WEBVTT 00:00:00.000 --> 00:00:02.440 [SQUEAKING] 00:00:02.440 --> 00:00:04.392 [RUSTLING] 00:00:04.392 --> 00:00:06.344 [CLICKING] 00:00:10.495 --> 00:00:11.620 FRANK SCHILBACH: All right. 00:00:11.620 --> 00:00:15.360 Welcome, everyone, to lecture 13 of 14.13. 00:00:15.360 --> 00:00:18.210 This is my last lecture on social preferences. 00:00:18.210 --> 00:00:21.870 I hope all of you are doing fine remotely. 00:00:21.870 --> 00:00:22.510 Hang in there. 00:00:22.510 --> 00:00:24.660 I know this is a very difficult time. 00:00:24.660 --> 00:00:27.090 This lecture is a bit of a more uplifting lecture 00:00:27.090 --> 00:00:28.620 than the previous ones. 00:00:28.620 --> 00:00:31.140 While the previous ones have looked at social preferences, 00:00:31.140 --> 00:00:33.450 how we can measure social preferences, 00:00:33.450 --> 00:00:37.200 and how can we perhaps isolate pure altruism-- 00:00:37.200 --> 00:00:41.490 do people really want others to be better for their own sake, 00:00:41.490 --> 00:00:46.350 even if nobody knows about it, either others or whether 00:00:46.350 --> 00:00:48.360 there's no feedback-- 00:00:48.360 --> 00:00:51.990 we sort of found out that like it seems like what looks a lot 00:00:51.990 --> 00:00:54.690 like altruism-- people are nice to each other-- 00:00:54.690 --> 00:00:56.760 is not pure altruism in the sense 00:00:56.760 --> 00:00:59.700 that like they really want others to do well. 00:00:59.700 --> 00:01:04.379 Instead, they want to either look good in front of others, 00:01:04.379 --> 00:01:06.810 or they're worried about reciprocity or sort 00:01:06.810 --> 00:01:09.480 of any other negative feedback from others, 00:01:09.480 --> 00:01:13.050 or it might be they might want to protect their self-image. 00:01:13.050 --> 00:01:15.270 They want to look good in front of themselves 00:01:15.270 --> 00:01:18.960 and they try to figure out ways to deceive themselves 00:01:18.960 --> 00:01:19.710 that they're nice. 00:01:19.710 --> 00:01:21.990 And given the opportunity to not be nice, 00:01:21.990 --> 00:01:24.480 people actually seem to be not as 00:01:24.480 --> 00:01:27.330 nice as one might think from pure dictator games 00:01:27.330 --> 00:01:30.270 or ultimatum games that we observe. 00:01:30.270 --> 00:01:33.090 This lecture is a bit more uplifting in the sense 00:01:33.090 --> 00:01:38.040 that it talks about field evidence on social preferences, 00:01:38.040 --> 00:01:40.440 and in particular you can ask the question 00:01:40.440 --> 00:01:46.890 about can policies increase pro-sociology in certain ways. 00:01:46.890 --> 00:01:51.275 We're going to talk about three broad sets of studies. 00:01:51.275 --> 00:01:53.400 First, we're going to talk about social preferences 00:01:53.400 --> 00:01:54.630 at the workplace. 00:01:54.630 --> 00:01:57.990 I'm going to talk about the impact of relative pay 00:01:57.990 --> 00:01:58.712 on productivity. 00:01:58.712 --> 00:02:00.420 We talked about this last time at the end 00:02:00.420 --> 00:02:01.920 of the lecture a little bit, but I'm 00:02:01.920 --> 00:02:05.190 going to just restart from scratch to sort of explain 00:02:05.190 --> 00:02:06.270 that a little bit better. 00:02:06.270 --> 00:02:08.280 Then we're going to talk about the morale 00:02:08.280 --> 00:02:10.020 effects of pay inequality. 00:02:10.020 --> 00:02:12.090 What happens to worker inequality 00:02:12.090 --> 00:02:17.657 when people are paid unequally, and is that a good idea? 00:02:17.657 --> 00:02:19.740 There's also a very nice paper in ethnic divisions 00:02:19.740 --> 00:02:21.940 and production in firms. 00:02:21.940 --> 00:02:25.620 We're going to talk about this a little bit in recitation. 00:02:25.620 --> 00:02:29.430 Number two-- we're going to talk about whether policies 00:02:29.430 --> 00:02:32.265 can induce pro-sociality. 00:02:32.265 --> 00:02:34.140 In particular we're going to focus on mixing. 00:02:34.140 --> 00:02:36.990 This is sort of called the contact hypothesis. 00:02:36.990 --> 00:02:39.360 When you mix people from different backgrounds, 00:02:39.360 --> 00:02:43.500 do they become nicer to each other in various ways? 00:02:43.500 --> 00:02:45.810 You'll discuss a very nice paper by Gautam Rao 00:02:45.810 --> 00:02:49.210 when mixing rich and poor children in school. 00:02:49.210 --> 00:02:52.980 And I talk very briefly about mixing cricket players in India 00:02:52.980 --> 00:02:57.630 to reduce tensions across castes or make people 00:02:57.630 --> 00:03:00.680 from different castes perhaps more prosocial to each other. 00:03:00.680 --> 00:03:02.430 And then I'm going to also briefly mention 00:03:02.430 --> 00:03:06.700 a study on mixing roommates in college by Corno et al. 00:03:06.700 --> 00:03:08.370 Finally, we're going to talk about 00:03:08.370 --> 00:03:10.950 whether people perhaps underestimate 00:03:10.950 --> 00:03:12.840 the benefits of prosociality. 00:03:12.840 --> 00:03:14.340 In particular, I'm going to show you 00:03:14.340 --> 00:03:17.880 a study on people undervaluing gratitude. 00:03:17.880 --> 00:03:21.150 So when people write letters of gratitude, 00:03:21.150 --> 00:03:23.430 do they understand how others might react, 00:03:23.430 --> 00:03:28.000 and is there perhaps some evidence that people are not 00:03:28.000 --> 00:03:32.430 prosocial enough because they misunderstand the effects 00:03:32.430 --> 00:03:34.920 that that might have on others? 00:03:34.920 --> 00:03:41.880 OK, so let me start with social preferences at the workplace. 00:03:41.880 --> 00:03:43.880 So Bandiera et al. is a very nice study 00:03:43.880 --> 00:03:48.030 that looks at the impact of relative pay on productivity. 00:03:48.030 --> 00:03:51.230 This is field evidence from the fruit farm in the UK. 00:03:51.230 --> 00:03:53.240 It's literally field evidence because it's 00:03:53.240 --> 00:03:56.840 from fruit-picking farms from fruit fields. 00:03:56.840 --> 00:04:00.060 The study looks at two types of payment schemes. 00:04:00.060 --> 00:04:01.700 These are piece rates-- 00:04:01.700 --> 00:04:04.550 workers are paid per unit of output. 00:04:04.550 --> 00:04:06.650 And the second scheme is relative pay-- workers 00:04:06.650 --> 00:04:09.200 are paid relative to others. 00:04:09.200 --> 00:04:12.140 We might think that relative pay is a good way 00:04:12.140 --> 00:04:15.020 to incentivize workers because you want to do better 00:04:15.020 --> 00:04:19.820 than your neighbor, and there's a computation of who does best. 00:04:19.820 --> 00:04:21.583 And in principle, it could be that that, 00:04:21.583 --> 00:04:23.750 really, if you only care about yourself, that really 00:04:23.750 --> 00:04:26.480 gets people to work very hard. 00:04:26.480 --> 00:04:28.070 Now, what Bandiera et al. look at 00:04:28.070 --> 00:04:31.940 is, how does the introduction of relative pay 00:04:31.940 --> 00:04:34.940 affect workers' effort and output? 00:04:34.940 --> 00:04:39.770 Now, one thing to notice is that while relative incentives might 00:04:39.770 --> 00:04:43.190 be good because you might induce people to want to be better 00:04:43.190 --> 00:04:46.460 than their neighbors or their co-workers, 00:04:46.460 --> 00:04:51.300 there's also a negative externality of relative pay. 00:04:51.300 --> 00:04:53.870 That is the idea that increasing your own pay 00:04:53.870 --> 00:04:56.235 comes at the costs of others pay. 00:04:56.235 --> 00:04:57.860 That is to say, if there's relative pay 00:04:57.860 --> 00:05:00.140 and I work really hard, I'm paid how 00:05:00.140 --> 00:05:03.515 I do relative to my friends or others that I'm working with. 00:05:03.515 --> 00:05:06.140 Well, if I'm working really hard and do better than my friends, 00:05:06.140 --> 00:05:07.430 I'm going to be paid more. 00:05:07.430 --> 00:05:11.060 But at the same time, for any effort that I put in, 00:05:11.060 --> 00:05:13.580 my friends are going to be paid less. 00:05:13.580 --> 00:05:19.370 And so workers anticipating these impacts on others 00:05:19.370 --> 00:05:22.105 might reduce their effort if they care about others. 00:05:22.105 --> 00:05:24.230 If I'm really worried about if there's relative pay 00:05:24.230 --> 00:05:26.750 and I work really hard, my friends are going to look bad. 00:05:26.750 --> 00:05:30.350 They're going to be relatively unproductive, 00:05:30.350 --> 00:05:33.740 and I worry that they might not be paid enough. 00:05:33.740 --> 00:05:37.250 Then I might reduce my effort. 00:05:37.250 --> 00:05:38.940 And then everybody might do that, 00:05:38.940 --> 00:05:40.940 and then the introduction of relative pay 00:05:40.940 --> 00:05:43.220 or the relative pay schemes might actually not 00:05:43.220 --> 00:05:48.290 be particularly effective or an effective way of incentivizing 00:05:48.290 --> 00:05:49.980 and motivating workers. 00:05:49.980 --> 00:05:53.210 So there's sort of this tension that relative pay in principle 00:05:53.210 --> 00:05:54.180 could work quite well. 00:05:54.180 --> 00:05:56.960 But if people really care a lot about others, 00:05:56.960 --> 00:05:58.250 then it might not work at all. 00:05:58.250 --> 00:06:00.800 It might actually backfire in some ways. 00:06:00.800 --> 00:06:05.150 Notice that it could be either that people care about others 00:06:05.150 --> 00:06:08.120 for their own sake in the sense that you might care 00:06:08.120 --> 00:06:11.870 about them because you want them to have high earnings, 00:06:11.870 --> 00:06:14.360 or you might care about others' earnings, 00:06:14.360 --> 00:06:17.395 not because you actually care about the outcomes per se, 00:06:17.395 --> 00:06:18.770 but rather because you're worried 00:06:18.770 --> 00:06:20.570 about negative repercussions. 00:06:20.570 --> 00:06:22.700 If these are your friends and you work really hard, 00:06:22.700 --> 00:06:25.190 well, at nights, your friends might really not be happy, 00:06:25.190 --> 00:06:28.160 and there might be retribution against you socially 00:06:28.160 --> 00:06:31.260 or in some other ways if you work too hard, if you caused 00:06:31.260 --> 00:06:35.020 them their earnings to go down. 00:06:35.020 --> 00:06:37.270 OK, so now, what does this paper do? 00:06:37.270 --> 00:06:39.610 It looks at how does switching to piece rates 00:06:39.610 --> 00:06:42.370 affect people's productivity? 00:06:42.370 --> 00:06:46.210 So the study has personnel data from a food farm in the UK. 00:06:46.210 --> 00:06:49.720 They measure productivity as a function of their compensation 00:06:49.720 --> 00:06:51.130 scheme. 00:06:51.130 --> 00:06:53.500 This is a quasi-field experiment. 00:06:53.500 --> 00:06:55.480 The timeline is as follows. 00:06:55.480 --> 00:06:58.750 For the first eight weeks of the 2002 picking season, 00:06:58.750 --> 00:07:02.830 fruit pickers were compensated on a relative performance 00:07:02.830 --> 00:07:03.520 scheme. 00:07:03.520 --> 00:07:09.010 So the per fruit piece rate is decreasing 00:07:09.010 --> 00:07:11.290 in average productivity. 00:07:11.290 --> 00:07:13.060 So if everybody works really hard 00:07:13.060 --> 00:07:15.490 and I don't, if I'm relatively bad compared to others, 00:07:15.490 --> 00:07:19.420 I get less than others. 00:07:19.420 --> 00:07:21.220 This is an incentive, as I discussed, 00:07:21.220 --> 00:07:25.390 to keep the productivity low if you care about others. 00:07:25.390 --> 00:07:27.070 And then in the next eight weeks, 00:07:27.070 --> 00:07:30.555 compensation will switch to a flat piece rate per fruit. 00:07:30.555 --> 00:07:31.930 So that's essentially just you're 00:07:31.930 --> 00:07:34.750 paid by however how much you produce, 00:07:34.750 --> 00:07:38.020 how many kilograms of fruit you collect. 00:07:38.020 --> 00:07:43.570 And the payment is entirely disjoint or independent of what 00:07:43.570 --> 00:07:44.390 other people do. 00:07:44.390 --> 00:07:47.140 That's a classic piece rate payment. 00:07:47.140 --> 00:07:49.810 So the externalities are entirely shut down. 00:07:49.810 --> 00:07:53.150 If I work really hard, I'm going to be paid. 00:07:53.150 --> 00:07:57.160 There's no effect whatsoever on my fellow workers. 00:07:57.160 --> 00:08:00.170 The switch was announced on the day the change took place, 00:08:00.170 --> 00:08:01.937 so it came as a surprise to workers. 00:08:01.937 --> 00:08:03.520 So we should not be worried about sort 00:08:03.520 --> 00:08:08.740 of pre-trends or productivity going up over time anyway. 00:08:08.740 --> 00:08:11.170 Now, what the paper finds is a dramatic increase 00:08:11.170 --> 00:08:14.620 in productivity with the introduction of piece rates. 00:08:14.620 --> 00:08:17.530 That is to say what you see here in the figure 00:08:17.530 --> 00:08:20.528 is the average worker productivity for two 00:08:20.528 --> 00:08:22.070 of the fields-- so that's essentially 00:08:22.070 --> 00:08:24.250 like broad, big fields where people work on, 00:08:24.250 --> 00:08:25.780 and I'm showing you the data for two 00:08:25.780 --> 00:08:28.690 of those fields, for which, essentially, on most days, 00:08:28.690 --> 00:08:31.720 there was actually production workers were picking fruit. 00:08:31.720 --> 00:08:37.090 What you see on the left side of the graph is when people were 00:08:37.090 --> 00:08:42.159 working under relative pay, there you see, essentially, 00:08:42.159 --> 00:08:44.169 relatively flat productivity. 00:08:44.169 --> 00:08:48.850 People picked about 5 kilograms per hour. 00:08:48.850 --> 00:08:51.740 There doesn't seem to be any trend upwards over time. 00:08:51.740 --> 00:08:53.830 So before the new policy was introduced, 00:08:53.830 --> 00:08:57.400 essentially, workers were producing or picking 00:08:57.400 --> 00:09:00.160 their fruit at a relatively constant rate over time, 00:09:00.160 --> 00:09:03.010 about 5 kilograms per hour. 00:09:03.010 --> 00:09:05.260 There was no trends, as I said. 00:09:05.260 --> 00:09:09.850 Now, you see the vertical line in the middle. 00:09:09.850 --> 00:09:13.480 This is when piece rates were, in fact, introduced. 00:09:13.480 --> 00:09:15.260 So there was no relative pay anymore, 00:09:15.260 --> 00:09:17.680 and now people were paid by piece rate. 00:09:17.680 --> 00:09:20.330 And that increased productivity by over 50%. 00:09:20.330 --> 00:09:24.310 So workers became a lot more productive after that. 00:09:24.310 --> 00:09:27.070 That's true for both of those fields. 00:09:27.070 --> 00:09:29.830 It is not the case that the average payment 00:09:29.830 --> 00:09:35.050 per unit of output per kilogram of fruit paid was actually 00:09:35.050 --> 00:09:36.100 was going up. 00:09:36.100 --> 00:09:38.350 So you might say, well, if the piece rate was actually 00:09:38.350 --> 00:09:42.400 higher under the piece rate payments 00:09:42.400 --> 00:09:45.250 on the right side of the screen, of the graph, 00:09:45.250 --> 00:09:47.440 then maybe workers would just be more productive. 00:09:47.440 --> 00:09:49.355 But in fact, if you look at the piece rate 00:09:49.355 --> 00:09:50.980 over time, if anything, the piece rates 00:09:50.980 --> 00:09:54.880 actually went down over time what the company was paying. 00:09:54.880 --> 00:09:57.610 So that is to say, the introduction 00:09:57.610 --> 00:10:01.480 of those piece rates, the productivity effect 00:10:01.480 --> 00:10:03.340 is not explained by workers being 00:10:03.340 --> 00:10:06.010 paid more per unit of output. 00:10:06.010 --> 00:10:08.050 Instead it seems to be really coming 00:10:08.050 --> 00:10:10.780 from the different incentives people workers have 00:10:10.780 --> 00:10:13.570 when it comes to how their productivity affects 00:10:13.570 --> 00:10:16.510 their fellow workers on the field. 00:10:16.510 --> 00:10:19.360 Now, you might have two potential explanations 00:10:19.360 --> 00:10:21.550 here for this evidence. 00:10:21.550 --> 00:10:23.260 One is social preferences. 00:10:23.260 --> 00:10:26.740 So you might work less to help others 00:10:26.740 --> 00:10:35.290 under relative incentives, even less when your friends benefit. 00:10:35.290 --> 00:10:38.340 You have many friends on a certain field 00:10:38.340 --> 00:10:39.600 that work with you. 00:10:39.600 --> 00:10:42.150 Well, then you might be particularly inclined 00:10:42.150 --> 00:10:48.220 to not work very hard because you care a lot more for them. 00:10:48.220 --> 00:10:50.778 And this is exactly what they find in this paper. 00:10:50.778 --> 00:10:52.320 These effects are stronger when there 00:10:52.320 --> 00:10:56.520 are more friends on the field for a particular worker. 00:10:56.520 --> 00:10:59.400 Second, however, it's also like a repeated game. 00:10:59.400 --> 00:11:03.180 That is to say, there's a low effort equilibrium, 00:11:03.180 --> 00:11:04.980 where, essentially, if there relative pay, 00:11:04.980 --> 00:11:07.590 we can just all agree everybody on the field 00:11:07.590 --> 00:11:10.800 might agree that not working particularly hard 00:11:10.800 --> 00:11:13.710 is a good idea because we are essentially 00:11:13.710 --> 00:11:15.270 paid relatively anyway. 