? 158 00:09:30,040 --> 00:09:32,220 Which news services do I choose to go on to? 159 00:09:32,580 --> 00:09:34,640 Who are the people that I communicate with? 160 00:09:34,640 --> 00:09:37,640 So it could be just which friends do I talk about politics with? 161 00:09:37,760 --> 00:09:39,720 Or it could be, you know, what's going on on my phone. 162 00:09:39,960 --> 00:09:40,600 Probably both. 163 00:09:41,280 --> 00:09:41,460 Yeah. 164 00:09:43,200 --> 00:09:47,560 So digging a little bit deeper on these echo chambers, they are often identified, 165 00:09:47,700 --> 00:09:52,260 and I think that makes sense, as a reason for this increased polarisation, right? 166 00:09:52,280 --> 00:09:55,300 If you always get the same good news about your preferred candidate. 167 00:09:55,760 --> 00:10:00,680 But how do these echo chambers change the strategic incentives of the candidates? 168 00:10:00,840 --> 00:10:02,200 You already mentioned that. 169 00:10:02,200 --> 00:10:05,580 When they choose their platforms, are there any trade-offs they're facing? 170 00:10:05,740 --> 00:10:06,900 How does it play out? 171 00:10:07,060 --> 00:10:09,040 Yeah, that's sort of at the heart of what we wanted to do. 172 00:10:09,160 --> 00:10:10,480 So we had in mind this idea. 173 00:10:10,900 --> 00:10:15,640 And I should say, this is largely a paper with Avi Acharya, Peter Boussire, and Floyd Zhang, 174 00:10:15,820 --> 00:10:17,040 various sort of co-authors. 175 00:10:17,620 --> 00:10:21,600 And so we wanted to think about what would happen in this classic model of elections 176 00:10:21,600 --> 00:10:23,140 when there aren't echo chambers. 177 00:10:23,320 --> 00:10:26,860 And then understand what would happen in equilibrium when there are echo chambers 178 00:10:26,860 --> 00:10:27,900 and make this comparison. 179 00:10:27,900 --> 00:10:30,580 And so the first thing we confronted is, 180 00:10:31,040 --> 00:10:35,540 well, if we're going to tell a story about how echo chambers affect candidate positioning, 181 00:10:36,000 --> 00:10:41,200 we have to think about the channel by which echo chambers influence voters in the model. 182 00:10:41,600 --> 00:10:44,640 And so this required that there's something for voters to learn 183 00:10:44,640 --> 00:10:47,740 that potentially gets influenced by the news that they tune into 184 00:10:47,740 --> 00:10:49,060 or the news that they listen to. 185 00:10:49,440 --> 00:10:54,340 And what I traced out before is how a candidate would face this trade-off 186 00:10:54,340 --> 00:10:57,320 between uncertainty about how voters are going to vote, 187 00:10:57,420 --> 00:10:59,560 where the median voter lies, and winning. 188 00:11:00,100 --> 00:11:02,260 And so we had to take a step back and think about 189 00:11:02,260 --> 00:11:06,620 how the presence of echo chambers influences voting behaviour 190 00:11:06,620 --> 00:11:08,840 or voter processing of information. 191 00:11:09,180 --> 00:11:12,160 And then how that would shape how candidates think about 192 00:11:12,160 --> 00:11:15,320 the probability that the median voter is in a particular place. 193 00:11:15,480 --> 00:11:17,480 So the way we go about this is thinking that, 194 00:11:17,540 --> 00:11:21,380 you know, an echo chamber could largely be a way that voters learn about 195 00:11:21,380 --> 00:11:25,940 the policies that candidates are going to enact, the horizontal spatial dimension. 196 00:11:26,460 --> 00:11:29,980 Or an echo chamber could be a way that voters are largely learning about 197 00:11:29,980 --> 00:11:33,060 these vertical dimensions, you know, which candidates are more liked, 198 00:11:33,200 --> 00:11:36,700 which ones are more competent, which ones are more likely to win, and so forth. 199 00:11:37,080 --> 00:11:38,180 And there's a little introspection. 200 00:11:38,440 --> 00:11:41,640 I think about everyone except the nerdiest political scientists. 201 00:11:41,840 --> 00:11:45,720 And I don't think we really have a media feed that's giving us detailed documentation 202 00:11:45,720 --> 00:11:48,280 of exactly what tax rate is going to be enacted 203 00:11:48,280 --> 00:11:51,780 and how, you know, various policy interventions are going to go about. 204 00:11:51,900 --> 00:11:55,640 We all get the sort of looser, more about the process 205 00:11:55,640 --> 00:11:58,900 and about what people think version of politics in our flow. 206 00:11:59,440 --> 00:12:03,660 So that made us think that the first order thing that's going on with these echo chambers is, 207 00:12:03,920 --> 00:12:05,660 it's about this vertical dimension. 208 00:12:06,080 --> 00:12:10,100 And so the starkest way that we could formalise this idea of, 209 00:12:10,280 --> 00:12:12,900 if you're right of centre, you get good news about the right candidate 210 00:12:12,900 --> 00:12:15,180 and bad news about the left candidate and vice versa, 211 00:12:15,800 --> 00:12:17,520 is to think that there's a media outlet. 212 00:12:17,520 --> 00:12:22,580 The media outlet learns true facts about which candidates are better or worse. 213 00:12:22,720 --> 00:12:23,840 So these vertical dimensions. 214 00:12:24,480 --> 00:12:27,040 And if I'm a right-wing person, a right-wing voter, 215 00:12:27,180 --> 00:12:29,140 and I'm choosing into a right-wing outlet, 216 00:12:29,320 --> 00:12:31,980 if there's in fact some good news about the right-wing candidate, 217 00:12:32,320 --> 00:12:33,500 it's going to show up in my feed. 218 00:12:33,780 --> 00:12:36,040 And if there's some bad news about the left candidate, 219 00:12:36,240 --> 00:12:37,400 it's going to show up in my feed. 220 00:12:37,700 --> 00:12:40,920 But if I'm a right-wing person tuning into news that makes me happy 221 00:12:40,920 --> 00:12:43,300 and there's bad news about the right-wing candidates, 222 00:12:43,580 --> 00:12:45,120 those stories just may not reach me. 223 00:12:45,120 --> 00:12:48,240 And if there's good news about the other party, the opposition party, 224 00:12:48,580 --> 00:12:50,020 those stories might not reach me. 225 00:12:50,400 --> 00:12:54,680 And so this is the way that the echo chamber is sending biassed information 226 00:12:54,680 --> 00:12:58,700 to voters based on whether or not they tend to be right or left of centre. 227 00:12:59,100 --> 00:13:04,100 And so this shapes how much the voters learn about the vertical dimension. 228 00:13:04,780 --> 00:13:08,580 And so now, given that, if we take a step back and think about 229 00:13:08,580 --> 00:13:12,840 the candidates balancing this probability of winning if they moderate 230 00:13:12,840 --> 00:13:16,340 versus how happy they are if they win by not moderating, 231 00:13:16,700 --> 00:13:19,780 what we have going on is in the world with an echo chamber, 232 00:13:20,300 --> 00:13:23,200 it seems like the voters are going to be learning less 233 00:13:23,200 --> 00:13:24,580 about these vertical dimensions 234 00:13:24,580 --> 00:13:28,300 because they don't get the bad news and they only... 235 00:13:28,300 --> 00:13:30,600 Okay, so if you're a candidate, 236 00:13:30,840 --> 00:13:34,120 if voters can only evaluate candidates based on two things, 237 00:13:34,160 --> 00:13:35,260 this vertical dimension, 238 00:13:35,480 --> 00:13:37,640 how good or bad they think one of the candidates are, 239 00:13:37,980 --> 00:13:39,140 and the spatial dimension, 240 00:13:39,140 --> 00:13:42,920 which candidate is closer to them on policy directions. 241 00:13:43,380 --> 00:13:45,160 If the echo chamber is reducing 242 00:13:45,160 --> 00:13:47,340 how much they learn about the vertical dimension, 243 00:13:47,840 --> 00:13:50,220 then it has to make voting more responsive 244 00:13:50,220 --> 00:13:52,520 to the horizontal, to the policy dimension. 245 00:13:53,040 --> 00:13:55,940 Okay, and so that means from the candidate's perspective, 246 00:13:56,200 --> 00:13:58,780 the impact of moderating is higher, right? 