00:11:15.270 --> 00:11:20.130 So if everybody doubles their effort, 00:11:20.130 --> 00:11:23.082 nothing is going to happen to a worker's compensation. 00:11:23.082 --> 00:11:25.290 So you might as well just decide everybody could just 00:11:25.290 --> 00:11:27.540 decide work half as hard, and you're 00:11:27.540 --> 00:11:28.680 going to be paid the same. 00:11:28.680 --> 00:11:32.760 And as long as we can sustain that equilibrium, 00:11:32.760 --> 00:11:33.870 that's a good idea. 00:11:33.870 --> 00:11:38.490 Of course, in any given period, any given worker 00:11:38.490 --> 00:11:41.597 might have the incentive to deviate from this equilibrium 00:11:41.597 --> 00:11:43.680 because on any given day, if you work really hard, 00:11:43.680 --> 00:11:47.230 you're going to be paid a lot more for that day. 00:11:47.230 --> 00:11:50.820 And so how might we be able to disentangle that? 00:11:50.820 --> 00:11:52.410 Well, what Bandiera et al. 00:11:52.410 --> 00:11:54.870 have is they have also some variation in the types 00:11:54.870 --> 00:11:57.910 of fruits that were collected. 00:11:57.910 --> 00:11:59.320 What do I mean by that? 00:11:59.320 --> 00:12:01.140 Well, if you think about what you 00:12:01.140 --> 00:12:04.770 need to be able to disentangle these, what you might want 00:12:04.770 --> 00:12:09.470 to have is differences in the observability of effort, right? 00:12:09.470 --> 00:12:11.980 So if I can see what my co-worker does, 00:12:11.980 --> 00:12:13.950 I can punish them very effectively. 00:12:13.950 --> 00:12:17.460 I can just know whenever Frank is working really hard, 00:12:17.460 --> 00:12:20.717 my co-workers might want to punish me. 00:12:20.717 --> 00:12:22.800 And then it's very hard for me to work really hard 00:12:22.800 --> 00:12:25.110 because I'm getting in trouble at nights 00:12:25.110 --> 00:12:27.810 after work with my co-workers. 00:12:27.810 --> 00:12:30.240 Or in the next day, everybody might also work really hard, 00:12:30.240 --> 00:12:32.460 and I don't want that. 00:12:32.460 --> 00:12:35.040 On the other hand, if I can hide how hard I'm working, 00:12:35.040 --> 00:12:38.370 well then, I might secretly work really hard, 00:12:38.370 --> 00:12:40.980 and my co-workers might not even find out. 00:12:40.980 --> 00:12:47.250 So then, if that's the case, in the repeated game equilibrium 00:12:47.250 --> 00:12:49.787 might just not be enforceable, as in I might just 00:12:49.787 --> 00:12:51.870 tell my friends, oh, I wasn't really working hard. 00:12:51.870 --> 00:12:53.010 But in fact, I was. 00:12:53.010 --> 00:12:58.590 And they have no way of finding out because I was unobservable. 00:12:58.590 --> 00:13:00.720 If, instead, I really care about them, 00:13:00.720 --> 00:13:04.740 if I'm really interested in their well-being 00:13:04.740 --> 00:13:06.600 overall and their payment, it doesn't 00:13:06.600 --> 00:13:08.730 matter whether my effort is observable. 00:13:08.730 --> 00:13:10.860 Regardless of how observable my effort is, 00:13:10.860 --> 00:13:13.440 I will not work very hard because not working 00:13:13.440 --> 00:13:15.430 hard makes them better off. 00:13:15.430 --> 00:13:18.420 So if you just had some differences in observability 00:13:18.420 --> 00:13:21.165 in effort, how hard people work, we 00:13:21.165 --> 00:13:25.260 could just disentangle these two explanations. 00:13:25.260 --> 00:13:27.435 So Bandiera et al. have this very nice variation. 00:13:27.435 --> 00:13:29.760 They have two fruits there, fruit type 00:13:29.760 --> 00:13:31.770 1, which are strawberries, and they 00:13:31.770 --> 00:13:34.620 have fruit of type 2, which are raspberries. 00:13:34.620 --> 00:13:36.870 If you ever have picked strawberries, 00:13:36.870 --> 00:13:39.270 you would know that strawberries is very easy. 00:13:39.270 --> 00:13:43.800 These are essentially very flat and low fruit 00:13:43.800 --> 00:13:45.810 that when somebody picks strawberries, 00:13:45.810 --> 00:13:48.330 it's very easy to see how fast somebody works, 00:13:48.330 --> 00:13:51.300 how much anybody picks and so on because you can just 00:13:51.300 --> 00:13:52.650 see across the entire field. 00:13:52.650 --> 00:13:56.010 You can see essentially what all of your coworkers are doing. 00:13:56.010 --> 00:14:00.930 In contrast, if you look at raspberries, these are bushes. 00:14:00.930 --> 00:14:04.860 So you can essentially literally hide behind the bushes and work 00:14:04.860 --> 00:14:08.250 really hard to pick really hard and fast 00:14:08.250 --> 00:14:09.870 without others noticing. 00:14:09.870 --> 00:14:13.440 So essentially, productivity is at least more unobserved 00:14:13.440 --> 00:14:15.780 than for strawberries. 00:14:15.780 --> 00:14:17.380 Now, what Bandiera et al. then find, 00:14:17.380 --> 00:14:21.750 in fact, is no impact of the piece rate on fruits of type 2, 00:14:21.750 --> 00:14:23.770 which is raspberries. 00:14:23.770 --> 00:14:27.960 So that suggests there is no evidence of pure altruism, 00:14:27.960 --> 00:14:32.340 and the effects, perhaps, could be driven by reciprocity. 00:14:32.340 --> 00:14:34.980 This is consistent with what we had found before. 00:14:34.980 --> 00:14:37.260 When you remember the dictator ultimatum games 00:14:37.260 --> 00:14:40.740 that we showed you, it looked a lot like people are altruistic. 00:14:40.740 --> 00:14:42.060 They care about others. 00:14:42.060 --> 00:14:44.820 But when you look at their actual motives, 00:14:44.820 --> 00:14:46.830 when you dig a little deeper, we find out that, 00:14:46.830 --> 00:14:50.640 in fact, this is driven by reciprocity, or by worries 00:14:50.640 --> 00:14:53.940 about repeated game effects, or worries about essentially 00:14:53.940 --> 00:14:57.450 just retribution, either on the field that other people will 00:14:57.450 --> 00:14:59.790 work also really hard if I work really hard right 00:14:59.790 --> 00:15:03.360 now, or socially after work, my coworkers 00:15:03.360 --> 00:15:04.635 might be really mean to me. 00:15:04.635 --> 00:15:06.510 They might even beat me up, or they might not 00:15:06.510 --> 00:15:10.260 be my friends anymore. 00:15:10.260 --> 00:15:12.840 Here is regression tables that show you this. 00:15:12.840 --> 00:15:16.350 The first column, you see fruit type 2. 00:15:16.350 --> 00:15:17.820 Again, these are the raspberries. 00:15:17.820 --> 00:15:20.700 This is when things are unobserved. 00:15:20.700 --> 00:15:23.130 There's no effective introducing the piece rate. 00:15:23.130 --> 00:15:27.120 In contrast, if you look at column 2, this is fruit type 1. 00:15:27.120 --> 00:15:29.850 These are strawberries, and there's large productivity 00:15:29.850 --> 00:15:33.600 effects of strawberries there. 00:15:33.600 --> 00:15:36.120 So these results, again, highlight the importance 00:15:36.120 --> 00:15:39.060 of setting the incentives carefully. 00:15:39.060 --> 00:15:40.800 Small details matter, and you really 00:15:40.800 --> 00:15:42.660 want to be very careful in how you 00:15:42.660 --> 00:15:45.720 said relative incentives that in principle seemed 00:15:45.720 --> 00:15:47.520 like a good idea because they provide 00:15:47.520 --> 00:15:51.900 steep incentives to workers, but in practice might actually not 00:15:51.900 --> 00:15:54.300 work particularly well. 00:15:57.490 --> 00:15:59.530 OK, so the second paper we're going to look at 00:15:59.530 --> 00:16:00.370 is Breza et al. 00:16:00.370 --> 00:16:03.190 These are the morale effects of pay inequality. 00:16:03.190 --> 00:16:06.790 This is a very nice randomized field experiment in rural India 00:16:06.790 --> 00:16:09.520 with low-skill manufacturing workers. 00:16:09.520 --> 00:16:12.700 The question the authors ask is do workers 00:16:12.700 --> 00:16:14.920 care about relative pay? 00:16:14.920 --> 00:16:19.720 And perhaps why do we see so little variation in pay 00:16:19.720 --> 00:16:20.440 across settings? 00:16:20.440 --> 00:16:23.350 So what we often see in many companies in India, 00:16:23.350 --> 00:16:27.940 but also in other places, also in certain villages in India, 00:16:27.940 --> 00:16:31.070 what we see is wages tend to be very compressed. 00:16:31.070 --> 00:16:34.480 So workers essentially earn the exact same wage, 00:16:34.480 --> 00:16:38.650 regardless of their underlying productivity, experience, 00:16:38.650 --> 00:16:39.520 and so on. 00:16:39.520 --> 00:16:43.000 Good and bad workers tend to earn 00:16:43.000 --> 00:16:47.140 exactly the same amount of money for per day 00:16:47.140 --> 00:16:49.270 or per hour and the like. 00:16:49.270 --> 00:16:50.890 Economists think that's inefficient 00:16:50.890 --> 00:16:52.450 because if you want to hire somebody 00:16:52.450 --> 00:16:53.950 who's really, really productive, you 00:16:53.950 --> 00:16:55.930 might want to pay them more. 00:16:55.930 --> 00:16:58.330 If you want to hire somebody who is not that productive, 00:16:58.330 --> 00:16:59.670 you might want to pay them less. 00:16:59.670 --> 00:17:02.170 You might actually want to pay them less than the prevailing 00:17:02.170 --> 00:17:04.270 wage, but you might not be able to do 00:17:04.270 --> 00:17:07.970 so because the norm is that you have to pay them the prevailing 00:17:07.970 --> 00:17:08.470 wage. 00:17:08.470 --> 00:17:10.960 And then you might end up not hiring that person, 00:17:10.960 --> 00:17:13.329 and that person might end up being unemployed, 00:17:13.329 --> 00:17:16.599 even though you would be very happy to pay them 00:17:16.599 --> 00:17:18.280 at a lower wage. 00:17:18.280 --> 00:17:20.780 This is a really relevant question. 00:17:20.780 --> 00:17:22.569 Sorry, the second question the authors ask 00:17:22.569 --> 00:17:26.260 is what is the notion of the underlying notion of fairness? 00:17:26.260 --> 00:17:28.900 What is fair and what's not fair? 00:17:28.900 --> 00:17:32.860 As in if there is wage dispersion across workers 00:17:32.860 --> 00:17:37.060 within teams or within a certain firm, under some circumstances 00:17:37.060 --> 00:17:38.385 under which that's OK to do? 00:17:38.385 --> 00:17:39.760 Are there some good justification 00:17:39.760 --> 00:17:47.380 that workers are OK with, or is it just always bad to do so? 00:17:47.380 --> 00:17:51.340 So can we find some situation under which it's justified 00:17:51.340 --> 00:17:54.730 for workers or not, or is it just like we cannot pay have 00:17:54.730 --> 00:17:57.760 pay inequality under any circumstances in those kinds 00:17:57.760 --> 00:17:59.000 of firms? 00:17:59.000 --> 00:18:00.580 This is potentially quite relevant 00:18:00.580 --> 00:18:03.170 for many features of the labor market. 00:18:03.170 --> 00:18:05.170 It might explain wage compression. 00:18:05.170 --> 00:18:06.670 That's what I'm saying, like wages 00:18:06.670 --> 00:18:09.460 tend to be fairly compressed, even if productivity is fairly 00:18:09.460 --> 00:18:11.500 dispersed across workers. 00:18:11.500 --> 00:18:13.450 There tends to be lots of wage rigidity. 00:18:13.450 --> 00:18:17.170 Wages tend to not move very much. 00:18:17.170 --> 00:18:20.500 Employers have trouble adjusting wages, 00:18:20.500 --> 00:18:23.080 or they are reluctant to adjust people's wages much, 00:18:23.080 --> 00:18:25.520 in particular when it comes to like downward rigidity. 00:18:25.520 --> 00:18:26.895 We talked about this a little bit 00:18:26.895 --> 00:18:31.120 already when we talked about reference dependence. 00:18:31.120 --> 00:18:33.460 This can help us think about like sorting of workers 00:18:33.460 --> 00:18:35.830 into firms and inequality. 00:18:35.830 --> 00:18:37.950 We can think about firm boundaries in a sense 00:18:37.950 --> 00:18:41.680 and say should they be really large firms 00:18:41.680 --> 00:18:43.420 or a number of small firms? 00:18:43.420 --> 00:18:48.070 If you think the relative comparison is across workers 00:18:48.070 --> 00:18:50.380 within firms, that would say, well, we should probably 00:18:50.380 --> 00:18:52.233 have a lot of small firms. 00:18:52.233 --> 00:18:53.650 And then each of these small firms 00:18:53.650 --> 00:18:55.510 could have different wages, as opposed 00:18:55.510 --> 00:18:58.360 to one large firm where essentially everybody has to be 00:18:58.360 --> 00:19:00.377 paid more or less the same. 00:19:00.377 --> 00:19:02.710 And we might think that's inefficient because as returns 00:19:02.710 --> 00:19:05.500 to scale of having a large firm. 00:19:05.500 --> 00:19:08.350 Think about it is also relevant for some HR policies 00:19:08.350 --> 00:19:10.810 in terms of how do you pay workers, 00:19:10.810 --> 00:19:13.130 how do you set wages, and so on. 00:19:13.130 --> 00:19:15.970 So as motivation in the study, the authors 00:19:15.970 --> 00:19:19.460 ask people the following questions. 00:19:19.460 --> 00:19:21.850 The questions are the following. 00:19:21.850 --> 00:19:29.080 This is rural India in Rissa, a relatively poor area with lots 00:19:29.080 --> 00:19:31.600 of small-scale manufacturing. 00:19:31.600 --> 00:19:33.430 The question is three people from a village 00:19:33.430 --> 00:19:36.760 get hired to work on a construction site together. 00:19:36.760 --> 00:19:39.940 The prevailing wage is 250 rupees. 00:19:39.940 --> 00:19:42.010 That's the standard wage in the village. 00:19:42.010 --> 00:19:43.960 Each village tends to have like a prevailing 00:19:43.960 --> 00:19:46.780 wage, which is how much people are paid usually. 00:19:46.780 --> 00:19:51.460 The contractor pays them 250 rupees per day. 00:19:51.460 --> 00:19:53.920 How well will they work together? 00:19:53.920 --> 00:19:56.050 And what you see is what respondents tend to say 00:19:56.050 --> 00:20:01.810 is 80% of respondents say people worked very well together. 00:20:01.810 --> 00:20:06.650 20% say as well as usual, and 0% say there will be conflict. 00:20:06.650 --> 00:20:08.620 So that's seems like paying everybody 00:20:08.620 --> 00:20:11.540 the same is the socially acceptable thing to do. 00:20:11.540 --> 00:20:14.260 Everybody seems happy with that. 00:20:14.260 --> 00:20:16.430 In contrast, if you ask the following, 00:20:16.430 --> 00:20:19.420 which is the contractor pays them different wages. 00:20:19.420 --> 00:20:23.890 If the contractor pays 250 per day, 270 per day, 00:20:23.890 --> 00:20:28.790 and 290 per day, how well will they work together? 00:20:28.790 --> 00:20:32.260 And so now, what you see here is when 00:20:32.260 --> 00:20:37.180 there is differences across pay, 94% of workers 00:20:37.180 --> 00:20:40.893 say there will be conflict. 00:20:40.893 --> 00:20:42.560 Listen, in this version of the question, 00:20:42.560 --> 00:20:45.850 it doesn't specify whether that's due with underlying 00:20:45.850 --> 00:20:47.000 productivity differences. 00:20:47.000 --> 00:20:50.060 So why is it that some workers are paid different than others? 00:20:50.060 --> 00:20:52.690 However, it's very strong suggestive evidence 00:20:52.690 --> 00:20:55.720 that workers might be really unhappy, 00:20:55.720 --> 00:20:59.320 and there will be conflict across workers 00:20:59.320 --> 00:21:03.340 if they're paid different wages in the same kind of team, 00:21:03.340 --> 00:21:06.700 or at least on the same construction site. 00:21:06.700 --> 00:21:08.840 And then, of course, if you're an employer, 00:21:08.840 --> 00:21:13.240 you might be quite worried about paying people different wages 00:21:13.240 --> 00:21:14.410 because of this conflict. 00:21:14.410 --> 00:21:17.770 Presumably, workers will not work or collaborate as well 00:21:17.770 --> 00:21:20.500 as they could potentially. 00:21:20.500 --> 00:21:23.920 OK, so now the authors set up a very nice field experiment 00:21:23.920 --> 00:21:25.070 that looks like this. 00:21:25.070 --> 00:21:30.850 So there's 10 production units of 3 workers in each factory. 00:21:30.850 --> 00:21:35.470 So think of factories like work sites, factories in quotation 00:21:35.470 --> 00:21:37.360 marks because it's really not just a factory 00:21:37.360 --> 00:21:38.860 as you might imagine it. 00:21:38.860 --> 00:21:41.410 It's really like a small work site, an office where people 00:21:41.410 --> 00:21:43.720 are working in different areas. 00:21:43.720 --> 00:21:45.670 There's 10 units of 3 workers each, 00:21:45.670 --> 00:21:47.350 so it's not a huge office. 00:21:47.350 --> 00:21:51.920 Each unit of three workers produces different products. 00:21:51.920 --> 00:21:54.700 So each unit has one product. 00:21:54.700 --> 00:21:56.650 For example, unit 1 makes brooms. 00:21:56.650 --> 00:21:59.140 Unit 2 makes incense sticks. 00:21:59.140 --> 00:22:03.000 Unit 3 makes leaf plates, and so on. 00:22:03.000 --> 00:22:06.442 So each unit has one item of production. 00:22:06.442 --> 00:22:07.900 Now, why are there different units? 00:22:07.900 --> 00:22:09.430 It's essentially to separate workers 00:22:09.430 --> 00:22:11.290 into different lines of work. 00:22:11.290 --> 00:22:13.390 So what we're going to expect is that workers 00:22:13.390 --> 00:22:15.700 are going to compare each other within unit, 00:22:15.700 --> 00:22:20.200 but not so much across units because if I make brooms 00:22:20.200 --> 00:22:22.040 and you make incense sticks, that's 00:22:22.040 --> 00:22:24.088 very different work in some ways. 00:22:24.088 --> 00:22:25.630 But if you make the exact same thing, 00:22:25.630 --> 00:22:27.910 the natural comparison now for workers 00:22:27.910 --> 00:22:30.640 to compare yourself with another worker. 00:22:30.640 --> 00:22:34.180 All unit members, so everybody in a given unit, 00:22:34.180 --> 00:22:35.930 makes the exact same product. 00:22:35.930 --> 00:22:40.810 So everybody in unit 1, who is three workers, makes brooms. 00:22:40.810 --> 00:22:44.390 Everybody in unit 2 makes incense sticks, and so on. 00:22:44.390 --> 00:22:47.980 Now the key experimental variation in the experiment 00:22:47.980 --> 00:22:52.900 is weight dispersion across workers within teams. 00:22:52.900 --> 00:22:56.290 Finally, there's also-- and I'll get back to that at the end-- 00:22:56.290 --> 00:23:02.000 production tasks vary in their observability of performance. 00:23:02.000 --> 00:23:04.300 So that is to say for some tasks it just 00:23:04.300 --> 00:23:07.240 happens to be that it's much easier 00:23:07.240 --> 00:23:10.150 to understand how productive your co-worker is than 00:23:10.150 --> 00:23:12.430 in other tasks. 00:23:12.430 --> 00:23:17.050 OK, so now what is the variation in relative pay. 00:23:17.050 --> 00:23:23.920 So here you see people are paid depending on their worker rank. 00:23:23.920 --> 00:23:26.260 There's four different regimes of pay 00:23:26.260 --> 00:23:28.990 in the different columns, heterogeneous, 00:23:28.990 --> 00:23:32.950 compressed L, compressed M, and compressed H. Workers 00:23:32.950 --> 00:23:33.753 are sorted. 00:23:33.753 --> 00:23:35.170 So there's a baseline period where 00:23:35.170 --> 00:23:38.380 workers who work for a few days, and the few days 00:23:38.380 --> 00:23:41.260 are used to assess their baseline productivity. 00:23:41.260 --> 00:23:44.560 They're ranked into three turnstyles, 00:23:44.560 --> 00:23:46.420 so there's like the low productive workers, 00:23:46.420 --> 00:23:49.060 median productivity, and high productivity workers. 00:23:49.060 --> 00:23:51.160 They're ranked essentially based on how well they 00:23:51.160 --> 00:23:52.570 did at baseline. 00:23:52.570 --> 00:23:54.755 OK, so if you're like really bad at the start, 00:23:54.755 --> 00:23:56.380 you would be a low productivity worker. 00:23:56.380 --> 00:23:57.760 If you're like an average worker, 00:23:57.760 --> 00:23:59.325 you would be median productivity. 00:23:59.325 --> 00:24:01.450 And if you're really like highly productive worker, 00:24:01.450 --> 00:24:06.280 you would be classified as high productivity workers. 00:24:06.280 --> 00:24:08.780 Notice that's always done for all of the different treatment 00:24:08.780 --> 00:24:09.530 groups. 00:24:09.530 --> 00:24:13.640 The difference now is across these different regimes 00:24:13.640 --> 00:24:15.530 how workers are paid. 00:24:15.530 --> 00:24:19.280 In their heterogeneous pay essentially, there's 00:24:19.280 --> 00:24:22.790 differences in cross workers in the sense 00:24:22.790 --> 00:24:27.830 of like the low productivity workers are paid low wage, 00:24:27.830 --> 00:24:29.570 the median productivity paid workers 00:24:29.570 --> 00:24:32.990 are paid the median weight, and the high productivity workers 00:24:32.990 --> 00:24:34.178 are paid the high wage. 00:24:34.178 --> 00:24:36.470 As you might expect, that's, sort of, the natural thing 00:24:36.470 --> 00:24:39.590 to do in terms if wanted more productivity, the most 00:24:39.590 --> 00:24:43.220 productive workers you're going to pay the most. 00:24:43.220 --> 00:24:46.130 Notice that these wage differences are modest. 00:24:46.130 --> 00:24:48.140 Even like the difference between W 00:24:48.140 --> 00:24:54.472 high and W low are like only about like up to 10%. 00:24:54.472 --> 00:24:56.180 So really these are not huge differences. 00:24:56.180 --> 00:24:58.680 It's not that the other guy if you are like low productivity 00:24:58.680 --> 00:25:01.870 worker if you get W low, it's not the other guy 00:25:01.870 --> 00:25:03.120 gets like twice as much. 00:25:03.120 --> 00:25:06.350 It really is like modestly more, but it is more money 00:25:06.350 --> 00:25:07.580 that they get. 00:25:07.580 --> 00:25:10.790 Now then if you look at the other three regimes, compressed 00:25:10.790 --> 00:25:15.140 L, compressed M, and compressed H, low productivity-- so 00:25:15.140 --> 00:25:18.980 in compressed L, everybody gets to low wage, in compressed M, 00:25:18.980 --> 00:25:22.040 everybody gets the minimum wage, and in compressed H, 00:25:22.040 --> 00:25:24.260 everybody gets the high wage. 00:25:24.260 --> 00:25:26.510 And now the study now lets the authors 00:25:26.510 --> 00:25:31.100 compare the different columns for holding exactly 00:25:31.100 --> 00:25:34.820 constant baseline productivity and wage levels. 00:25:34.820 --> 00:25:39.800 That is to say, for instance, we can compare the heterogeneous-- 00:25:39.800 --> 00:25:41.810 the low productivity workers that 00:25:41.810 --> 00:25:44.600 happen to be randomized into like the heterogeneous 00:25:44.600 --> 00:25:49.130 treatment, they receive W low or compare them 00:25:49.130 --> 00:25:51.530 with like low productivity workers at baseline who 00:25:51.530 --> 00:25:54.530 happen to be randomized into a compressed L. 00:25:54.530 --> 00:25:56.240 So I should have said more clearly 00:25:56.240 --> 00:25:59.750 people are randomized into any of these four groups, 00:25:59.750 --> 00:26:02.570 and then depending on what your baseline productivity is 00:26:02.570 --> 00:26:05.970 you get the wages as I show here in the table. 00:26:05.970 --> 00:26:08.780 So for example, if you are like a low productivity worker, 00:26:08.780 --> 00:26:13.670 you might be randomized into the heterogeneous treatment group. 00:26:13.670 --> 00:26:16.130 You might receive W low, or you might 00:26:16.130 --> 00:26:18.260 be randomized into the low-- 00:26:18.260 --> 00:26:23.150 compressed, low treatment where you also receive W low. 00:26:23.150 --> 00:26:25.760 Notice that in both cases, that worker 00:26:25.760 --> 00:26:28.100 is a low productivity worker. 00:26:28.100 --> 00:26:31.430 In both cases, the worker receives W low. 00:26:31.430 --> 00:26:34.850 However, what's different here now is that his co-worker-- 00:26:34.850 --> 00:26:36.230 his or her-- 00:26:36.230 --> 00:26:38.060 in this case, these are all men-- 00:26:38.060 --> 00:26:41.210 co-workers are receiving either the same, which 00:26:41.210 --> 00:26:44.900 is in compressed L, or they receive higher wages 00:26:44.900 --> 00:26:49.020 in the heterogeneous weight treatment, right. 00:26:49.020 --> 00:26:50.720 So now we can look at workers who 00:26:50.720 --> 00:26:53.000 have the exact same baseline productivity on average 00:26:53.000 --> 00:26:56.330 at least and have the exact same weight, but what's being varied 00:26:56.330 --> 00:26:58.700 is like how much other workers are earning. 00:26:58.700 --> 00:27:01.698 We can do this for low productivity workers. 00:27:01.698 --> 00:27:03.740 We can also do it for median productivity workers 00:27:03.740 --> 00:27:05.667 or also for high productivity workers. 00:27:05.667 --> 00:27:07.250 I skip the median productivity worker, 00:27:07.250 --> 00:27:08.810 but that's exactly the same. 00:27:08.810 --> 00:27:10.220 For high productivity workers, we 00:27:10.220 --> 00:27:15.800 can look at the workers who had been randomized 00:27:15.800 --> 00:27:22.340 into the heterogeneous treatment group or heterogeneous worker 00:27:22.340 --> 00:27:23.870 group-- 00:27:23.870 --> 00:27:26.870 sorry, heterogeneous wage group where 00:27:26.870 --> 00:27:29.150 the worker gets a WH but everybody 00:27:29.150 --> 00:27:31.460 else gets like a lower wage, or we 00:27:31.460 --> 00:27:35.120 can compare that to compressed age where everybody 00:27:35.120 --> 00:27:37.340 gets WH in that group. 00:27:43.510 --> 00:27:45.820 OK, so now what do you what do the authors find? 00:27:45.820 --> 00:27:48.820 Let's start with the low productivity 00:27:48.820 --> 00:27:50.170 workers at baseline. 00:27:50.170 --> 00:27:52.120 Remember the collinear comparison here 00:27:52.120 --> 00:27:56.200 is between pay disparity-- this is a heterogeneous group-- pay 00:27:56.200 --> 00:27:58.330 disparity and compressed L, which 00:27:58.330 --> 00:28:02.710 is like the group where everybody is paid the same. 00:28:02.710 --> 00:28:05.500 Now what the authors find-- and you see this in the graph 00:28:05.500 --> 00:28:09.272 fairly nicely, which is like the productivity-- 00:28:09.272 --> 00:28:11.230 these are like the red line and the blue line-- 00:28:11.230 --> 00:28:12.910 the productivity on the left side- 00:28:12.910 --> 00:28:17.440 this is before the treatment starts for about over 10 days-- 00:28:17.440 --> 00:28:20.180 the productivity looks pretty much the same. 00:28:20.180 --> 00:28:22.390 But then when you go to the right side of the graph, 00:28:22.390 --> 00:28:26.350 not immediately but after a few days they seem to be a gap 00:28:26.350 --> 00:28:30.310 or a gap emerges between the red line and the blue line, which 00:28:30.310 --> 00:28:36.160 is exactly the gap as you would expect if the disparity makes 00:28:36.160 --> 00:28:40.120 workers less productive, which is to say the workers who 00:28:40.120 --> 00:28:43.000 receive a low wage but others in that group 00:28:43.000 --> 00:28:46.510 are receiving higher wages are becoming less productive 00:28:46.510 --> 00:28:50.830 compared to workers that receive a low wage where others have 00:28:50.830 --> 00:28:53.530 the exact same productivity. 00:28:53.530 --> 00:28:57.340 And you see this also in the regression tables, 00:28:57.340 --> 00:29:01.690 you see about a 22% reduction in mean output 00:29:01.690 --> 00:29:05.903 and a 9% reduction in earnings, which are pretty large effects. 00:29:05.903 --> 00:29:07.570 So these are like large effects compared 00:29:07.570 --> 00:29:10.810 to like other interventions that people have tried. 00:29:10.810 --> 00:29:12.550 Interesting, you see a little bit 00:29:12.550 --> 00:29:15.820 like while the treatment effect when you look at the graph 00:29:15.820 --> 00:29:19.360 initially looks not particularly large, if anything it might not 00:29:19.360 --> 00:29:22.660 even be there often like a few days, that treatment effect 00:29:22.660 --> 00:29:23.890 increases over time. 00:29:23.890 --> 00:29:27.610 It becomes larger so over time people 00:29:27.610 --> 00:29:33.550 become less and less productive compared to the compressed wage 00:29:33.550 --> 00:29:34.690 treatment. 00:29:34.690 --> 00:29:36.940 Interestingly, we find-- the authors 00:29:36.940 --> 00:29:40.720 find similar effects for high ranked, high productivity 00:29:40.720 --> 00:29:41.740 workers. 00:29:41.740 --> 00:29:43.870 So notice that these are not workers 00:29:43.870 --> 00:29:46.540 who either who are high productivity workers 00:29:46.540 --> 00:29:49.600 to receive the same wage in the pay disparity treatment 00:29:49.600 --> 00:29:53.200 and the compressed H treatment that have the same wage. 00:29:53.200 --> 00:29:56.140 In one case, they are paid exactly the same. 00:29:56.140 --> 00:29:58.630 In the other case, they're paid more 00:29:58.630 --> 00:30:00.910 compared to their coworkers. 00:30:00.910 --> 00:30:03.190 And what happens what seems to be the case 00:30:03.190 --> 00:30:08.410 is that the compressed high paid workers are, in fact, more 00:30:08.410 --> 00:30:10.420 productive compared to others. 00:30:10.420 --> 00:30:13.498 You might have expected in some ways like if one guy gets paid 00:30:13.498 --> 00:30:15.040 high payment but then others get paid 00:30:15.040 --> 00:30:17.802 less, that makes the high pay worker more productive, 00:30:17.802 --> 00:30:19.510 because maybe he feels good about himself 00:30:19.510 --> 00:30:20.800 that he's a high paid workers. 00:30:20.800 --> 00:30:23.710 And maybe he feels he can prove himself or the like, 00:30:23.710 --> 00:30:26.050 but instead what seems to be the case the group that 00:30:26.050 --> 00:30:31.270 works in his team that does not work well together or they just 00:30:31.270 --> 00:30:34.030 becomes uncomfortable to work with somebody else who is like 00:30:34.030 --> 00:30:37.500 really mad at you for earning more than they do. 00:30:37.500 --> 00:30:41.950 And so now the high pay workers in the compressed treatment, 00:30:41.950 --> 00:30:44.140 in fact, earn more or produce more 00:30:44.140 --> 00:30:49.160 than the high paid workers in the pay disparity treatment. 00:30:49.160 --> 00:30:52.510 So that is to say the pay disparity, 00:30:52.510 --> 00:30:55.630 pay inequality does not only reduce 00:30:55.630 --> 00:30:58.360 worker's productivity for the low pay workers, 00:30:58.360 --> 00:31:00.400 so not only the people-- 00:31:00.400 --> 00:31:04.533 the workers who are earning less compared to their co-workers, 00:31:04.533 --> 00:31:06.200 and that's in some sense to be expected. 00:31:06.200 --> 00:31:08.492 You might just be mad at everybody else or other people 00:31:08.492 --> 00:31:11.350 in your group are earning more. 00:31:11.350 --> 00:31:13.190 Then you might be unhappy and mad about that 00:31:13.190 --> 00:31:14.950 and just then produce less. 00:31:14.950 --> 00:31:17.890 Instead it seems to be the case even the high productivity 00:31:17.890 --> 00:31:21.130 workers, the workers who earn more than others in their group 00:31:21.130 --> 00:31:25.330 are producing less compared to the control group where 00:31:25.330 --> 00:31:27.890 everybody earns the same. 00:31:27.890 --> 00:31:29.320 So what did we learn from that? 00:31:29.320 --> 00:31:33.070 Well, pay disparity lowers worker performance for all team 00:31:33.070 --> 00:31:36.790 members, and so the interpretation of that 00:31:36.790 --> 00:31:42.070 is that pay disparity undermines workers ability to cooperate 00:31:42.070 --> 00:31:46.000 in their own self-interest. 00:31:46.000 --> 00:31:47.740 The paper has some additional evidence 00:31:47.740 --> 00:31:50.050 with some cooperative tasks where 00:31:50.050 --> 00:31:53.060 essentially workers are worse at cooperating with each other. 00:31:53.060 --> 00:31:55.480 It seems to really be that workers are not 00:31:55.480 --> 00:31:57.760 happy to work in the same place with somebody 00:31:57.760 --> 00:32:00.550 else who earns less. 00:32:00.550 --> 00:32:03.100 Importantly, the perceived justification 00:32:03.100 --> 00:32:05.800 is essential in mediating these effects. 00:32:05.800 --> 00:32:08.590 That is to say I told you previously that for some task 00:32:08.590 --> 00:32:12.680 it was easier to see who was more productive than others, 00:32:12.680 --> 00:32:15.400 especially at baseline just because in the task 00:32:15.400 --> 00:32:19.570 the difference is across highly and less highly productive. 00:32:19.570 --> 00:32:22.480 Or high and low productivity workers was just wider, 00:32:22.480 --> 00:32:25.270 and it's just easy to see that Frank is really bad at this 00:32:25.270 --> 00:32:26.800 and somebody else is really good. 00:32:26.800 --> 00:32:29.320 And for some task, it's really easy to see that, 00:32:29.320 --> 00:32:30.800 for others not. 00:32:30.800 --> 00:32:32.410 So if it's really easy-- if there's 00:32:32.410 --> 00:32:34.397 a perceived justification for workers, 00:32:34.397 --> 00:32:35.980 if workers, sort of, thought, OK, it's 00:32:35.980 --> 00:32:39.550 really easy to see who is more productive than others, 00:32:39.550 --> 00:32:42.250 the effects are much weaker than when there 00:32:42.250 --> 00:32:44.440 is no receive justification. 00:32:44.440 --> 00:32:46.473 That is to say workers themselves are saying, 00:32:46.473 --> 00:32:47.890 like, if I'm saying, look, there's 00:32:47.890 --> 00:32:50.000 this other worker who is really productive 00:32:50.000 --> 00:32:51.740 and if I'm sort of saying, hey, look, 00:32:51.740 --> 00:32:53.198 this is really obvious that they're 00:32:53.198 --> 00:32:54.910 more productive than I am, it's only fair 00:32:54.910 --> 00:32:57.910 that they're earning more, then the affects 00:32:57.910 --> 00:33:01.190 are much damage-- then there's much less than an effect that's 00:33:01.190 --> 00:33:04.120 accepted, and there's much less of a productivity 00:33:04.120 --> 00:33:08.048 effect, if any, if there's pay inequality. 