247 00:13:58,820 --> 00:14:00,980 If I moderate, this increases the probability 248 00:14:00,980 --> 00:14:04,120 that I win by a lot more in the presence of an echo chamber 249 00:14:04,120 --> 00:14:06,560 than in the world without an echo chamber 250 00:14:06,560 --> 00:14:07,940 where voters are learning a lot 251 00:14:07,940 --> 00:14:09,260 about these vertical dimensions, 252 00:14:09,520 --> 00:14:10,660 and so that matters a lot. 253 00:14:10,860 --> 00:14:13,800 And so what happens is in the model with the echo chamber, 254 00:14:14,140 --> 00:14:16,780 candidate voting turns out to be more responsive 255 00:14:16,780 --> 00:14:18,020 to policy platforms, 256 00:14:18,520 --> 00:14:19,900 and so this makes the way 257 00:14:19,900 --> 00:14:23,400 that candidates balance these things out more moderate, 258 00:14:23,640 --> 00:14:25,380 right, because if the probability of winning 259 00:14:25,380 --> 00:14:27,900 is much more responsive to their policy platforms, 260 00:14:28,280 --> 00:14:30,820 they'll moderate more, polarise less. 261 00:14:31,200 --> 00:14:33,100 And so in fact, when there's less learning 262 00:14:33,100 --> 00:14:34,480 because of these echo chambers, 263 00:14:34,480 --> 00:14:36,580 there's more responsiveness to policy, 264 00:14:36,800 --> 00:14:39,420 and so the candidates wind up being more moderate 265 00:14:39,420 --> 00:14:42,500 in a model without these biassed or selective exposure. 266 00:14:42,960 --> 00:14:44,340 Yeah, so the punchline is sort of, 267 00:14:44,580 --> 00:14:46,620 I think, the opposite of conventional wisdom. 268 00:14:46,880 --> 00:14:50,160 As you make the media a more biassed source of information 269 00:14:50,160 --> 00:14:51,620 about these vertical dimensions, 270 00:14:52,100 --> 00:14:54,600 you necessarily make candidates believe 271 00:14:54,600 --> 00:14:56,040 that voters are more responsive 272 00:14:56,040 --> 00:14:57,500 to the horizontal dimension, 273 00:14:57,720 --> 00:14:59,520 and this causes them to moderate. 274 00:15:00,160 --> 00:15:02,020 And so it's sort of the opposite mechanism 275 00:15:02,020 --> 00:15:04,240 that you would get compared to a model 276 00:15:04,240 --> 00:15:07,020 in which selective exposure was just about policy. 277 00:15:07,400 --> 00:15:09,100 Okay, so the additional dimension 278 00:15:09,100 --> 00:15:12,540 leads to this a little bit counterintuitive fact 279 00:15:12,540 --> 00:15:14,320 that actually the polarisation 280 00:15:14,320 --> 00:15:16,920 does not increase through the echo chambers, 281 00:15:17,180 --> 00:15:20,160 but the candidates actually choose the moderate platforms. 282 00:15:20,560 --> 00:15:21,600 Right, and so, you know, 283 00:15:21,620 --> 00:15:23,140 I think there's plenty of work out there 284 00:15:23,140 --> 00:15:25,520 that traces out sort of the conventional wisdom 285 00:15:25,520 --> 00:15:28,920 that polarisation has to be the result of echo chambers. 286 00:15:29,520 --> 00:15:30,700 You know, what this work says is, 287 00:15:30,740 --> 00:15:31,420 well, wait a second, 288 00:15:31,500 --> 00:15:33,960 if we try to think about that direct channel, 289 00:15:33,960 --> 00:15:36,740 it matters what we think selective exposure is about, 290 00:15:36,980 --> 00:15:39,340 what we think the first-order effects of media are. 291 00:15:39,900 --> 00:15:41,860 And so if what we're learning the most about 292 00:15:41,860 --> 00:15:45,160 from social media are vertical dimensions, 293 00:15:45,460 --> 00:15:46,880 then it shouldn't be the case 294 00:15:46,880 --> 00:15:49,720 that echo chambers are the causal reason for polarisation. 295 00:15:50,420 --> 00:15:52,720 Now, I think polarisation has been happening, 296 00:15:52,900 --> 00:15:53,760 certainly in the US, 297 00:15:53,940 --> 00:15:56,220 probably in European cases as well, 298 00:15:56,620 --> 00:15:58,920 and we think echo chambers are probably happening, 299 00:15:59,080 --> 00:16:00,980 but this makes me less convinced 300 00:16:00,980 --> 00:16:03,720 that it's echo chambers causing polarisation 301 00:16:03,720 --> 00:16:06,340 rather there must be some other things floating around, 302 00:16:06,860 --> 00:16:08,800 and simultaneously we're getting polarisation 303 00:16:09,320 --> 00:16:11,280 and we're getting echo chambers increasing, 304 00:16:11,740 --> 00:16:13,840 but that's not a simple causal story. 