00:33:08.048 --> 00:33:10.090 However, if it seems to be the case that they're, 00:33:10.090 --> 00:33:12.550 kind of, like producing the same thing, 00:33:12.550 --> 00:33:16.150 if we like similarly producing the same thing 00:33:16.150 --> 00:33:19.270 and we're equally good at it or approximately equally good 00:33:19.270 --> 00:33:22.610 and one person is paid more than I am for no good reason, 00:33:22.610 --> 00:33:26.200 so it appears that workers are really not happy about that, 00:33:26.200 --> 00:33:33.160 then work performance, in particular output falls. 00:33:33.160 --> 00:33:34.430 So what are the implications? 00:33:34.430 --> 00:33:39.370 Well, this evidence suggests that weight compression 00:33:39.370 --> 00:33:42.490 may be more likely in some settings than in others, right. 00:33:42.490 --> 00:33:45.220 That's to say like if it's easy to justify, 00:33:45.220 --> 00:33:48.280 if the production function or like the production process is 00:33:48.280 --> 00:33:52.030 such that all aspects of the production and the output 00:33:52.030 --> 00:33:55.750 or the performance are easy, measurable, and observable 00:33:55.750 --> 00:33:58.720 for workers, then workers might say 00:33:58.720 --> 00:34:02.510 it's OK if some workers are earning more than others. 00:34:02.510 --> 00:34:05.740 However, if that's not the case, if there's a bunch of stuff 00:34:05.740 --> 00:34:08.020 in the work process that's really not observed-- 00:34:08.020 --> 00:34:11.170 maybe it's verbal or maybe just really hard to justify or hard 00:34:11.170 --> 00:34:14.199 to sort of assess in some tangible way-- 00:34:14.199 --> 00:34:16.750 then workers might be really unhappy 00:34:16.750 --> 00:34:19.060 and that might lead to weight compression, 00:34:19.060 --> 00:34:23.110 because the employer might anticipate that. 00:34:23.110 --> 00:34:28.239 So overall, this, sort of, says that like relative comparisons 00:34:28.239 --> 00:34:29.500 matter quite a bit. 00:34:29.500 --> 00:34:32.860 Fairness matters quite a bit, and what's 00:34:32.860 --> 00:34:34.909 really key for an employer and anybody who 00:34:34.909 --> 00:34:39.130 sets incentives is to take into account these fairness 00:34:39.130 --> 00:34:43.090 considerations, and one wants to really understand what is fair 00:34:43.090 --> 00:34:44.300 and what is not. 00:34:44.300 --> 00:34:48.460 And upsetting workers by violating these fairness 00:34:48.460 --> 00:34:52.030 considerations or norms might be really, really costly 00:34:52.030 --> 00:34:55.060 for an employer. 00:34:55.060 --> 00:34:58.980 So the good news here is that if one understands that well, one 00:34:58.980 --> 00:35:01.890 can really sort of produce increased productivity quite 00:35:01.890 --> 00:35:06.660 a bit, and in some cases weight inequalities actually fine. 00:35:06.660 --> 00:35:09.600 One just has to be careful in figuring out when exactly is 00:35:09.600 --> 00:35:12.790 that the case, OK. 00:35:12.790 --> 00:35:14.290 As I said, there's a very nice paper 00:35:14.290 --> 00:35:17.350 by Jonas Hjort on ethnic divisions 00:35:17.350 --> 00:35:18.760 and production in firms. 00:35:18.760 --> 00:35:20.740 You're going to talk about that in recitation. 00:35:23.960 --> 00:35:26.880 OK, the second part of this lecture 00:35:26.880 --> 00:35:31.620 we'll talk about policies to increase pro-sociality. 00:35:31.620 --> 00:35:34.290 And so the first of these papers is 00:35:34.290 --> 00:35:37.320 very nice recent paper by Gautam Rao that looks at the question 00:35:37.320 --> 00:35:39.660 are social preferences malleable. 00:35:39.660 --> 00:35:41.920 Like, what are the origins of social preferences? 00:35:41.920 --> 00:35:46.930 Why is it that some people appear nicer than others? 00:35:46.930 --> 00:35:48.660 And then once we understand that, 00:35:48.660 --> 00:35:51.960 perhaps, we can also understand what policies, if anything, 00:35:51.960 --> 00:35:53.837 can affect social preferences. 00:35:56.460 --> 00:35:58.350 The main question that this paper asks 00:35:58.350 --> 00:36:01.560 is how does being mixed with poor students in school 00:36:01.560 --> 00:36:05.040 affect the social preferences of rich students. 00:36:05.040 --> 00:36:08.790 Now in a lot of the, kind of, policies 00:36:08.790 --> 00:36:11.430 where poor students are mixed with rich students, 00:36:11.430 --> 00:36:13.320 you might ask the question, how does 00:36:13.320 --> 00:36:14.500 it affect the poor student. 00:36:14.500 --> 00:36:16.410 If a poor student for whatever reason 00:36:16.410 --> 00:36:20.130 would not be able to afford like a rich school or rich student's 00:36:20.130 --> 00:36:22.320 school and you, sort of, allow them 00:36:22.320 --> 00:36:26.610 by giving them scholarships or other policies, 00:36:26.610 --> 00:36:28.110 often the question that people ask 00:36:28.110 --> 00:36:30.450 is like what are the benefits of doing that. 00:36:30.450 --> 00:36:34.035 Is the poor student doing better in school, 00:36:34.035 --> 00:36:36.660 or do they have different types of friends, different networks, 00:36:36.660 --> 00:36:39.520 and does it lead to better jobs and so on and so forth? 00:36:39.520 --> 00:36:41.790 This question, this paper ask a different question. 00:36:41.790 --> 00:36:45.270 It asks the question about what is the effect of rich students 00:36:45.270 --> 00:36:47.100 if they randomly or quasi randomly 00:36:47.100 --> 00:36:52.200 are exposed to being in class with additional or some more 00:36:52.200 --> 00:36:54.150 poor students. 00:36:54.150 --> 00:36:56.490 The paper exploits a policy change 00:36:56.490 --> 00:37:01.800 that introduced an admissions quota of 20% for poor students 00:37:01.800 --> 00:37:04.000 in primary schools in Delhi. 00:37:04.000 --> 00:37:08.280 In Delhi, these are rich primary schools. 00:37:08.280 --> 00:37:11.410 The paper looks at two sources of variation. 00:37:11.410 --> 00:37:14.100 There's variation across classrooms 00:37:14.100 --> 00:37:17.400 that allows the author to look at the overall effects. 00:37:17.400 --> 00:37:21.370 So you can look at, like, within schools there's 00:37:21.370 --> 00:37:26.650 going to be treated and controlled cohorts, 00:37:26.650 --> 00:37:30.010 because the policy was introduced at some point. 00:37:30.010 --> 00:37:33.280 So for some students, they enjoyed 00:37:33.280 --> 00:37:36.760 the benefits of being in class with poor students 00:37:36.760 --> 00:37:38.050 or the costs. 00:37:38.050 --> 00:37:39.550 We'll see about that. 00:37:39.550 --> 00:37:42.130 And for other schools, they're already too far advanced. 00:37:42.130 --> 00:37:44.230 They were like essentially before the-- 00:37:44.230 --> 00:37:46.780 they went to school before the policy was enacted, 00:37:46.780 --> 00:37:49.390 and therefore they were not exposed 00:37:49.390 --> 00:37:52.270 to having poor students in school. 00:37:52.270 --> 00:37:53.830 For these treated schools there is 00:37:53.830 --> 00:37:57.340 essentially variation within school across cohorts 00:37:57.340 --> 00:37:59.890 that we can look at. 00:37:59.890 --> 00:38:01.250 Second within cohorts. 00:38:01.250 --> 00:38:04.000 We can look at treated and control schools. 00:38:04.000 --> 00:38:08.300 So some schools were getting additional poor kids and others 00:38:08.300 --> 00:38:08.800 did not. 00:38:08.800 --> 00:38:11.180 I'll tell you about this in a second. 00:38:11.180 --> 00:38:14.020 And then there's variation within classrooms 00:38:14.020 --> 00:38:17.800 to allow us to look at the role of personal interactions. 00:38:17.800 --> 00:38:19.780 There are, sort of, idiosyncratic assignments 00:38:19.780 --> 00:38:22.210 to study groups, so some students 00:38:22.210 --> 00:38:25.240 happen to be in study groups with a poor kid, 00:38:25.240 --> 00:38:26.830 and others were not. 00:38:32.795 --> 00:38:34.170 Now what does this study measure. 00:38:34.170 --> 00:38:36.750 The study measures three broad set of outcomes. 00:38:36.750 --> 00:38:39.420 It measures pro-social behavior and generosity. 00:38:39.420 --> 00:38:41.670 This is very much what we have already looked at. 00:38:41.670 --> 00:38:45.810 In particular, it looks at dictator games and volunteering 00:38:45.810 --> 00:38:47.670 for charities at school. 00:38:47.670 --> 00:38:51.318 The paper has a very nice mix of laboratory outcomes, 00:38:51.318 --> 00:38:52.860 sort of, measured in the field, which 00:38:52.860 --> 00:38:55.795 is, sort of, the dictator game that you're all familiar with. 00:38:55.795 --> 00:38:57.420 And field outcomes, were just, sort of, 00:38:57.420 --> 00:39:00.480 trying to collect real world outcomes 00:39:00.480 --> 00:39:02.880 in the sense of like things that are, perhaps, 00:39:02.880 --> 00:39:07.320 somewhat less contrived from the perspective of kids in school. 00:39:07.320 --> 00:39:09.240 And what's very nice about the study is 00:39:09.240 --> 00:39:12.780 that it seems to be that the results are very much aligned 00:39:12.780 --> 00:39:15.540 between, sort of, these live outcomes, the dictator game 00:39:15.540 --> 00:39:18.250 type outcomes, and the field outcomes, 00:39:18.250 --> 00:39:21.870 which is volunteering for charity at school in this case. 00:39:21.870 --> 00:39:26.010 Second, the author looks at discrimination 00:39:26.010 --> 00:39:27.570 in social interactions. 00:39:27.570 --> 00:39:30.780 In particular, he does it all sports contests and looks 00:39:30.780 --> 00:39:33.660 at teammates selection among these students 00:39:33.660 --> 00:39:38.010 and then willingness to attend play dates with poor students. 00:39:38.010 --> 00:39:41.370 Finally, the author also looks at academic outcomes, 00:39:41.370 --> 00:39:44.970 in particular test scores and disciplinary infractions. 00:39:44.970 --> 00:39:47.910 Now why might one want to look at academic outcomes, 00:39:47.910 --> 00:39:51.960 and why we're interested in social preferences, 00:39:51.960 --> 00:39:54.270 there's several reasons for that. 00:39:54.270 --> 00:39:55.920 In particular, an important reason 00:39:55.920 --> 00:39:59.830 here is that if you're against this type of policy, 00:39:59.830 --> 00:40:04.560 I might, sort of, show you that, oh, adding poor children 00:40:04.560 --> 00:40:07.140 might affect the social preference of rich children 00:40:07.140 --> 00:40:08.400 and so on in various ways. 00:40:08.400 --> 00:40:10.830 Maybe discretization goes down, but perhaps it's 00:40:10.830 --> 00:40:13.780 the case that all comes at the cost of academic performance, 00:40:13.780 --> 00:40:14.280 right. 00:40:14.280 --> 00:40:16.405 If you're trying to, sort of, implement this policy 00:40:16.405 --> 00:40:23.100 and persuade policymakers, maybe teachers or parents of like, 00:40:23.100 --> 00:40:27.690 OK, let's have more poor students in your school. 00:40:27.690 --> 00:40:29.640 Well, parents might, sort of, be OK fine. 00:40:29.640 --> 00:40:32.310 There's going to be like some change in social preferences, 00:40:32.310 --> 00:40:34.110 some change in discrimination, but really 00:40:34.110 --> 00:40:39.390 what we care about is test scores or discipline at school, 00:40:39.390 --> 00:40:42.180 and these poor kids might not be-- 00:40:42.180 --> 00:40:44.310 might be worse in terms of test scores, 00:40:44.310 --> 00:40:46.560 and they might have negative peer effects 00:40:46.560 --> 00:40:51.450 in terms of test scores but also in terms 00:40:51.450 --> 00:40:53.400 of disciplinary infractions. 00:40:53.400 --> 00:40:55.300 And, sort of, as a policy question, 00:40:55.300 --> 00:40:57.360 then it's really important to understand 00:40:57.360 --> 00:40:59.010 if there are some benefits in terms 00:40:59.010 --> 00:41:02.280 of poor sociology or reduced discrimination, 00:41:02.280 --> 00:41:04.050 do these benefits come at the cost 00:41:04.050 --> 00:41:07.005 of reduced academic performance? 00:41:12.093 --> 00:41:14.010 So now what is, in fact, the policy innovation 00:41:14.010 --> 00:41:15.690 in Delhi in 2007. 00:41:15.690 --> 00:41:20.340 There was a 20% admissions quota in private schools introduced 00:41:20.340 --> 00:41:22.980 for poor students in some of these private schools. 00:41:22.980 --> 00:41:26.737 There's a household income cutoff of $2,000 per year. 00:41:26.737 --> 00:41:28.320 So these are not, sort of, the poorest 00:41:28.320 --> 00:41:32.820 of the pole of the students who are qualified. 00:41:32.820 --> 00:41:35.760 Schools which received subsidized land 00:41:35.760 --> 00:41:39.870 from the government were essentially 00:41:39.870 --> 00:41:41.820 included in this policy change. 00:41:41.820 --> 00:41:44.693 That's over 90% of elite private schools. 00:41:44.693 --> 00:41:46.360 So think of these elite private schools. 00:41:46.360 --> 00:41:50.730 These are really, sort of, like very rich kids or parents 00:41:50.730 --> 00:41:52.013 of these kids. 00:41:52.013 --> 00:41:53.430 There's no fees for the poor kids, 00:41:53.430 --> 00:41:57.000 because they would not be able to afford these fees anyway. 00:41:57.000 --> 00:41:58.980 There was importantly also no tracking. 00:41:58.980 --> 00:42:03.870 So it wasn't that high ability or high performance kids 00:42:03.870 --> 00:42:05.310 would do very well where they like 00:42:05.310 --> 00:42:08.310 tracked into the high tracks and the good tracks verses 00:42:08.310 --> 00:42:09.540 lower tracks. 00:42:09.540 --> 00:42:11.730 Instead everybody was mixed together, 00:42:11.730 --> 00:42:15.600 poor and rich, high performing, low performing students. 00:42:15.600 --> 00:42:18.720 And the poor kids were selected using lotteries, 00:42:18.720 --> 00:42:21.600 which in principle also allows the author or others to look 00:42:21.600 --> 00:42:24.330 at the effective on those poor kids, the students who 00:42:24.330 --> 00:42:25.860 are selected versus not. 00:42:25.860 --> 00:42:29.040 To be clear this is not the subject of this paper. 00:42:29.040 --> 00:42:33.900 Just to give you a sense of the magnitude of like the mixing, 00:42:33.900 --> 00:42:38.400 you see is that the average beneficiary was a, sort of, 00:42:38.400 --> 00:42:45.690 at the 25th percentile of the income distribution in Delhi, 00:42:45.690 --> 00:42:51.130 and wealthy students are very much on the right tail-- 00:42:51.130 --> 00:42:54.540 sorry on the right tail of the distribution, something 00:42:54.540 --> 00:42:58.170 like the 90th percentile, 95th, which is about-- 00:42:58.170 --> 00:43:02.410 the US equivalent of that would be about $200,000 per year. 00:43:02.410 --> 00:43:06.420 The average beneficiary had like the US equivalent 00:43:06.420 --> 00:43:08.340 income of like $23,000. 00:43:08.340 --> 00:43:10.530 Again, these are not the poorest of the pole 00:43:10.530 --> 00:43:14.460 but relatively poor as about like an order of magnitude 00:43:14.460 --> 00:43:19.230 poorer than the average person in the actual school. 00:43:19.230 --> 00:43:23.640 Now this policy then induced large variation 00:43:23.640 --> 00:43:25.260 across classrooms. 00:43:25.260 --> 00:43:30.150 If you look at poor students in the fraction of the numbers 00:43:30.150 --> 00:43:33.810 of poor students in by grade in 2011, 00:43:33.810 --> 00:43:38.010 I told you the policy innovation was in 2007. 00:43:38.010 --> 00:43:42.540 So if you look at 2011-- that's four years later-- 00:43:42.540 --> 00:43:47.100 anybody who is in grade four in 2011 00:43:47.100 --> 00:43:51.420 has essentially no poor student in that class 00:43:51.420 --> 00:43:53.310 in those rich schools. 00:43:53.310 --> 00:43:57.870 So for students who were in 2011 in grade four, five, or six. 00:43:57.870 --> 00:43:59.610 That policy came too late. 00:43:59.610 --> 00:44:02.730 They did not have any poor classmates. 00:44:02.730 --> 00:44:07.140 In contrast, if you look at the lower grades, 00:44:07.140 --> 00:44:10.430 grades 3, 2, 1, and 0, which essentially preschool 00:44:10.430 --> 00:44:13.500 or minus 1, which is preschool grades, 00:44:13.500 --> 00:44:15.660 there's lots of additional poor students 00:44:15.660 --> 00:44:19.750 now in those rich schools. 00:44:19.750 --> 00:44:22.560 Now again, there's variation within schools 00:44:22.560 --> 00:44:23.490 and across schools. 00:44:23.490 --> 00:44:26.190 The variation within schools is across classrooms. 00:44:26.190 --> 00:44:29.190 That's, kind of, what I'm showing you here. 00:44:29.190 --> 00:44:31.883 But it's also a variation across schools. 00:44:31.883 --> 00:44:33.300 So just to be clear, the variation 00:44:33.300 --> 00:44:39.030 within schools across classrooms is like in the treated schools 00:44:39.030 --> 00:44:41.850 there's some classes like the fourth grade have 00:44:41.850 --> 00:44:45.090 like essentially zero poor kids in their classes, 00:44:45.090 --> 00:44:48.210 and the third grade, in contrast, 00:44:48.210 --> 00:44:52.390 have lots of poor kids in the classroom. 00:44:52.390 --> 00:44:55.080 They can do that comparison across classrooms 00:44:55.080 --> 00:44:56.340 within school. 00:44:56.340 --> 00:44:59.040 In addition, has also variation across schools. 00:44:59.040 --> 00:45:01.920 There is, in particular, there's treatment schools, which 00:45:01.920 --> 00:45:04.590 comply in 2007 as they were supposed to, 00:45:04.590 --> 00:45:07.200 but then there's also delayed treatment schools, 00:45:07.200 --> 00:45:08.940 which complied in 2008. 00:45:08.940 --> 00:45:11.100 So the jump shifted by one cohort. 00:45:11.100 --> 00:45:13.350 So the cohort, the jump that I showed you here, 00:45:13.350 --> 00:45:17.820 this is for the treatment schools that complied 00:45:17.820 --> 00:45:20.580 or this for everybody that complied in 2007. 00:45:20.580 --> 00:45:22.540 The delayed one complied a year later, 00:45:22.540 --> 00:45:26.820 so you'll see them essentially, somewhat later increasing-- 00:45:26.820 --> 00:45:28.590 or one year later exactly increasing 00:45:28.590 --> 00:45:31.020 the fraction of poor kids in their classroom. 00:45:31.020 --> 00:45:32.970 And then there is the control schools 00:45:32.970 --> 00:45:36.630 that were not subject to the policy, at least until 2013. 00:45:36.630 --> 00:45:39.625 So they had received-- 00:45:39.625 --> 00:45:42.000 they had received land from federal government or private 00:45:42.000 --> 00:45:44.140 foundations. 00:45:44.140 --> 00:45:46.410 So essentially there was no treatment. 00:45:46.410 --> 00:45:50.700 That is all students are rich in these types of classes. 00:45:50.700 --> 00:45:53.630 So that is-- the land receiving is essentially just-- 00:45:53.630 --> 00:45:57.180 that was like a fairly arbitrary rule in some ways. 00:45:57.180 --> 00:46:00.470 So that, sort of, leads to at least quasi random variation. 00:46:00.470 --> 00:46:07.280 And some schools received the treatment, and others did not. 00:46:07.280 --> 00:46:09.330 Now let me show you some outcomes. 00:46:09.330 --> 00:46:11.960 So one outcome is our friend the dictator game which you have 00:46:11.960 --> 00:46:15.350 seen quite a bit and played already as well is students are 00:46:15.350 --> 00:46:18.860 allowed with 10 rupees, which is not very much-- 00:46:18.860 --> 00:46:21.380 that's about like $0.15-- 00:46:21.380 --> 00:46:23.540 and they choose to share-- 00:46:23.540 --> 00:46:25.760 notice that these are pretty young kids, 00:46:25.760 --> 00:46:28.040 so 10 rupees is quite a bit of money for them. 00:46:28.040 --> 00:46:33.170 They choose to share an amount as before between 0 and 10. 00:46:33.170 --> 00:46:36.203 And Gautam sets this up such that students could-- 00:46:36.203 --> 00:46:38.120 it was not just about the money, but they also 00:46:38.120 --> 00:46:41.850 can exchange the money for candy later, add 1 rupee per piece. 00:46:41.850 --> 00:46:45.190 So that was a pretty good deal. 00:46:45.190 --> 00:46:48.490 Then the dictator games for the order was randomized. 00:46:48.490 --> 00:46:50.350 The kids would play two games. 00:46:50.350 --> 00:46:53.800 Again, so all rich kids would do that. 00:46:53.800 --> 00:46:57.340 They play first game one where the recipient is a poor student 00:46:57.340 --> 00:46:59.170 in a school for poor children. 00:46:59.170 --> 00:47:01.990 And in game 2, it's a rich student 00:47:01.990 --> 00:47:03.880 in a private control school. 00:47:03.880 --> 00:47:07.750 Notice it's a control school to be clear, 00:47:07.750 --> 00:47:10.960 so it's not a kid that they might know. 00:47:10.960 --> 00:47:15.430 So the kids are chosen such that the kids were 00:47:15.430 --> 00:47:18.640 given the names and the photographs of the school shown 00:47:18.640 --> 00:47:20.440 to the subject, to the kids. 00:47:20.440 --> 00:47:21.850 So you could sort of see-- 00:47:21.850 --> 00:47:23.470 and this is verified in the briefing 00:47:23.470 --> 00:47:26.710 by the author-- is that the children understand 00:47:26.710 --> 00:47:30.370 very well who is a rich kid and who's a poor kid, 00:47:30.370 --> 00:47:33.290 but these are not kids that they would actually know. 00:47:33.290 --> 00:47:35.290 And something that's a less interesting question 00:47:35.290 --> 00:47:38.530 if you have gone to school with somebody for a long time, 00:47:38.530 --> 00:47:41.740 you might have seen them around in your school yard, 00:47:41.740 --> 00:47:44.450 you might be nicer to them or less nice to them. 00:47:44.450 --> 00:47:46.330 That's not what the study is looking at. 00:47:46.330 --> 00:47:50.200 The study is looking at other kids in other schools 00:47:50.200 --> 00:47:53.530 that look rich or look poor both in terms 00:47:53.530 --> 00:47:56.930 of the way they're dressed and so on, 00:47:56.930 --> 00:47:59.570 but in particular the way their school looks like. 00:47:59.570 --> 00:48:01.085 So essentially these are-- 00:48:01.085 --> 00:48:02.710 now we're looking at social preferences 00:48:02.710 --> 00:48:07.310 towards other people, people that these kids don't know. 00:48:07.310 --> 00:48:09.410 OK, so now what do we see. 00:48:09.410 --> 00:48:11.570 So this is a generosity to the pole. 00:48:11.570 --> 00:48:14.690 So like this is essentially the results from game number one. 00:48:14.690 --> 00:48:18.500 Again, remember these are all rich kids who are giving now 00:48:18.500 --> 00:48:22.190 to poor kids, and the question now is being exposed to 00:48:22.190 --> 00:48:27.860 or does being exposed to poor children for several years, 00:48:27.860 --> 00:48:32.480 does that affect giving in the form of dictator 00:48:32.480 --> 00:48:36.230 games towards poor children to start with. 00:48:36.230 --> 00:48:39.270 And we're going to talk about rich students after that. 00:48:39.270 --> 00:48:43.430 So here you see the generosity to the poor in control schools. 00:48:43.430 --> 00:48:47.450 You see this is the percent to the poor recipient. 00:48:47.450 --> 00:48:50.340 It's remarkably similar to what we've seen previously, 00:48:50.340 --> 00:48:52.120 which is about 20% to 30%. 00:48:52.120 --> 00:48:56.570 About 25% is given to the pole kid in the control 00:48:56.570 --> 00:48:59.930 groups in the control schools. 00:48:59.930 --> 00:49:04.190 This fraction seems to trend up a little bit over time, 00:49:04.190 --> 00:49:07.370 but it's also quite constant over time across grades. 00:49:07.370 --> 00:49:10.970 You see here by grades two, three, four, and five 00:49:10.970 --> 00:49:11.990 on the x-axis. 00:49:11.990 --> 00:49:15.320 On the y-axis, it's the percent given to the poor recipients. 00:49:15.320 --> 00:49:17.240 Now adding the treatment schools, 00:49:17.240 --> 00:49:20.600 now we find that for two grades-- 00:49:20.600 --> 00:49:23.030 for grade two and grade three, remember the graph 00:49:23.030 --> 00:49:24.940 that I showed you here. 00:49:24.940 --> 00:49:27.610 You see that in grades three, two, one, 00:49:27.610 --> 00:49:29.200 and so on, these are the kids who 00:49:29.200 --> 00:49:33.130 are exposed to the poor children in class in contrast 00:49:33.130 --> 00:49:35.410 to grades four and five that have not been exposed 00:49:35.410 --> 00:49:37.540 to poor kids in their class. 00:49:37.540 --> 00:49:41.620 So what this graph shows is that exactly the grades 00:49:41.620 --> 00:49:44.470 three and two that have been exposed-- 00:49:44.470 --> 00:49:46.000 in the treatment groups that have 00:49:46.000 --> 00:49:48.280 been exposed to poor classmates, these 00:49:48.280 --> 00:49:55.030 are exactly the students, the grades where giving is higher 00:49:55.030 --> 00:49:58.690 towards the poor compared to the control groups 00:49:58.690 --> 00:50:02.710 but also compared to the older cohorts four and five, 00:50:02.710 --> 00:50:04.090 the grade four and five. 00:50:04.090 --> 00:50:07.070 When you look at grades four and five, 00:50:07.070 --> 00:50:09.320 they see essentially no difference between treatment 00:50:09.320 --> 00:50:10.640 and control schools. 00:50:10.640 --> 00:50:12.530 So they seem to be very similar. 00:50:12.530 --> 00:50:15.170 In contrast, if you look at grades three 00:50:15.170 --> 00:50:18.770 and through where the treatment kids are exposed 00:50:18.770 --> 00:50:22.430 to the treatment by having the poor kids in their classrooms, 00:50:22.430 --> 00:50:27.020 these are exactly the years where giving goes up and goes 00:50:27.020 --> 00:50:27.890 up by quite a bit. 00:50:27.890 --> 00:50:31.730 While the average was, as I said before, about 25%, 00:50:31.730 --> 00:50:34.010 that goes up to about 35%. 00:50:34.010 --> 00:50:36.455 So that's a pretty large relative increase. 00:50:39.650 --> 00:50:42.530 Now in addition, as I told you, there 00:50:42.530 --> 00:50:45.800 are also delayed treatment groups, treatment schools. 00:50:45.800 --> 00:50:48.968 For those schools, if you were in grade three, 00:50:48.968 --> 00:50:51.260 it was also too late for you, because the treatment was 00:50:51.260 --> 00:50:53.490 introduced only a year later. 00:50:53.490 --> 00:50:55.850 So if you were in grade three, you 00:50:55.850 --> 00:50:58.880 did not have any poor kids in your school. 00:50:58.880 --> 00:51:00.800 Now if you look at the yellow line here, 00:51:00.800 --> 00:51:03.620 the yellow line looks very much like the green line. 00:51:03.620 --> 00:51:06.200 That is for the delay treatment schools, 00:51:06.200 --> 00:51:08.180 there's no impact either. 00:51:08.180 --> 00:51:11.360 They look very much like the control group of schools, 00:51:11.360 --> 00:51:13.710 as you would expect. 00:51:13.710 --> 00:51:17.960 However, in the delayed treatment schools 00:51:17.960 --> 00:51:22.460 in grade number two, that's the grade where 00:51:22.460 --> 00:51:23.640 the students are exposed-- 00:51:23.640 --> 00:51:26.450 the rich students are exposed to the poor kids using 00:51:26.450 --> 00:51:31.340 the essentially similar effects to the treatment schools, 00:51:31.340 --> 00:51:32.520 as you expect. 00:51:32.520 --> 00:51:34.070 So that's fairly compelling evidence 00:51:34.070 --> 00:51:39.230 that really the differences that you see across these schools 00:51:39.230 --> 00:51:43.790 are driven by the treatment that was-- 00:51:43.790 --> 00:51:48.770 the timing of the treatment as opposed to other potential 00:51:48.770 --> 00:51:52.100 effects due to like selection or maybe these schools 00:51:52.100 --> 00:51:53.390 are different and so on. 00:51:53.390 --> 00:51:55.070 We see essentially no differences 00:51:55.070 --> 00:51:58.650 in grades four and five for any of the treatment and control 00:51:58.650 --> 00:51:59.330 schools. 00:51:59.330 --> 00:52:02.150 You also see no difference in grades three 00:52:02.150 --> 00:52:03.890 for the delayed treatment schools 00:52:03.890 --> 00:52:05.430 compared to the control group. 00:52:05.430 --> 00:52:07.970 In contrast, we see clear differences 00:52:07.970 --> 00:52:11.180 in grades two for the delay in the treatment school as 00:52:11.180 --> 00:52:12.770 compared to the control schools. 00:52:12.770 --> 00:52:16.040 We also see a clear difference for the treatment schools 00:52:16.040 --> 00:52:19.280 compared to both the control schools and the delayed 00:52:19.280 --> 00:52:22.650 treatment schools as well. 00:52:22.650 --> 00:52:25.400 So that's fairly compelling evidence 00:52:25.400 --> 00:52:28.550 that game or play, dictator games 00:52:28.550 --> 00:52:32.690 changes when students are exposed 00:52:32.690 --> 00:52:37.750 for quite a long time to poor kids in their classroom. 00:52:37.750 --> 00:52:41.830 OK so now in addition, we will have 00:52:41.830 --> 00:52:44.270 variation within classroom. 00:52:44.270 --> 00:52:46.690 So what I show you here, this is all variation 00:52:46.690 --> 00:52:48.400 across classrooms. 00:52:48.400 --> 00:52:51.640 That is to say, if you have a poor kid in your classroom, 00:52:51.640 --> 00:52:55.760 you will behave differently, or that will change your attitude 00:52:55.760 --> 00:52:58.840 toward your giving and dictator games 00:52:58.840 --> 00:53:03.370 towards the poor or other poor children. 00:53:03.370 --> 00:53:06.490 Now, in addition, the author also 00:53:06.490 --> 00:53:10.750 has evidence of variation within classroom. 00:53:10.750 --> 00:53:12.500 Now, he lets you think about for a second, 00:53:12.500 --> 00:53:15.590 why is it important to also have variation within classroom? 00:53:15.590 --> 00:53:18.410 Isn't the evidence that we have here already compelling enough? 00:53:18.410 --> 00:53:20.643 Why do we need more? 00:53:20.643 --> 00:53:22.310 And you can think about it for a second. 00:53:29.530 --> 00:53:33.250 The reason for that is that the diff-in-diff, the difference 00:53:33.250 --> 00:53:36.790 in differences comparing essentially across classrooms, 00:53:36.790 --> 00:53:39.190 identifies the overall effective having 00:53:39.190 --> 00:53:41.380 poor classmates in class. 00:53:41.380 --> 00:53:44.920 That might be the result of personal interactions. 00:53:44.920 --> 00:53:48.160 But it might also be the result of the teacher. 00:53:48.160 --> 00:53:49.390 The curriculum might change. 00:53:49.390 --> 00:53:52.010 Your parents might change in some ways in just telling you, 00:53:52.010 --> 00:53:52.510 look. 00:53:52.510 --> 00:53:54.460 There's these poor kids in your class. 00:53:54.460 --> 00:53:56.350 The teachers might tell everybody, oh, you 00:53:56.350 --> 00:53:58.120 need to be nice to poor kids. 00:53:58.120 --> 00:53:59.320 The curriculum might change. 00:53:59.320 --> 00:54:00.820 It might sort of include things that 00:54:00.820 --> 00:54:02.770 tell you to be nice to poor kids and so on. 00:54:02.770 --> 00:54:04.937 You might be sort of exposed to just different types 00:54:04.937 --> 00:54:05.860 of material. 00:54:05.860 --> 00:54:08.050 So really, given the evidence that we have here, 00:54:08.050 --> 00:54:10.480 when you just compare across classrooms, 00:54:10.480 --> 00:54:14.170 you cannot disentangle whether this is driven from being 00:54:14.170 --> 00:54:18.430 personally exposed in terms of interacting with poor children 00:54:18.430 --> 00:54:20.830 personally by just talking to them in certain ways, 00:54:20.830 --> 00:54:24.290 or being exposed to them, or learning about them and so on. 00:54:24.290 --> 00:54:27.820 So you cannot disentangle personal interactions from 00:54:27.820 --> 00:54:30.430 teacher or curriculum changes that are essentially 00:54:30.430 --> 00:54:32.960 at the classroom level. 00:54:32.960 --> 00:54:35.320 So in addition, then, Gautam also 00:54:35.320 --> 00:54:40.360 has a within-classroom strategy that 00:54:40.360 --> 00:54:43.180 exploits the assignments of study groups. 00:54:43.180 --> 00:54:44.860 What is that doing is essentially, 00:54:44.860 --> 00:54:48.190 it isolates the role of direct personal interactions 00:54:48.190 --> 00:54:50.630 and is not subject to sorting concerns. 00:54:50.630 --> 00:54:53.260 Another concern might be that you might sort of say, 00:54:53.260 --> 00:54:57.820 oh, well, in this school, they're 00:54:57.820 --> 00:55:01.000 having poor kids that are allowed 00:55:01.000 --> 00:55:02.260 into this private school. 00:55:02.260 --> 00:55:05.120 I'm going to send my kids to somebody else. 00:55:05.120 --> 00:55:10.150 So that's also an issue perhaps potentially in the evidence 00:55:10.150 --> 00:55:12.250 that I showed you before across classrooms. 00:55:12.250 --> 00:55:14.080 The within-classroom strategy is now 00:55:14.080 --> 00:55:17.170 looking at everybody who is in that classroom 00:55:17.170 --> 00:55:22.900 and looking at some kids were just randomly assigned 00:55:22.900 --> 00:55:25.810 to be in study groups with poor kids, and others were not. 00:55:25.810 --> 00:55:28.240 And that sort of allowed us to disentangle 00:55:28.240 --> 00:55:30.085 or allows us to isolate the effect 00:55:30.085 --> 00:55:32.870 of personal interactions. 00:55:32.870 --> 00:55:36.670 So one hour a day, kids were working in small groups 00:55:36.670 --> 00:55:38.530 of 2 to 4 students. 00:55:38.530 --> 00:55:41.200 Remember, this was for a very long period of time. 00:55:41.200 --> 00:55:43.450 So it's one hour per day, but it was also 00:55:43.450 --> 00:55:44.450 for quite a bit of time. 00:55:44.450 --> 00:55:46.090 It's not just doing that once or twice. 00:55:46.090 --> 00:55:48.610 It's more an extended period of time. 00:55:48.610 --> 00:55:51.760 And crucially, some schools use the alphabetic order 00:55:51.760 --> 00:55:54.450 of first names to assign study groups. 00:55:54.450 --> 00:55:56.350 So that's essentially exogenous variation 00:55:56.350 --> 00:55:58.090 in personal interactions. 00:55:58.090 --> 00:56:04.120 Other schools frequently shuffled the groups and only-- 00:56:04.120 --> 00:56:08.020 which essentially sort of-- 00:56:08.020 --> 00:56:11.080 everybody was equally exposed to those kinds of-- 00:56:11.080 --> 00:56:13.300 to different students. 