305 00:16:14,120 --> 00:16:14,860 Do you have a theory 306 00:16:14,860 --> 00:16:17,200 about what the other influences could be? 307 00:16:17,340 --> 00:16:19,200 I have ideas, I don't have a theory, 308 00:16:19,420 --> 00:16:20,780 and I don't have an empirical story, 309 00:16:20,880 --> 00:16:23,020 although I will admit it's sort of a mixed bag. 310 00:16:23,140 --> 00:16:24,260 On the simplistic view, 311 00:16:24,380 --> 00:16:27,200 this is a reason why I think there's plenty of job security 312 00:16:27,200 --> 00:16:28,320 for people like me. 313 00:16:28,540 --> 00:16:30,320 Like, there's still a lot of open questions 314 00:16:30,320 --> 00:16:32,680 for us to think through equilibrium forces 315 00:16:32,680 --> 00:16:36,020 and think through what we can learn from data, right? 316 00:16:36,060 --> 00:16:37,000 There's a challenge here, 317 00:16:37,140 --> 00:16:39,280 which is polarisation is a story, 318 00:16:39,460 --> 00:16:40,620 I think for most people, 319 00:16:40,780 --> 00:16:42,640 about preferences of individuals 320 00:16:42,640 --> 00:16:44,580 and preferences of parties, 321 00:16:44,580 --> 00:16:48,060 or maybe it's a story about behaviour that we observe, 322 00:16:48,400 --> 00:16:51,420 and the challenge is I can see who votes in elections 323 00:16:51,420 --> 00:16:53,240 and I can see how they vote 324 00:16:53,240 --> 00:16:56,100 and I can see the platforms parties take, 325 00:16:56,560 --> 00:16:59,340 but I don't directly see the preferences of the parties 326 00:16:59,340 --> 00:17:00,840 or the preferences of the voters, 327 00:17:01,400 --> 00:17:04,020 and what game theory teaches me is I need a model 328 00:17:04,020 --> 00:17:05,520 and I need to let an equilibrium story 329 00:17:05,520 --> 00:17:08,780 to think about how primitives like preferences 330 00:17:08,780 --> 00:17:11,520 or information map into observables 331 00:17:11,520 --> 00:17:13,440 like platforms and voting, 332 00:17:13,660 --> 00:17:15,579 and unless I can think about those channels, 333 00:17:15,760 --> 00:17:18,300 I have a hard time figuring out what's driving what. 334 00:17:18,420 --> 00:17:20,300 I have co-authors and friends 335 00:17:20,300 --> 00:17:22,940 that are much better statisticians than I am 336 00:17:22,940 --> 00:17:24,380 that are sort of bridging this gap 337 00:17:24,380 --> 00:17:25,680 and doing structural work 338 00:17:25,680 --> 00:17:28,720 to try to reverse engineer from equilibrium analysis 339 00:17:28,720 --> 00:17:30,100 what the data teaches us, 340 00:17:30,100 --> 00:17:31,260 and so I'm optimistic 341 00:17:31,260 --> 00:17:33,380 that people are trying to answer these questions, 342 00:17:33,620 --> 00:17:34,780 but I don't think we know why. 343 00:17:35,100 --> 00:17:37,140 Okay, there's a lot of mystery left then. 344 00:17:37,980 --> 00:17:38,540 That's fair. 345 00:17:38,700 --> 00:17:40,700 A lot of room for research also, of course. 346 00:17:40,940 --> 00:17:43,420 Talking about mystery maybe a little bit more, 347 00:17:43,560 --> 00:17:45,800 so there is one thing you did look at, 348 00:17:45,920 --> 00:17:47,520 and that is what happens in your model 349 00:17:47,520 --> 00:17:49,920 when the voters are not fully rational, 350 00:17:50,240 --> 00:17:51,940 at least in the standard economic sense, 351 00:17:52,060 --> 00:17:53,660 so what does happen 352 00:17:53,660 --> 00:17:56,400 when you assume bounded rationality by the voters? 353 00:17:56,560 --> 00:17:57,920 Does that change your findings 354 00:17:57,920 --> 00:18:00,000 or does it not impact them at all? 355 00:18:00,220 --> 00:18:02,400 Yeah, this was a sort of a fun feature 356 00:18:02,400 --> 00:18:04,620 of this analysis for my co-authors and I, 357 00:18:05,140 --> 00:18:06,780 and so if we take a step back 358 00:18:06,780 --> 00:18:09,400 and think about how we're modelling echo chambers, 359 00:18:09,720 --> 00:18:12,300 I had this idea that I tune into a source 360 00:18:12,300 --> 00:18:14,380 that's only going to tell me things that I want to hear, 361 00:18:14,500 --> 00:18:16,560 it's going to hide things that I don't want to hear, 362 00:18:17,100 --> 00:18:19,800 and so there's sort of this second-order question of, 363 00:18:20,220 --> 00:18:23,040 is no news bad news or is no news no news, right? 364 00:18:23,100 --> 00:18:25,940 So if I've chosen to listen to people 365 00:18:25,940 --> 00:18:27,700 that will tell me what I want to hear, 366 00:18:27,700 --> 00:18:29,280 and I'm not hearing very much, 367 00:18:29,460 --> 00:18:32,440 presumably that means that the facts are against me, right? 368 00:18:32,680 --> 00:18:33,580 If I don't hear anything, 369 00:18:33,660 --> 00:18:35,620 that must be because there's some really compelling story 370 00:18:35,620 --> 00:18:37,440 that my candidate's not very good 371 00:18:37,440 --> 00:18:39,160 and the other candidate is much better, 372 00:18:39,320 --> 00:18:41,040 and so there's a question for us, 373 00:18:41,100 --> 00:18:44,400 do I choose to live in an echo chamber or information silo 374 00:18:44,400 --> 00:18:46,020 and only hear what I want to hear, 375 00:18:46,260 --> 00:18:48,040 but on election day do I remember 376 00:18:48,040 --> 00:18:50,980 that I chose to get this biassed source of information 377 00:18:50,980 --> 00:18:53,320 and think that no news is bad news, 378 00:18:53,580 --> 00:18:55,060 or do I forget about this process? 379 00:18:55,280 --> 00:18:57,040 And so we don't want to jump in 380 00:18:57,040 --> 00:19:00,440 and take the hardcore full rationality perspective 381 00:19:00,440 --> 00:19:03,680 and say that our citizens have chosen to have echo chambers, 382 00:19:03,860 --> 00:19:05,920 but they're fully aware of the selection effects. 383 00:19:06,180 --> 00:19:08,520 Instead, what we do is we have a parameter. 384 00:19:08,840 --> 00:19:09,920 On one end of the spectrum, 385 00:19:10,180 --> 00:19:13,100 our voters are quite cognisant of what's going on. 386 00:19:13,460 --> 00:19:15,120 They're fully aware of these selection effects, 387 00:19:15,280 --> 00:19:16,840 and so if they don't hear news, 388 00:19:17,080 --> 00:19:18,980 they know that's bad news and they update accordingly. 389 00:19:19,360 --> 00:19:20,640 On the other end of the spectrum, 390 00:19:20,820 --> 00:19:22,980 they completely forget about these selection effects. 391 00:19:23,140 --> 00:19:24,180 They pretend they don't happen, 392 00:19:24,180 --> 00:19:27,200 and so if they hear good news about their candidate, 393 00:19:27,460 --> 00:19:28,560 that impacts them a lot. 394 00:19:28,660 --> 00:19:30,600 If they hear bad news about the other candidate, 395 00:19:30,780 --> 00:19:31,740 that impacts them a lot. 396 00:19:31,820 --> 00:19:32,820 They don't hear very much. 397 00:19:33,040 --> 00:19:35,200 They say, oh, there was nothing to learn, okay? 398 00:19:35,560 --> 00:19:37,640 And so we then go through and solve, 399 00:19:37,840 --> 00:19:39,480 so what is the equilibrium to this game 400 00:19:39,480 --> 00:19:42,440 for all the possible levels of how rational 401 00:19:42,440 --> 00:19:44,020 or Bayesian our people are 402 00:19:44,020 --> 00:19:46,600 versus how boundedly rational they are 403 00:19:46,600 --> 00:19:48,340 in this very particular sense? 404 00:19:48,480 --> 00:19:50,020 And what we found is, as I mentioned, 405 00:19:50,220 --> 00:19:53,800 so compared to neutral, unbiased information, 406 00:19:53,800 --> 00:19:56,100 echo chambers cause the equilibria 407 00:19:56,100 --> 00:19:57,280 to have more moderation. 408 00:19:58,120 --> 00:20:01,180 However, if I think about the presence of an echo chamber, 409 00:20:01,560 --> 00:20:03,640 as I increase the bias, 410 00:20:03,800 --> 00:20:06,200 and so as I move away from full rationality, 411 00:20:06,460 --> 00:20:09,160 I do increase the level of polarisation, okay? 412 00:20:09,160 --> 00:20:11,840 And so there is this partial explanation 413 00:20:11,840 --> 00:20:13,600 for maybe what's driving polarisation, 414 00:20:14,000 --> 00:20:16,780 which is the presence of echo chambers 415 00:20:16,780 --> 00:20:20,280 with people becoming increasingly uncognizant 416 00:20:20,280 --> 00:20:21,980 that they're getting biassed information, 417 00:20:21,980 --> 00:20:24,720 that does lead to a pattern of increased polarisation. 418 00:20:25,140 --> 00:20:26,540 So that could be one explanation 419 00:20:26,540 --> 00:20:28,860 for why we actually see the polarisation, yeah. 420 00:20:29,020 --> 00:20:31,700 Yeah, selective exposure and full cognition 421 00:20:31,700 --> 00:20:33,200 that we had selective exposure 422 00:20:33,200 --> 00:20:34,440 doesn't get you to polarisation, 423 00:20:34,920 --> 00:20:36,980 but given we have selective exposure, 424 00:20:37,440 --> 00:20:39,560 people that are forgetful about the fact 425 00:20:39,560 --> 00:20:40,700 that there's a selection bias, 426 00:20:40,700 --> 00:20:42,560 that does lead to changes 427 00:20:42,560 --> 00:20:44,280 in how much polarisation we should observe 428 00:20:44,280 --> 00:20:45,000 from the candidates. 