00:56:13.300 --> 00:56:16.270 So you are sometimes in study groups with rich students 00:56:16.270 --> 00:56:18.760 only and sometimes with poor students only. 00:56:18.760 --> 00:56:23.170 In the first group where the alphabetic order was used, 00:56:23.170 --> 00:56:25.600 you were either sort of in the study group 00:56:25.600 --> 00:56:32.170 where you happened to be next to somebody who's 00:56:32.170 --> 00:56:34.360 a poor kid because they happened to be next to you 00:56:34.360 --> 00:56:38.000 in the alphabet, or you're not, and that stayed the same. 00:56:38.000 --> 00:56:40.370 Now, what does this let us look at? 00:56:40.370 --> 00:56:44.190 Essentially, the alphabetical order predicts study partners. 00:56:44.190 --> 00:56:46.790 So now you might say people with different names might also 00:56:46.790 --> 00:56:47.520 be different. 00:56:47.520 --> 00:56:49.400 So it could just be that like my name-- 00:56:49.400 --> 00:56:51.350 because my name is different, I'm 00:56:51.350 --> 00:56:55.890 more likely to be next in the alphabet to a poor student. 00:56:55.890 --> 00:56:58.250 For example, if I'm a rich kid and there 00:56:58.250 --> 00:57:02.000 are some names that are more similar to poor kids' names, 00:57:02.000 --> 00:57:04.610 it could be that my parents are particularly tolerant. 00:57:04.610 --> 00:57:09.050 They also gave me a name that sounds like a poor person. 00:57:09.050 --> 00:57:11.510 And therefore, I'm more likely to be in a study group 00:57:11.510 --> 00:57:14.420 with the poor kids. 00:57:14.420 --> 00:57:16.910 And that's not the impact of being in the study group 00:57:16.910 --> 00:57:18.170 or rather sort of selection. 00:57:18.170 --> 00:57:20.075 It could be that names are different 00:57:20.075 --> 00:57:22.440 or people with different names are different. 00:57:22.440 --> 00:57:25.050 So that's why having this control group is really nice. 00:57:25.050 --> 00:57:26.900 So what we can look at now is to say, 00:57:26.900 --> 00:57:33.200 you can look at kids that have names adjacent to rich students 00:57:33.200 --> 00:57:36.810 and kids that have names adjacent to poor students. 00:57:36.810 --> 00:57:39.095 So if you look at the left side, this 00:57:39.095 --> 00:57:41.060 is essentially looking at students 00:57:41.060 --> 00:57:45.890 that have their names adjacent to rich students only 00:57:45.890 --> 00:57:48.350 or students that have names adjacent to also 00:57:48.350 --> 00:57:49.670 some poor students. 00:57:49.670 --> 00:57:53.460 When you look at that, where the alphabetic order was not used, 00:57:53.460 --> 00:57:55.400 the outcome here is what's your share 00:57:55.400 --> 00:57:56.833 of having poor study partners? 00:57:56.833 --> 00:57:58.250 And so there, essentially, there's 00:57:58.250 --> 00:58:01.490 no difference, as it should be, because the alphabetic order 00:58:01.490 --> 00:58:02.210 was not used. 00:58:02.210 --> 00:58:04.040 There was a shuffling around the kids 00:58:04.040 --> 00:58:05.630 all the time, so it didn't really 00:58:05.630 --> 00:58:07.040 matter what your alphabet was. 00:58:07.040 --> 00:58:08.040 It didn't really matter. 00:58:08.040 --> 00:58:10.370 But crucially, we can still look at the alphabet. 00:58:10.370 --> 00:58:14.787 We can look at kids that have names 00:58:14.787 --> 00:58:16.370 where, Jason, in the alphabet, there's 00:58:16.370 --> 00:58:18.800 only rich kids, rich students. 00:58:18.800 --> 00:58:22.160 Or they can look at kids that have 00:58:22.160 --> 00:58:24.380 names where, adjacent to the alphabets, 00:58:24.380 --> 00:58:25.650 there are poor students. 00:58:25.650 --> 00:58:27.650 And, here you see on the left side of the graph, 00:58:27.650 --> 00:58:31.890 you see essentially no difference across these groups. 00:58:31.890 --> 00:58:35.270 So they were equally likely to have a poor kid in their study 00:58:35.270 --> 00:58:36.200 group. 00:58:36.200 --> 00:58:38.690 In contrast, on the right side of the graph, 00:58:38.690 --> 00:58:45.150 you see the alphabetic order was used to assign study groups. 00:58:45.150 --> 00:58:50.450 So if you had a name adjacent to a poor student in the alphabet, 00:58:50.450 --> 00:58:54.260 you're very, very likely to have a poor kid assigned to you 00:58:54.260 --> 00:58:55.400 or be in your study group. 00:58:55.400 --> 00:58:58.440 Not always because sometimes, these are groups of three. 00:58:58.440 --> 00:59:00.980 So sometimes, you were lucky or unlucky 00:59:00.980 --> 00:59:02.930 depending on how you view it. 00:59:02.930 --> 00:59:06.350 The group was just above you or below you in the alphabet. 00:59:06.350 --> 00:59:07.880 But most of the time, if you have 00:59:07.880 --> 00:59:10.520 a name adjacent to a poor student, that student 00:59:10.520 --> 00:59:15.290 or any student would be in your study group. 00:59:15.290 --> 00:59:19.564 If you didn't have a name adjacent to you or f 00:59:19.564 --> 00:59:21.530 you only had rich students adjacent to you, 00:59:21.530 --> 00:59:23.030 there's still a chance that you have 00:59:23.030 --> 00:59:24.380 a poor student in your study group 00:59:24.380 --> 00:59:26.720 because it could be just like not the person next to you 00:59:26.720 --> 00:59:31.350 but another person further down is, in fact, a poor student. 00:59:31.350 --> 00:59:33.920 You end up in that group with that student. 00:59:33.920 --> 00:59:36.347 Remember, these are groups of 2 to 4 students. 00:59:36.347 --> 00:59:38.930 So it could just be that you're in a group of 3 to 4 students, 00:59:38.930 --> 00:59:42.230 and not the person adjacent to you but the person after that 00:59:42.230 --> 00:59:44.820 ends up in your study group. 00:59:44.820 --> 00:59:47.570 So you're still likely to have-- 00:59:47.570 --> 00:59:50.570 you have still like about like a 40% chance of having 00:59:50.570 --> 00:59:54.500 a poor kid in your study group. 00:59:54.500 --> 00:59:56.600 But there's a huge difference between the two 00:59:56.600 --> 00:59:58.190 types of groups. 00:59:58.190 --> 01:00:00.710 The name adjacent to the study group 01:00:00.710 --> 01:00:05.330 has a huge fraction, about 90%, versus 40%. 01:00:05.330 --> 01:00:07.520 So crucially, now, we can compare 01:00:07.520 --> 01:00:10.040 for both types of schools, for the schools 01:00:10.040 --> 01:00:13.010 on the right-hand side where, essentially, the names are very 01:00:13.010 --> 01:00:18.200 predictive of having a poor kid in your study group, 01:00:18.200 --> 01:00:20.780 versus on the left-hand side where the names are not 01:00:20.780 --> 01:00:24.920 predicative at all whether you have a poor kid in your study 01:00:24.920 --> 01:00:25.680 group. 01:00:25.680 --> 01:00:27.350 And so we can-- 01:00:27.350 --> 01:00:30.260 that way, we can keep selection the same, the types of name 01:00:30.260 --> 01:00:31.040 are the same. 01:00:31.040 --> 01:00:33.080 We are isolating the impact of having 01:00:33.080 --> 01:00:35.540 a poor kid in your study group. 01:00:35.540 --> 01:00:36.920 Now, what does the author find? 01:00:36.920 --> 01:00:39.570 He finds that-- this is what we showed you before, 01:00:39.570 --> 01:00:43.430 which is in the control group, we had about, 01:00:43.430 --> 01:00:46.109 in the dictator game, these kids gave about 27%. 01:00:51.350 --> 01:00:53.060 Having a poor kid in your classroom. 01:00:53.060 --> 01:00:54.977 That's the evidence that I already showed you. 01:00:54.977 --> 01:00:58.980 The treatment effect is about 12 percentage points. 01:00:58.980 --> 01:00:59.940 That's fairly large. 01:00:59.940 --> 01:01:02.900 That's almost like a 50% increase. 01:01:02.900 --> 01:01:04.290 A little bit less than that. 01:01:04.290 --> 01:01:06.620 So that's a huge increase. 01:01:06.620 --> 01:01:08.960 And then in addition, when you look 01:01:08.960 --> 01:01:14.330 at kids that have a poor study partner, 01:01:14.330 --> 01:01:16.580 notice that these are things not necessarily additive, 01:01:16.580 --> 01:01:19.220 but if you have a poor study partner versus not, 01:01:19.220 --> 01:01:23.340 there's an effective about 7.5 percentage points. 01:01:23.340 --> 01:01:26.100 That's a pretty large effect. 01:01:26.100 --> 01:01:29.690 So both of these things seem to matter. 01:01:29.690 --> 01:01:32.240 Personal interactions seem to be quite important. 01:01:32.240 --> 01:01:33.890 But in addition, there seem to be also 01:01:33.890 --> 01:01:35.600 some additional effects perhaps coming 01:01:35.600 --> 01:01:37.640 from the classroom level. 01:01:37.640 --> 01:01:39.740 You might-- so there's two types of effects 01:01:39.740 --> 01:01:40.940 that are possible here. 01:01:40.940 --> 01:01:44.030 Some types of effects could be from the teacher, parents, 01:01:44.030 --> 01:01:45.090 curriculum, et cetera. 01:01:45.090 --> 01:01:46.190 Might be different. 01:01:46.190 --> 01:01:47.720 Or other things could be just like, 01:01:47.720 --> 01:01:50.690 even if you don't have a poor kid in your study group, 01:01:50.690 --> 01:01:52.940 you might still play with them or be friends with them 01:01:52.940 --> 01:01:55.880 or just seeing them around in the classroom might affect you. 01:01:55.880 --> 01:01:59.690 But crucially, the personal interactions 01:01:59.690 --> 01:02:03.305 seem to be very important as well. 01:02:03.305 --> 01:02:05.430 Now, one question you might ask is, well, so far, I 01:02:05.430 --> 01:02:08.490 showed you generosity towards poor children. 01:02:08.490 --> 01:02:09.690 So that was game number 1. 01:02:09.690 --> 01:02:14.190 We looked at what happens in the dictator games 01:02:14.190 --> 01:02:16.170 when a rich kid plays with a poor kid 01:02:16.170 --> 01:02:19.810 and the rich kids become nicer towards the poor kids. 01:02:19.810 --> 01:02:23.640 Now, how about generosity towards other wealthy children? 01:02:23.640 --> 01:02:27.000 Well, it turns out that that increases as well. 01:02:27.000 --> 01:02:29.850 So it's a smaller effect size, but it's still substantial, 01:02:29.850 --> 01:02:32.830 and, in fact, statistically significant. 01:02:32.830 --> 01:02:34.230 So why might that be? 01:02:34.230 --> 01:02:37.030 You might sort of ask, well, what's going on here? 01:02:37.030 --> 01:02:38.790 So here's the evidence. 01:02:38.790 --> 01:02:42.130 You sort of see the change in giving to rich participants. 01:02:42.130 --> 01:02:43.630 This is sort of like a distribution. 01:02:43.630 --> 01:02:47.550 You see this is the share giving to the rich participant. 01:02:47.550 --> 01:02:49.013 It seems to be what happens-- 01:02:49.013 --> 01:02:50.430 these are essentially the fraction 01:02:50.430 --> 01:02:53.190 of students in the different treatment versus control 01:02:53.190 --> 01:02:56.760 who give 0%, 10%, 20%, 30%, 40%, 50%. 01:02:56.760 --> 01:02:59.850 And what you see-- there is about a 10 percentage points 01:02:59.850 --> 01:03:04.350 decrease in the fraction who give 0% and about 01:03:04.350 --> 01:03:07.110 an 8-something percentage points increase 01:03:07.110 --> 01:03:09.750 in the fraction who give 50%. 01:03:09.750 --> 01:03:12.510 So it seems to be that there's quite a few students who 01:03:12.510 --> 01:03:14.310 move from 0%. 01:03:14.310 --> 01:03:19.830 About 10% of them move from 0% to 50%. 01:03:19.830 --> 01:03:23.400 And 50-50 is kind of like the equal allocation. 01:03:23.400 --> 01:03:26.880 So rich kids also becoming nicer towards rich kids 01:03:26.880 --> 01:03:28.620 in these dictator games. 01:03:28.620 --> 01:03:31.230 So what's going on here? 01:03:31.230 --> 01:03:33.810 There's different potential explanations. 01:03:33.810 --> 01:03:37.590 Perhaps the most plausible explanation-- the study 01:03:37.590 --> 01:03:39.540 has some evidence on that. 01:03:39.540 --> 01:03:42.060 At the end of the day, it's hard to sort of entirely 01:03:42.060 --> 01:03:47.230 nail this or rule out all other potential explanation, 01:03:47.230 --> 01:03:50.160 but it's quite plausible that what's happening here 01:03:50.160 --> 01:03:53.160 is that inequity, students, when you're 01:03:53.160 --> 01:03:58.530 exposed to poor children, you are essentially a bit more 01:03:58.530 --> 01:04:05.010 averse to inequality or inequity across people, even in things, 01:04:05.010 --> 01:04:08.610 in fairly trivial things such as dictator games. 01:04:08.610 --> 01:04:11.130 Now, I told you before, that's a little bit funny 01:04:11.130 --> 01:04:13.620 because dictator games obviously are very narrowly 01:04:13.620 --> 01:04:16.710 framing people and looking at essentially very 01:04:16.710 --> 01:04:18.900 narrow outcomes. 01:04:18.900 --> 01:04:22.560 If we have, like, a 50/50 outcome in the dictator game, 01:04:22.560 --> 01:04:25.410 that doesn't mean that our life is the same. 01:04:25.410 --> 01:04:27.660 You might be still much richer than I am. 01:04:27.660 --> 01:04:30.450 So having sort of equal outcomes in the dictator, 01:04:30.450 --> 01:04:32.910 the 50-50 allocation in the dictator game, 01:04:32.910 --> 01:04:35.910 might be narrowly framed, in that particular game, fair. 01:04:35.910 --> 01:04:38.730 But of course, it's not fair in the grand scheme of things. 01:04:38.730 --> 01:04:41.180 But what seems to be the case is that-- 01:04:41.180 --> 01:04:43.230 and there's some other evidence in the paper 01:04:43.230 --> 01:04:45.420 that you can read if you would like. 01:04:45.420 --> 01:04:49.320 It seems to be the case that rich children become 01:04:49.320 --> 01:04:53.160 more averse to unequal outcomes in the world in general 01:04:53.160 --> 01:04:56.220 because they essentially see these poor kids who 01:04:56.220 --> 01:04:59.160 are very smart and are disadvantaged 01:04:59.160 --> 01:05:02.190 in terms of various ways from having 01:05:02.190 --> 01:05:03.780 lower wealth of their parents. 01:05:03.780 --> 01:05:06.750 The rich kids become sort of more adverse against that, 01:05:06.750 --> 01:05:09.660 and that translates even into dictator games 01:05:09.660 --> 01:05:13.260 with other rich kids in a very sort of minor thing 01:05:13.260 --> 01:05:15.250 in the world and in these dictator games, 01:05:15.250 --> 01:05:17.250 again, even though these dictator games actually 01:05:17.250 --> 01:05:18.210 don't change very much. 01:05:18.210 --> 01:05:21.550 But they really seem to be averse to inequality 01:05:21.550 --> 01:05:25.650 of these outcomes and move them from the 10-0 allocation 01:05:25.650 --> 01:05:31.910 to 50-50, even with these rich kids. 01:05:31.910 --> 01:05:32.660 OK. 01:05:32.660 --> 01:05:35.240 So that was evidence on social preferences 01:05:35.240 --> 01:05:37.070 as measured by dictator games. 01:05:37.070 --> 01:05:41.120 But people also had some evidence on discrimination. 01:05:41.120 --> 01:05:45.210 In particular, a small field experiment on team selection. 01:05:45.210 --> 01:05:47.390 So what does this study do? 01:05:47.390 --> 01:05:49.790 So its subjects are, again, students 01:05:49.790 --> 01:05:51.660 from two elite private schools. 01:05:51.660 --> 01:05:55.970 So now, it's like two of these selected schools are selected. 01:05:55.970 --> 01:05:57.650 One is a treatment school. 01:05:57.650 --> 01:05:59.060 One is a control group. 01:05:59.060 --> 01:06:03.830 And in addition, Gautam invited athletic poor students 01:06:03.830 --> 01:06:04.790 from a public school. 01:06:04.790 --> 01:06:07.280 Importantly, he said, athletic students-- these 01:06:07.280 --> 01:06:10.240 are students who are better at sports than the rich kids. 01:06:10.240 --> 01:06:11.990 You might think the poorer kids are better 01:06:11.990 --> 01:06:14.940 at sports than the rich kids anyway, which is probably true. 01:06:14.940 --> 01:06:18.380 But now, these are particularly athletic students on purpose 01:06:18.380 --> 01:06:22.940 who are invited to attend a sports event as well. 01:06:22.940 --> 01:06:25.920 Now, students in this experiment, in this game, 01:06:25.920 --> 01:06:29.570 must choose teammates to run a relay race. 01:06:29.570 --> 01:06:35.480 Now, when you're a rich kid who is thinking about who 01:06:35.480 --> 01:06:37.310 should be in my team, you can either 01:06:37.310 --> 01:06:41.780 choose a rich kid who is kind of like similar to you 01:06:41.780 --> 01:06:46.730 in social ways, or you can choose a poor kid 01:06:46.730 --> 01:06:48.110 who you don't know very much. 01:06:48.110 --> 01:06:50.660 You might actually not like the poor kids. 01:06:50.660 --> 01:06:54.840 But the poor kid is a lot better in the running, 01:06:54.840 --> 01:06:56.990 so it might be much better for you. 01:06:56.990 --> 01:06:59.150 You might be much more likely to win in the game 01:06:59.150 --> 01:07:02.190 because now your partner in the relay race is much faster. 01:07:02.190 --> 01:07:04.520 So it's a very nice trade-off between ability-- 01:07:04.520 --> 01:07:08.960 choosing the fast runner versus social similarity. 01:07:08.960 --> 01:07:12.440 And now, what Gautam then is doing 01:07:12.440 --> 01:07:16.340 is like, if you sort of choose the rich kid, that's 01:07:16.340 --> 01:07:19.430 then a measure of discrimination because essentially, you're 01:07:19.430 --> 01:07:23.870 choosing a worse runner in favor of 01:07:23.870 --> 01:07:26.660 or because you want more social similarity. 