429 00:20:45,320 --> 00:20:47,560 Yeah, those people that are maybe not suspicious 430 00:20:47,560 --> 00:20:51,260 about getting no news about any candidate, 431 00:20:51,260 --> 00:20:53,300 those are the ones that could drive this 432 00:20:53,300 --> 00:20:55,540 rather than the right direction of polarisation, yeah. 433 00:20:55,680 --> 00:20:56,820 Right, and the pathway is that 434 00:20:56,820 --> 00:20:59,080 if I'm cognisant about the selection effects, 435 00:20:59,240 --> 00:21:01,220 I know that I didn't learn very much from the media, 436 00:21:01,420 --> 00:21:02,560 and so I don't know very much 437 00:21:02,560 --> 00:21:03,560 about the vertical dimension, 438 00:21:03,960 --> 00:21:05,900 and so I have to sort of base my decision 439 00:21:05,900 --> 00:21:07,480 a bit more on the horizontal dimension. 440 00:21:07,960 --> 00:21:10,880 But if I'm not aware of the selection effects 441 00:21:10,880 --> 00:21:11,700 and I get fooled, 442 00:21:12,060 --> 00:21:13,260 I think I'm learning a lot 443 00:21:13,260 --> 00:21:14,360 about the vertical dimension, 444 00:21:14,500 --> 00:21:16,620 so I underweight the horizontal dimension, 445 00:21:16,820 --> 00:21:19,240 and so candidates would be relatively more polarised. 446 00:21:19,240 --> 00:21:20,920 Interesting explanation, yeah. 447 00:21:21,120 --> 00:21:23,120 Probably also have to look at some more research 448 00:21:23,120 --> 00:21:26,480 to find out whether that is really the true one reason. 449 00:21:26,680 --> 00:21:27,600 Exactly, yeah. 450 00:21:27,740 --> 00:21:29,720 You have worked in this field 451 00:21:29,720 --> 00:21:30,820 of electoral competition, 452 00:21:30,920 --> 00:21:32,660 also beyond this echo chambers, 453 00:21:33,040 --> 00:21:34,660 and is there anything else 454 00:21:34,660 --> 00:21:36,200 that could help explain why 455 00:21:36,200 --> 00:21:38,200 and how candidates choose platforms 456 00:21:38,200 --> 00:21:39,220 in your research? 457 00:21:39,460 --> 00:21:40,620 Yeah, so an intuition 458 00:21:40,620 --> 00:21:42,460 that came out of this analysis, 459 00:21:42,640 --> 00:21:44,180 which we don't emphasise in the paper 460 00:21:44,180 --> 00:21:45,660 that I guess I've been summarising, 461 00:21:45,940 --> 00:21:48,740 is a source for added strategic uncertainty 462 00:21:48,740 --> 00:21:50,000 in elections in general, 463 00:21:50,560 --> 00:21:52,020 and so the logic is this. 464 00:21:52,380 --> 00:21:53,860 Depending on the uncertainty 465 00:21:53,860 --> 00:21:55,980 that the candidates face 466 00:21:55,980 --> 00:21:57,220 about where voters are, 467 00:21:57,380 --> 00:21:59,560 it's possible that they're pretty convinced 468 00:21:59,560 --> 00:22:00,780 there's a pretty tight band 469 00:22:00,780 --> 00:22:02,540 of where the median is in the middle, 470 00:22:02,660 --> 00:22:03,940 but then there are these fat tails, 471 00:22:04,180 --> 00:22:05,920 and so there's non-trivial chances 472 00:22:05,920 --> 00:22:08,240 that something dramatic is going to happen 473 00:22:08,240 --> 00:22:09,040 between the time 474 00:22:09,040 --> 00:22:11,520 that they sketch out platforms and voting 475 00:22:11,520 --> 00:22:13,020 in which they could win 476 00:22:13,020 --> 00:22:14,240 with very extreme positions, 477 00:22:14,600 --> 00:22:15,980 and so the logic is that 478 00:22:15,980 --> 00:22:17,440 I'm a right-wing party. 