01:07:26.660 --> 01:07:28.820 You don't want to hang out with a poor kid. 01:07:28.820 --> 01:07:30.680 Instead, you choose the rich kid, 01:07:30.680 --> 01:07:32.990 which reduces your chances in the race 01:07:32.990 --> 01:07:36.350 but increases your time spent with the rich kids 01:07:36.350 --> 01:07:38.360 compared to poor kids. 01:07:38.360 --> 01:07:41.900 Let me tell you a little bit more detail of the experiment. 01:07:41.900 --> 01:07:43.220 Stage 1 is randomization. 01:07:43.220 --> 01:07:47.810 So the people were randomized to sessions with varying stakes. 01:07:47.810 --> 01:07:53.630 There's 50 rupees, 20 rupees, 500 rupees per student 01:07:53.630 --> 01:07:55.340 for the winning team. 01:07:55.340 --> 01:07:58.300 This is a lot of money compared to students usual pocket money, 01:07:58.300 --> 01:07:58.800 right? 01:07:58.800 --> 01:08:02.450 So they would get something like $10 or 500 rupees 01:08:02.450 --> 01:08:04.690 is like one month's of pocket money. 01:08:04.690 --> 01:08:07.370 So that's really high stakes for these kids. 01:08:07.370 --> 01:08:09.140 The price is varied, essentially, 01:08:09.140 --> 01:08:10.760 because it lets us price out. 01:08:10.760 --> 01:08:12.770 It gives us a price of discrimination. 01:08:12.770 --> 01:08:15.620 Lets us understand how much are students willing to give up 01:08:15.620 --> 01:08:19.500 in order to not have to-- 01:08:19.500 --> 01:08:23.450 or in order to be able to socialize with the rich kids 01:08:23.450 --> 01:08:24.890 compared to the poor kid. 01:08:24.890 --> 01:08:28.250 There was a brief mixing to start with to judge 01:08:28.250 --> 01:08:30.170 socioeconomic status. 01:08:30.170 --> 01:08:32.810 So the kids were allowed to mingle a little bit. 01:08:32.810 --> 01:08:36.050 That would allow them to fairly easily understand 01:08:36.050 --> 01:08:39.109 who was a poor kid and who's a rich kid. 01:08:39.109 --> 01:08:40.069 OK. 01:08:40.069 --> 01:08:43.670 Stage number 2 was ability revelation and team selection. 01:08:43.670 --> 01:08:47.029 So you could essentially observe a two-person race. 01:08:47.029 --> 01:08:49.850 Usually, it's one poor and one rich students. 01:08:49.850 --> 01:08:51.283 Neither is from the old school. 01:08:51.283 --> 01:08:52.700 So these are not students that you 01:08:52.700 --> 01:08:56.540 would know from school anyway. 01:08:56.540 --> 01:08:59.149 But the uniforms make the school identifiable. 01:08:59.149 --> 01:09:01.819 You kind of know who is the rich kid and who's the poor kid. 01:09:01.819 --> 01:09:05.029 Now, then you can pick which of these two runners 01:09:05.029 --> 01:09:07.140 you want to have as your partners. 01:09:07.140 --> 01:09:10.939 So that is to say you see them run one by one. 01:09:10.939 --> 01:09:13.609 It's very easy to see who's faster and who's not. 01:09:13.609 --> 01:09:15.290 And so again, discrimination here 01:09:15.290 --> 01:09:18.600 is interpreted as picking the slow runner. 01:09:18.600 --> 01:09:22.189 So if you pick the slower runner, 01:09:22.189 --> 01:09:25.370 then it must be because you like something 01:09:25.370 --> 01:09:28.250 some other characteristics about that person more. 01:09:28.250 --> 01:09:29.840 The obvious characteristic here is 01:09:29.840 --> 01:09:34.380 that it's most likely going to be that kid is rich. 01:09:34.380 --> 01:09:35.250 OK. 01:09:35.250 --> 01:09:41.189 Then stages 3 and 4 are the choice implementation 01:09:41.189 --> 01:09:42.270 relay race. 01:09:42.270 --> 01:09:43.896 So students are randomly picked to have 01:09:43.896 --> 01:09:44.979 their choices implemented. 01:09:44.979 --> 01:09:46.260 So some of those choices-- 01:09:46.260 --> 01:09:48.029 this is, again, the strategy method. 01:09:48.029 --> 01:09:50.790 Some of these choices were actually randomly implemented. 01:09:50.790 --> 01:09:54.660 So there's plausible deniability for the students 01:09:54.660 --> 01:10:01.440 in the sense of you could just happen to be randomized 01:10:01.440 --> 01:10:05.520 or you happen to pick some students versus another. 01:10:05.520 --> 01:10:07.080 It could just be by chance that you 01:10:07.080 --> 01:10:09.058 are with one student versus another. 01:10:09.058 --> 01:10:11.100 So you could-- your freedom to choose essentially 01:10:11.100 --> 01:10:12.030 provide you cover. 01:10:12.030 --> 01:10:14.070 Sometimes, as we discussed before, 01:10:14.070 --> 01:10:16.380 the computer is choosing, so you always have an excuse. 01:10:16.380 --> 01:10:20.280 So it's intended to reveal student's true preferences as 01:10:20.280 --> 01:10:22.140 opposed to perhaps what they think 01:10:22.140 --> 01:10:26.190 that their friends want them to choose or the other runners. 01:10:26.190 --> 01:10:28.020 Then the relay races were actually held 01:10:28.020 --> 01:10:30.630 and prizes were distributed as promised. 01:10:30.630 --> 01:10:34.840 Number 4, crucially, there was a social interaction. 01:10:34.840 --> 01:10:37.770 So if you picked your teammate, you actually 01:10:37.770 --> 01:10:40.980 had to spend two hours of playing with a teammate. 01:10:40.980 --> 01:10:43.930 Board games, sports, playgrounds, and so on. 01:10:43.930 --> 01:10:45.690 Importantly, this was preannounce. 01:10:45.690 --> 01:10:49.680 So now, again, as I said before, when picking your partner, 01:10:49.680 --> 01:10:52.350 you have the choice between either picking 01:10:52.350 --> 01:10:55.230 the fast, poor kid, which will really 01:10:55.230 --> 01:10:57.960 increase your probability of winning, 01:10:57.960 --> 01:10:59.820 or the rich kid, who is kind of slow 01:10:59.820 --> 01:11:03.240 and will reduce your probability of winning. 01:11:03.240 --> 01:11:05.670 But if you pick that poor kid or the rich kid, 01:11:05.670 --> 01:11:08.850 you have to actually spend two hours with that teammate 01:11:08.850 --> 01:11:11.370 playing board games, sports, playground, and so on, 01:11:11.370 --> 01:11:16.540 and you might not want to do that with a poor kid. 01:11:16.540 --> 01:11:17.380 OK. 01:11:17.380 --> 01:11:21.310 So now, first, what is the demand for discrimination? 01:11:21.310 --> 01:11:22.660 You can look at this graph. 01:11:22.660 --> 01:11:25.330 It shows very nicely for the different prices. 01:11:25.330 --> 01:11:28.840 As I said, 500 rupees for winning the race, 200 rupees, 01:11:28.840 --> 01:11:30.340 or 500 rupees. 01:11:30.340 --> 01:11:32.440 If you look at 500 rupees, this is again 01:11:32.440 --> 01:11:34.120 one months of pocket money. 01:11:34.120 --> 01:11:36.430 There's no difference between treated and untreated 01:11:36.430 --> 01:11:37.190 classrooms. 01:11:37.190 --> 01:11:38.800 That is to say-- 01:11:38.800 --> 01:11:41.560 so there's 10% of people are, as you 01:11:41.560 --> 01:11:42.820 want, discriminating the poor. 01:11:42.820 --> 01:11:48.820 So this is less than 10% is about 7%, 8% of students 01:11:48.820 --> 01:11:55.120 pick the rich kids, even in the really, really high stakes 01:11:55.120 --> 01:11:55.870 race. 01:11:55.870 --> 01:11:58.360 That is to say, when the stakes are 500 rupees, 01:11:58.360 --> 01:12:00.970 there are, like, about 6%, 7%, 8% 01:12:00.970 --> 01:12:04.210 of students who still pick the rich kids, 01:12:04.210 --> 01:12:06.130 and they sort of take into account 01:12:06.130 --> 01:12:11.800 the chance that that might lose them the race. 01:12:11.800 --> 01:12:16.450 But they don't have to spend two hours with a poor kid then 01:12:16.450 --> 01:12:17.710 socializing. 01:12:17.710 --> 01:12:20.560 Now, when you look at the lower prices, the fraction, 01:12:20.560 --> 01:12:23.020 as you expect-- this is the red line. 01:12:23.020 --> 01:12:25.960 Sorry, this is the green line, the upper line. 01:12:25.960 --> 01:12:28.270 The fraction who choose the rich kids, 01:12:28.270 --> 01:12:29.770 the fraction who were discriminating 01:12:29.770 --> 01:12:33.360 against the poor, increases as you expect. 01:12:33.360 --> 01:12:35.200 So now, it becomes cheaper. 01:12:35.200 --> 01:12:37.990 The race is only 200 rupees or 50 rupees. 01:12:37.990 --> 01:12:41.110 You might be more inclined to pick the rich kid because you 01:12:41.110 --> 01:12:43.510 know the value of socializing stays the same, 01:12:43.510 --> 01:12:45.520 but the costs of picking the rich kid, 01:12:45.520 --> 01:12:47.680 the cost of losing the race, potentially 01:12:47.680 --> 01:12:51.760 at least is reduced. 01:12:51.760 --> 01:12:53.620 If you look at 50 rupees, the price 01:12:53.620 --> 01:12:59.890 of 50 rupees for the game, for the relay race, that's 01:12:59.890 --> 01:13:03.460 about almost 40% of students now pick the rich kid, 01:13:03.460 --> 01:13:05.650 even though that might lose them the race. 01:13:05.650 --> 01:13:09.400 And now, crucially, in red, the dashed line 01:13:09.400 --> 01:13:11.920 below, you can see the treated classrooms. 01:13:11.920 --> 01:13:14.860 And what he finds is that for 500 rupees, when 01:13:14.860 --> 01:13:17.518 the stakes are really, really high, there's no effect. 01:13:17.518 --> 01:13:19.060 Essentially, it doesn't really matter 01:13:19.060 --> 01:13:21.393 whether you are in a treated or a not treated classroom, 01:13:21.393 --> 01:13:25.060 in part perhaps because there's not much room for going lower 01:13:25.060 --> 01:13:25.940 than that. 01:13:25.940 --> 01:13:29.382 So essentially, there's no effect on there 01:13:29.382 --> 01:13:31.840 because you know the stakes are really, really high anyway. 01:13:31.840 --> 01:13:33.460 There's not much discrimination. 01:13:33.460 --> 01:13:35.555 And having had a treated-- 01:13:35.555 --> 01:13:37.180 having had a poor kid in your classroom 01:13:37.180 --> 01:13:39.220 doesn't really change that. 01:13:39.220 --> 01:13:42.820 But then, very clearly, for 250 rupees and 50 rupees, the lower 01:13:42.820 --> 01:13:45.670 stakes, there's a clear difference 01:13:45.670 --> 01:13:48.050 between the green and the red lines. 01:13:48.050 --> 01:13:51.400 That is to say, there's a lot less discrimination 01:13:51.400 --> 01:13:52.780 towards the poor kids. 01:13:52.780 --> 01:13:54.970 Poor kids are a lot more likely to be 01:13:54.970 --> 01:14:00.430 chosen when a student has-- in treated classrooms 01:14:00.430 --> 01:14:03.070 when somebody had a poor kid, another poor kid 01:14:03.070 --> 01:14:05.480 in that classroom for several years. 01:14:05.480 --> 01:14:07.840 So that's to say there's-- 01:14:07.840 --> 01:14:09.880 being exposed to these poor children 01:14:09.880 --> 01:14:14.740 reduces discrimination among the poor among the rich students 01:14:14.740 --> 01:14:15.520 subsequently. 01:14:18.260 --> 01:14:20.723 This is sort of the same graph that I showed you before. 01:14:20.723 --> 01:14:23.140 Again, that's sort of consistent with what we have before. 01:14:23.140 --> 01:14:26.170 It gets a little messier than we had seen previously. 01:14:26.170 --> 01:14:29.440 But similarly, even within classrooms, 01:14:29.440 --> 01:14:33.520 if you look at grades 2 and 3 versus 4 and 5, 01:14:33.520 --> 01:14:36.250 the effects seem to be concentrated more 01:14:36.250 --> 01:14:40.680 pronounced in grades 2 and 3. 01:14:40.680 --> 01:14:46.290 Now, finally, as I said, we look at test scores and discipline. 01:14:46.290 --> 01:14:48.390 So arguably, there are some positive effects 01:14:48.390 --> 01:14:49.590 on social preferences. 01:14:49.590 --> 01:14:53.040 And as I said before, the policy question now is, 01:14:53.040 --> 01:14:55.950 does that come at the cost of academic achievement 01:14:55.950 --> 01:14:57.070 in some ways? 01:14:57.070 --> 01:14:59.520 So is it that the rich kid now, by being exposed 01:14:59.520 --> 01:15:02.010 to the poor kids, may be somewhat nicer and more 01:15:02.010 --> 01:15:05.460 friendly and less discriminating against the poor? 01:15:05.460 --> 01:15:06.330 That's all and good. 01:15:06.330 --> 01:15:09.960 But is it the case now that test scores go down? 01:15:09.960 --> 01:15:14.250 So Gautam finds no effect on aggregate test score index 01:15:14.250 --> 01:15:16.440 or zero effect in Hindi and math. 01:15:16.440 --> 01:15:22.230 There's a little bit of a reduction in English scores 01:15:22.230 --> 01:15:25.320 of 0.9 standard deviations. 01:15:25.320 --> 01:15:27.480 That's marginally significant. 01:15:27.480 --> 01:15:34.810 That's suggestive but not perhaps particularly large. 01:15:34.810 --> 01:15:36.807 So these effects are not particularly large. 01:15:36.807 --> 01:15:39.390 And in particular, in aggregate, so if you agree to everything 01:15:39.390 --> 01:15:43.350 together, there don't seem to be any significant effects. 01:15:43.350 --> 01:15:47.550 So perhaps the English scores are suggestive but sort of not 01:15:47.550 --> 01:15:48.870 particularly large. 01:15:48.870 --> 01:15:51.000 There's some mild effect on discipline. 01:15:51.000 --> 01:15:52.980 Interestingly, there's an increase in swearing. 01:15:52.980 --> 01:15:54.188 You think that's good or bad. 01:15:54.188 --> 01:15:55.260 You can think about that. 01:15:55.260 --> 01:15:58.710 But there seems to be a little bit of an effect in terms 01:15:58.710 --> 01:16:01.620 of language uses. 01:16:01.620 --> 01:16:04.020 There's no effect on violent and disruptive behavior, 01:16:04.020 --> 01:16:11.200 which you might think is a lot more damaging, potentially. 01:16:11.200 --> 01:16:11.860 OK. 01:16:11.860 --> 01:16:14.230 So just summarizing, what does the paper find? 01:16:14.230 --> 01:16:17.260 Well, having poor classmates makes wealthy students 01:16:17.260 --> 01:16:19.180 more prosocial and generous. 01:16:19.180 --> 01:16:22.385 They're more likely to volunteer for charities. 01:16:22.385 --> 01:16:24.760 I didn't show you that evidence, but that's another piece 01:16:24.760 --> 01:16:26.110 of evidence that he finds. 01:16:26.110 --> 01:16:29.500 They're more likely to give in money in dictator games 01:16:29.500 --> 01:16:32.500 to give them more higher fractions of their shares 01:16:32.500 --> 01:16:33.820 in dictator games. 01:16:33.820 --> 01:16:37.940 They also choose more equitable outcomes 01:16:37.940 --> 01:16:41.890 in sort of disinterested third party games 01:16:41.890 --> 01:16:45.730 where essentially, you choose between two other students 01:16:45.730 --> 01:16:47.073 and their allocation. 01:16:47.073 --> 01:16:48.490 So it's not just that they're more 01:16:48.490 --> 01:16:51.310 likely to be willing to give up money that others get, 01:16:51.310 --> 01:16:55.390 but also, they're more likely to choose equal allocations 01:16:55.390 --> 01:16:57.845 in third party games and disinterested games. 01:16:57.845 --> 01:16:59.470 Again, I didn't show you that evidence. 01:16:59.470 --> 01:17:01.870 But it seems to be that what's increasing here 01:17:01.870 --> 01:17:04.780 is sort of like inequality aversion 01:17:04.780 --> 01:17:11.290 in these sort of disinterested dictator or the types of games, 01:17:11.290 --> 01:17:14.080 where essentially, these students, by being exposed 01:17:14.080 --> 01:17:19.420 to poor kids, are now more averse against unequal outcomes 01:17:19.420 --> 01:17:21.130 in these types of games. 01:17:21.130 --> 01:17:23.710 Second, there is less discrimination 01:17:23.710 --> 01:17:27.640 and more higher willingness to socialize with the poor. 01:17:27.640 --> 01:17:31.300 They're more likely to choose poor teammates 01:17:31.300 --> 01:17:32.830 in sports contests. 01:17:32.830 --> 01:17:34.810 They're also more willing to attend playdates 01:17:34.810 --> 01:17:35.650 with poor children. 01:17:35.650 --> 01:17:38.200 Again, I didn't show you that evidence. 01:17:38.200 --> 01:17:40.600 And then there's some small, negative effect 01:17:40.600 --> 01:17:46.960 on academic outcomes that I think are mostly negligible. 01:17:46.960 --> 01:17:50.230 Now, there's other work on the contact hypothesis 01:17:50.230 --> 01:17:50.980 that I showed you. 01:17:50.980 --> 01:17:53.470 So the contact-- what is the contact hypothesis? 01:17:53.470 --> 01:17:57.190 The contact hypothesis goes back to at least Allport 01:17:57.190 --> 01:18:01.600 in 1954, which is the idea that interpersonal contacts 01:18:01.600 --> 01:18:05.020 reduces prejudice until certain conditions. 01:18:05.020 --> 01:18:08.290 And not just prejudice, but also changes 01:18:08.290 --> 01:18:11.440 attitudes and social preferences potentially. 01:18:11.440 --> 01:18:14.290 So Matt Lowe, who was a PhD student here at MIT, 01:18:14.290 --> 01:18:17.680 has a very nice paper that looks at cricket tournaments 01:18:17.680 --> 01:18:20.950 and asks the question whether cricket leagues in India 01:18:20.950 --> 01:18:25.900 can increase cross-class interaction in pro-sociology. 