479 00:22:17,440 --> 00:22:18,880 I think it's pretty likely 480 00:22:18,880 --> 00:22:20,640 that I know where the median voter is, 481 00:22:20,740 --> 00:22:22,900 and so if both parties announce 482 00:22:22,900 --> 00:22:24,520 very centrist, not equivalent, 483 00:22:24,660 --> 00:22:25,960 but very centrist platforms, 484 00:22:26,340 --> 00:22:28,220 where the median is somewhere in the middle 485 00:22:28,220 --> 00:22:29,720 is going to determine which one wins, 486 00:22:29,980 --> 00:22:31,280 but there's a chance 487 00:22:31,280 --> 00:22:32,900 that something dramatic happens, 488 00:22:33,200 --> 00:22:33,920 and I could win 489 00:22:33,920 --> 00:22:35,420 with a very extreme platform, 490 00:22:35,480 --> 00:22:37,100 very close to the edge of the spectrum 491 00:22:37,100 --> 00:22:38,100 where I'd like to be. 492 00:22:38,240 --> 00:22:39,500 That's an unlikely outcome, 493 00:22:39,820 --> 00:22:41,040 so I'm not likely to win 494 00:22:41,040 --> 00:22:42,400 if I take this extreme platform, 495 00:22:42,620 --> 00:22:44,300 but I'm super happy if I do win, 496 00:22:44,520 --> 00:22:46,240 so there's sort of this Hail Mary logic 497 00:22:46,240 --> 00:22:48,480 of maybe I sometimes announce 498 00:22:48,480 --> 00:22:49,660 very extreme platforms. 499 00:22:50,120 --> 00:22:51,360 It's very likely that I lose, 500 00:22:51,780 --> 00:22:53,200 but the difference between losing 501 00:22:53,200 --> 00:22:54,260 with a moderate platform 502 00:22:54,260 --> 00:22:56,700 and winning with a moderate platform 503 00:22:56,700 --> 00:22:57,600 is pretty negligible. 504 00:22:57,780 --> 00:22:58,860 We're getting centrist policy. 505 00:22:59,500 --> 00:23:00,480 The upside from winning 506 00:23:00,480 --> 00:23:01,660 with a very extreme platform 507 00:23:01,660 --> 00:23:02,680 that I like is huge, 508 00:23:02,880 --> 00:23:04,820 and so there are this sort of forces 509 00:23:04,820 --> 00:23:06,440 in equilibrium that require 510 00:23:06,440 --> 00:23:08,020 that every once in a while, 511 00:23:08,260 --> 00:23:09,220 with some probability, 512 00:23:09,540 --> 00:23:10,320 candidates are taking 513 00:23:10,320 --> 00:23:11,380 quite extreme platforms, 514 00:23:11,740 --> 00:23:14,840 and so there's sort of regularity conditions 515 00:23:14,840 --> 00:23:15,780 that game theorists 516 00:23:15,780 --> 00:23:17,060 have tended to use 517 00:23:17,060 --> 00:23:18,120 that make the models 518 00:23:18,120 --> 00:23:19,540 have nice, tight predictions 519 00:23:19,540 --> 00:23:21,160 about where the candidates locate, 520 00:23:21,640 --> 00:23:22,740 but there are substantively 521 00:23:22,740 --> 00:23:24,000 compelling stories about 522 00:23:24,000 --> 00:23:25,580 maybe there's a bit more uncertainty 523 00:23:25,580 --> 00:23:26,440 about voters 524 00:23:26,440 --> 00:23:28,040 that sort of kill those results, 525 00:23:28,380 --> 00:23:30,400 and then the equilibrium story 526 00:23:30,400 --> 00:23:32,240 has to have more strategic uncertainty 527 00:23:32,240 --> 00:23:32,900 that we thought, 528 00:23:33,360 --> 00:23:34,820 more low probability 529 00:23:34,820 --> 00:23:36,580 that candidates do crazy stuff 530 00:23:36,580 --> 00:23:38,200 because those crazy things might win, 531 00:23:38,480 --> 00:23:39,600 and the downside from losing 532 00:23:39,600 --> 00:23:41,360 with a crazy platform isn't very high. 533 00:23:41,740 --> 00:23:42,280 At the time, 534 00:23:42,400 --> 00:23:43,560 we sort of thought through this. 535 00:23:43,740 --> 00:23:45,240 I had an undergrad at Yale. 536 00:23:45,240 --> 00:23:47,520 He's now a PhD student at Princeton, 537 00:23:47,720 --> 00:23:48,380 Justice Hirasha. 538 00:23:48,600 --> 00:23:49,600 We're sort of fleshing out 539 00:23:49,600 --> 00:23:50,360 some of this logic 540 00:23:50,360 --> 00:23:51,840 and talking with some friends 541 00:23:51,840 --> 00:23:53,300 about doing some empirical work 542 00:23:53,300 --> 00:23:54,380 based on this as well. 543 00:23:54,700 --> 00:23:55,020 Yeah, great. 544 00:23:55,100 --> 00:23:56,760 That would be my next question, 545 00:23:56,860 --> 00:23:58,280 whether there's anything planned 546 00:23:58,280 --> 00:23:59,700 to test your predictions 547 00:23:59,700 --> 00:24:01,060 also about the echo chambers, 548 00:24:01,260 --> 00:24:02,920 of course, in the empirical sense. 549 00:24:03,260 --> 00:24:03,660 Yeah, no. 550 00:24:03,720 --> 00:24:04,720 In the narrow sense, 551 00:24:04,820 --> 00:24:05,640 we're sort of trying 552 00:24:05,640 --> 00:24:06,780 to come up with a model 553 00:24:06,780 --> 00:24:08,460 that lets us track 554 00:24:08,460 --> 00:24:11,840 how variation in a few different things 555 00:24:11,840 --> 00:24:14,080 might explain polarisation, right? 556 00:24:14,080 --> 00:24:15,400 Not just looking at 557 00:24:15,400 --> 00:24:17,180 one causal path at a time, 558 00:24:17,320 --> 00:24:18,000 but thinking about 559 00:24:18,000 --> 00:24:19,360 as many different pieces 560 00:24:19,360 --> 00:24:20,680 of the model as we can 561 00:24:20,680 --> 00:24:22,040 and seeing how they co-vary 562 00:24:22,040 --> 00:24:23,920 and seeing if that can explain 563 00:24:23,920 --> 00:24:26,460 a pattern of what causes polarisation. 564 00:24:26,900 --> 00:24:28,000 This is work to be done, 565 00:24:28,100 --> 00:24:29,060 so I can't really advertise 566 00:24:29,060 --> 00:24:29,920 what we're going to get to. 567 00:24:30,320 --> 00:24:31,160 I don't know the answer. 568 00:24:31,560 --> 00:24:33,040 But then, as I alluded to before, 569 00:24:33,200 --> 00:24:35,020 there's sort of this larger research programme 570 00:24:35,020 --> 00:24:36,800 that I'm occasionally involved in, 571 00:24:36,840 --> 00:24:37,800 but there are other people 572 00:24:37,800 --> 00:24:39,000 that are much more engaged 573 00:24:39,000 --> 00:24:41,480 of trying to use equilibrium models 574 00:24:41,480 --> 00:24:43,320 and careful empirical work 575 00:24:43,320 --> 00:24:45,000 to do some structural analysis 576 00:24:45,000 --> 00:24:46,300 and figure these things out. 577 00:24:46,520 --> 00:24:48,000 So a lot of work to be done. 578 00:24:48,140 --> 00:24:49,800 I think we genuinely don't know 579 00:24:49,800 --> 00:24:51,100 what's causing polarisation. 580 00:24:51,380 --> 00:24:52,740 The character of the empirical work 581 00:24:52,740 --> 00:24:54,200 is a lot of really good papers, 582 00:24:54,400 --> 00:24:55,420 each of which tells us 583 00:24:55,420 --> 00:24:56,400 that any one story 584 00:24:56,400 --> 00:24:57,540 about what causes polarisation 585 00:24:57,540 --> 00:24:58,620 can't be correct. 586 00:24:58,780 --> 00:25:00,060 So we observe polarisation, 587 00:25:00,400 --> 00:25:01,700 we observe lots of things varying, 588 00:25:01,900 --> 00:25:03,040 and we don't quite understand 589 00:25:03,040 --> 00:25:04,320 the determinants yet. 590 00:25:04,720 --> 00:25:06,540 Yeah, it's a very current topic, 591 00:25:06,800 --> 00:25:08,700 so I'm sure we can expect 592 00:25:08,700 --> 00:25:10,560 a lot more research in that area 593 00:25:10,560 --> 00:25:11,980 in solving some of these mysteries. 594 00:25:11,980 --> 00:25:13,480 But for now, 595 00:25:13,600 --> 00:25:14,940 thank you very much for giving us 596 00:25:14,940 --> 00:25:16,360 this very interesting insight 597 00:25:16,360 --> 00:25:18,120 into your research and the outcomes. 598 00:25:18,540 --> 00:25:19,440 And thank you very much 599 00:25:19,440 --> 00:25:20,440 for coming on the podcast 600 00:25:20,440 --> 00:25:22,100 and giving us your time, of course. 601 00:25:22,460 --> 00:25:23,040 Oh, thank you. 602 00:25:23,120 --> 00:25:24,120 This is a lot of fun. 603 00:25:24,320 --> 00:25:25,160 I like the idea that 604 00:25:25,160 --> 00:25:26,900 we're all thinking about strategy 605 00:25:26,900 --> 00:25:27,580 and incentives. 606 00:25:28,020 --> 00:25:29,000 So, cool. 607 00:25:29,140 --> 00:25:29,560 That's great. 608 00:25:29,660 --> 00:25:29,980 Me too. 609 00:25:31,500 --> 00:25:32,920 We hope you have enjoyed 610 00:25:32,920 --> 00:25:33,880 this week's episode 611 00:25:33,880 --> 00:25:35,800 and we would love to hear your feedback, 612 00:25:36,280 --> 00:25:37,300 comments, or questions. 613 00:25:37,680 --> 00:25:38,920 Just send us an email 614 00:25:38,920 --> 00:25:40,900 to gamechanger at TWS 615 00:25:40,900 --> 00:25:42,980 minus partners dot com. 616 00:25:43,420 --> 00:25:44,720 If you have enjoyed this episode, 617 00:25:45,040 --> 00:25:46,500 please also consider 618 00:25:46,500 --> 00:25:48,020 subscribing to our podcast 619 00:25:48,020 --> 00:25:49,800 or leaving a review or rating 620 00:25:49,800 --> 00:25:52,600 using Spotify, Apple Podcasts, 621 00:25:52,640 --> 00:25:54,320 or your favourite podcast app. 622 00:25:54,600 --> 00:25:55,900 Many thanks for listening 623 00:25:55,900 --> 00:25:58,420 and we hope you also tune in 624 00:25:58,420 --> 00:26:00,800 for the next episode of Game Changer.