01:18:25.900 --> 01:18:28.810 So what he does is he randomizes, essentially, 01:18:28.810 --> 01:18:31.300 cricket leagues and teams in cricket leagues 01:18:31.300 --> 01:18:33.370 where people across different castes 01:18:33.370 --> 01:18:36.710 are now more or less likely to play with each other, 01:18:36.710 --> 01:18:39.790 both in terms of-- he varies or there's variation 01:18:39.790 --> 01:18:42.010 in the study within teams. 01:18:42.010 --> 01:18:44.080 So are you more or less likely to have 01:18:44.080 --> 01:18:46.780 a-- or some people are more or less likely to have 01:18:46.780 --> 01:18:48.640 a lower or higher caste. 01:18:48.640 --> 01:18:51.550 So a person from a different cast in their team. 01:18:51.550 --> 01:18:55.870 And he has variation in what he calls adversarial contact, 01:18:55.870 --> 01:19:00.850 which is they're more or less likely to be exposed to players 01:19:00.850 --> 01:19:04.450 from other teams in higher or different castes 01:19:04.450 --> 01:19:05.840 from themselves. 01:19:05.840 --> 01:19:08.770 So if you're on the team and have a person 01:19:08.770 --> 01:19:11.500 from a different caste on your team, you're on the same team. 01:19:11.500 --> 01:19:15.170 You share the same objective, and you want to win together. 01:19:15.170 --> 01:19:17.410 So now, having somebody from a different caste 01:19:17.410 --> 01:19:20.230 or just somebody who's different in various ways in your team 01:19:20.230 --> 01:19:23.620 might make you like them better. 01:19:23.620 --> 01:19:26.500 You might be sort of more positive about them. 01:19:26.500 --> 01:19:28.900 You might be sort of more likely to talk to them. 01:19:28.900 --> 01:19:30.233 You might learn about them. 01:19:30.233 --> 01:19:31.900 You might sort of see some sides in them 01:19:31.900 --> 01:19:33.245 that you hadn't seen before. 01:19:33.245 --> 01:19:35.620 So you might be more like they sort of empathize and look 01:19:35.620 --> 01:19:37.300 at sort of nice characteristic of them 01:19:37.300 --> 01:19:39.790 and sort of update them positively about people 01:19:39.790 --> 01:19:42.910 from other tasks, and that changes your attitudes 01:19:42.910 --> 01:19:45.340 towards them in general. 01:19:45.340 --> 01:19:48.737 If, however, you play against somebody from different castes, 01:19:48.737 --> 01:19:51.070 you might, actually-- that might, actually, if anything, 01:19:51.070 --> 01:19:54.790 backfire, because you really don't like your opponents. 01:19:54.790 --> 01:19:56.260 You might see them very negatively. 01:19:56.260 --> 01:19:57.802 You might be aggressive towards them. 01:19:57.802 --> 01:19:59.620 You might be unfriendly towards them. 01:19:59.620 --> 01:20:01.660 You might sort of not like that they 01:20:01.660 --> 01:20:03.320 win against you or the like. 01:20:03.320 --> 01:20:06.190 So I'll just add, these adversarial interactions 01:20:06.190 --> 01:20:07.990 might actually backfire in the sense 01:20:07.990 --> 01:20:09.760 that they might not foster integration 01:20:09.760 --> 01:20:14.900 but actually sort of make things worse. 01:20:14.900 --> 01:20:21.490 So Matt runs this experiment and finds evidence of increased 01:20:21.490 --> 01:20:22.820 cross-caste interactions. 01:20:22.820 --> 01:20:25.810 So people are more likely to be friends, more likely 01:20:25.810 --> 01:20:27.280 to hang out with others. 01:20:27.280 --> 01:20:30.850 They also are more generous in dictator and other types 01:20:30.850 --> 01:20:31.870 of games. 01:20:31.870 --> 01:20:35.980 They also are more likely to engage in trade 01:20:35.980 --> 01:20:37.630 or in economic exchange. 01:20:37.630 --> 01:20:40.390 So what Matt does is he sort of randomizes gloves. 01:20:40.390 --> 01:20:42.490 It's like left gloves and right gloves. 01:20:42.490 --> 01:20:43.990 And he does the same for flip-flops. 01:20:43.990 --> 01:20:46.360 Left flip-flops and right flip-flops. 01:20:46.360 --> 01:20:49.120 And people are more likely to trade with somebody 01:20:49.120 --> 01:20:53.300 from another caste if they ever been on the same team 01:20:53.300 --> 01:20:54.550 with people from other castes. 01:20:54.550 --> 01:20:56.217 If they have a higher fraction of people 01:20:56.217 --> 01:21:00.610 on their team of people from other castes, 01:21:00.610 --> 01:21:03.400 they're more likely to engage in all these behaviors. 01:21:03.400 --> 01:21:05.710 They're more likely to have cross-class interaction. 01:21:05.710 --> 01:21:07.690 More likely to be prosocial. 01:21:07.690 --> 01:21:10.060 More likely to engage in economic exchange 01:21:10.060 --> 01:21:13.110 with people from other castes. 01:21:13.110 --> 01:21:17.270 So that's all true for collaborative contact. 01:21:17.270 --> 01:21:21.470 That is to say, that's contact with people on the same team. 01:21:21.470 --> 01:21:24.470 In contrast, for adversarial interactions, when people 01:21:24.470 --> 01:21:28.460 are on the opposite team, having more people from other castes 01:21:28.460 --> 01:21:31.730 or being exposed to more people from other castes 01:21:31.730 --> 01:21:33.860 does not have these positive effects 01:21:33.860 --> 01:21:36.630 and, for some of these outcomes, have even negative effects. 01:21:36.630 --> 01:21:38.660 So if anything, that sort of backfires. 01:21:38.660 --> 01:21:42.050 It doesn't-- just-- a mere exposure to others, 01:21:42.050 --> 01:21:45.320 if you are sort of in an adversarial contact situation, 01:21:45.320 --> 01:21:49.580 does not really foster prosociology or any of these 01:21:49.580 --> 01:21:50.930 types of integration. 01:21:50.930 --> 01:21:52.970 If anything, it backfires. 01:21:52.970 --> 01:21:54.188 Now, why is that important? 01:21:54.188 --> 01:21:55.730 If you think about like, for example, 01:21:55.730 --> 01:21:58.280 attitudes towards immigrants, it really 01:21:58.280 --> 01:22:01.670 matters hugely what types of contacts people are exposed to. 01:22:01.670 --> 01:22:04.880 If people have worked together in the same team, if they have, 01:22:04.880 --> 01:22:09.440 perhaps, team pay, if they were to work towards the same goal, 01:22:09.440 --> 01:22:11.990 really, it seems this evidence suggests 01:22:11.990 --> 01:22:15.260 can foster prosociology and so on and so forth 01:22:15.260 --> 01:22:18.590 that leads to integration, reduce discrimination, 01:22:18.590 --> 01:22:19.980 and so on and so forth. 01:22:19.980 --> 01:22:22.010 So it's sort of the incentives are aligned 01:22:22.010 --> 01:22:25.010 or if you could sort of set up incentives that are aligned, 01:22:25.010 --> 01:22:26.990 people might become nicer to each other 01:22:26.990 --> 01:22:31.040 and interactions might be fostered. 01:22:31.040 --> 01:22:34.248 In contrast, if contact is adversarial, 01:22:34.248 --> 01:22:36.290 if you're worried that immigrants are taking away 01:22:36.290 --> 01:22:39.410 your jobs, being exposed to immigrants 01:22:39.410 --> 01:22:41.240 might just do the opposite. 01:22:41.240 --> 01:22:43.010 So you might see a lot of immigrants. 01:22:43.010 --> 01:22:44.840 But in a way, if you feel like you're 01:22:44.840 --> 01:22:46.910 in computation with them, if they're adversarial, 01:22:46.910 --> 01:22:49.010 if they're sort of your enemies in some ways 01:22:49.010 --> 01:22:54.440 or your opponents in some computation for a job, 01:22:54.440 --> 01:22:59.300 being exposed to them might things, in fact, if anything, 01:22:59.300 --> 01:23:01.650 worse. 01:23:01.650 --> 01:23:04.790 Finally, there's another piece of very nice evidence 01:23:04.790 --> 01:23:06.470 by Corno et al. 01:23:06.470 --> 01:23:09.710 This paper is considering the impact 01:23:09.710 --> 01:23:11.750 of random interracial interactions 01:23:11.750 --> 01:23:16.850 among college roommates in South Africa on stereotypes, 01:23:16.850 --> 01:23:19.470 attitudes, and performance. 01:23:19.470 --> 01:23:21.230 So what they look at, essentially, 01:23:21.230 --> 01:23:26.820 is roommates of different race reduces-- 01:23:26.820 --> 01:23:28.460 so these are black and white students-- 01:23:28.460 --> 01:23:31.880 reduces white students stereotypes towards blacks 01:23:31.880 --> 01:23:35.360 and increases interracial friendships. 01:23:35.360 --> 01:23:39.410 It also improves grades and lowers dropouts among blacks. 01:23:39.410 --> 01:23:42.740 So there's sort of a number of positive effects 01:23:42.740 --> 01:23:46.670 of a very simple policy of increasing 01:23:46.670 --> 01:23:48.770 contacts among roommates. 01:23:48.770 --> 01:23:50.783 And again, if you think about this evidence, 01:23:50.783 --> 01:23:53.450 it seems to suggest that perhaps that's kind of like if you have 01:23:53.450 --> 01:23:56.090 a roommate that you're going to live with for quite 01:23:56.090 --> 01:23:58.978 a while, that feels a lot like collaborative contact 01:23:58.978 --> 01:24:00.770 in the sense of, like, you're sort of stuck 01:24:00.770 --> 01:24:02.312 with that roommate for quite a while, 01:24:02.312 --> 01:24:05.030 so you might as well sort make the best out of it, 01:24:05.030 --> 01:24:07.103 even if you initially don't like that person 01:24:07.103 --> 01:24:09.770 and you have the same objectives of being happy together, living 01:24:09.770 --> 01:24:10.460 together. 01:24:10.460 --> 01:24:12.530 And that really seems to foster and have 01:24:12.530 --> 01:24:14.570 some positive benefits. 01:24:17.680 --> 01:24:20.760 So taking together, that sort of says, so A, 01:24:20.760 --> 01:24:24.270 the contact hypothesis seems to be broadly right. 01:24:24.270 --> 01:24:27.480 Contact to people who are different from you 01:24:27.480 --> 01:24:31.195 might make you more tolerant or more prosocial towards them. 01:24:31.195 --> 01:24:32.820 There might be more sort of integration 01:24:32.820 --> 01:24:34.180 of different groups. 01:24:34.180 --> 01:24:36.180 But what really seems to matter quite a bit 01:24:36.180 --> 01:24:40.890 is the type of contact that people are exposed to. 01:24:40.890 --> 01:24:43.170 So finally and relatively quickly, 01:24:43.170 --> 01:24:45.840 I'll tell you a little bit about whether people underestimate 01:24:45.840 --> 01:24:47.850 the benefits of prosociology. 01:24:47.850 --> 01:24:50.910 I should mention that problem set 01:24:50.910 --> 01:24:54.960 3 question 2 is, in fact, trying to ask you to do something 01:24:54.960 --> 01:24:56.220 related to that. 01:24:56.220 --> 01:24:58.470 So you might want to actually do the problems at first 01:24:58.470 --> 01:25:00.280 before you finish this lecture. 01:25:00.280 --> 01:25:02.280 At least sort of-- sorry, not the entire problem 01:25:02.280 --> 01:25:05.670 set, but question 2 of problems set 3. 01:25:05.670 --> 01:25:09.810 That will not take you very long to do so, 01:25:09.810 --> 01:25:11.310 and it's more like a fun exercise 01:25:11.310 --> 01:25:15.460 that I thought would be nice for you to engage in. 01:25:15.460 --> 01:25:20.670 But anyway, this is a very nice paper by Kumar and Epley. 01:25:20.670 --> 01:25:22.200 That's a typo here. 01:25:22.200 --> 01:25:25.230 It should say Kumar and Epley. 01:25:25.230 --> 01:25:27.660 And the officer is asked the question about, 01:25:27.660 --> 01:25:32.110 do we have correct beliefs about the impacts of generosity? 01:25:32.110 --> 01:25:34.260 And so what's the underlying reason here is, 01:25:34.260 --> 01:25:37.710 well, many prosocial acts require estimating the impacts 01:25:37.710 --> 01:25:39.150 on the recipient. 01:25:39.150 --> 01:25:41.400 If you give money to somebody in Africa, 01:25:41.400 --> 01:25:46.800 if you help anybody and people in need, if you donate money 01:25:46.800 --> 01:25:50.860 in general, if you write letters of gratitude 01:25:50.860 --> 01:25:53.100 or if you do random acts of kindness, 01:25:53.100 --> 01:25:55.970 it requires some sort of estimation of, 01:25:55.970 --> 01:26:01.140 how is the other person going to feel if they receive-- 01:26:01.140 --> 01:26:04.380 if they are on the receiving end of this prosocial act? 01:26:04.380 --> 01:26:06.750 Now, what Epley and Kumar argue is 01:26:06.750 --> 01:26:10.710 people are subject to egocentric bias, which 01:26:10.710 --> 01:26:13.290 may lead them to systematically underestimate 01:26:13.290 --> 01:26:15.390 the positive impact of prosociology. 01:26:15.390 --> 01:26:18.960 In this case, gratitude letters. 01:26:18.960 --> 01:26:20.040 And so why is that? 01:26:20.040 --> 01:26:23.130 Well, it's because predicting others mental states is 01:26:23.130 --> 01:26:27.090 difficult. It's really hard for people 01:26:27.090 --> 01:26:29.640 to understand how another person might feel. 01:26:29.640 --> 01:26:31.710 Usually, people would sort think about themselves 01:26:31.710 --> 01:26:35.590 and sort think about how would I feel, and what would things be, 01:26:35.590 --> 01:26:36.690 and how are things like. 01:26:36.690 --> 01:26:42.100 And it's very hard to understand how others might react to you. 01:26:42.100 --> 01:26:43.890 It sort of requires perspective taking, 01:26:43.890 --> 01:26:48.300 and that's sometimes tricky for people to do. 01:26:48.300 --> 01:26:50.790 So this is again a very nice experiment 01:26:50.790 --> 01:26:54.390 by Kumar and Nicholas Epley. 01:26:54.390 --> 01:26:57.300 Not Nicholas last name, but Nicholas Epley. 01:26:57.300 --> 01:27:00.630 And they test whether people misunderstand the consequences 01:27:00.630 --> 01:27:02.530 of showing appreciation. 01:27:02.530 --> 01:27:04.410 And so what are these experiments? 01:27:04.410 --> 01:27:06.510 There's a series of experiments that look at this. 01:27:06.510 --> 01:27:07.830 What do they look like? 01:27:07.830 --> 01:27:10.920 But what they do is they ask people, 01:27:10.920 --> 01:27:14.280 MBA students or subjects in experiments, 01:27:14.280 --> 01:27:15.900 to pick a prosocial act. 01:27:15.900 --> 01:27:18.990 That's writing a gratitude letter, for example. 01:27:18.990 --> 01:27:22.603 And then ask how the giver and the recipient will be affected. 01:27:22.603 --> 01:27:25.020 In some cases, it's only the recipient, and in some cases, 01:27:25.020 --> 01:27:27.300 also the giver. 01:27:27.300 --> 01:27:30.580 Then they perform the prosocial act by, for example, 01:27:30.580 --> 01:27:32.350 writing a letter of gratitude. 01:27:32.350 --> 01:27:34.290 And then assess how the giver under recipients 01:27:34.290 --> 01:27:36.850 were actually affected by doing that. 01:27:36.850 --> 01:27:38.970 And then you can compare that to item number 2. 01:27:38.970 --> 01:27:42.750 To the estimation x ante. 01:27:42.750 --> 01:27:44.130 Now, what do they find? 01:27:44.130 --> 01:27:47.548 They find clear evidence that when you look at, 01:27:47.548 --> 01:27:49.590 like, in particular, when you look at the giver-- 01:27:49.590 --> 01:27:53.670 so these graphs all show kind of the predicted ratings. 01:27:53.670 --> 01:27:59.700 The ratings is between 0 and 10 about what's the experience? 01:27:59.700 --> 01:28:02.070 How happy is the other person? 01:28:02.070 --> 01:28:05.820 What is the surprise about receiving this letter 01:28:05.820 --> 01:28:07.630 of gratitude in this example? 01:28:07.630 --> 01:28:10.200 So people are more surprised than predicted. 01:28:10.200 --> 01:28:12.300 They are also more surprised about the content 01:28:12.300 --> 01:28:15.180 than predicted. 01:28:15.180 --> 01:28:17.550 They're also-- the recipients mood 01:28:17.550 --> 01:28:19.830 is actually better than predicted. 01:28:19.830 --> 01:28:23.670 And the awkwardness is also better than predicted. 01:28:23.670 --> 01:28:26.760 So when you ask people to write letters of gratitude, 01:28:26.760 --> 01:28:29.640 people tend to say, oh, you know, 01:28:29.640 --> 01:28:33.810 it's going to be tedious to do, and the other person 01:28:33.810 --> 01:28:35.550 might feel it's awkward. 01:28:35.550 --> 01:28:37.210 And what am I going to say? 01:28:37.210 --> 01:28:39.330 And is it going to be weird when I'm going to say? 01:28:39.330 --> 01:28:41.850 Is the person-- don't they know already 01:28:41.850 --> 01:28:45.360 anyway that I'm really appreciate of that person? 01:28:45.360 --> 01:28:46.330 And so on. 01:28:46.330 --> 01:28:49.830 So people come up with all sorts of reasons why they might not 01:28:49.830 --> 01:28:50.910 want to do that. 01:28:50.910 --> 01:28:53.580 If you sort of introspect and ask yourself how many letters 01:28:53.580 --> 01:28:57.930 of gratitude have you written in the last year, that doesn't-- 01:28:57.930 --> 01:29:00.510 few people actually do that in practice. 01:29:00.510 --> 01:29:03.390 And perhaps some of these reasons like the perceived 01:29:03.390 --> 01:29:05.910 awkwardness or perhaps sort of the underestimation 01:29:05.910 --> 01:29:09.973 of the recipient's mood might contribute to that. 01:29:09.973 --> 01:29:11.640 Now, if you also think for a little bit, 01:29:11.640 --> 01:29:15.210 like, how would one actually feel to receive these letters, 01:29:15.210 --> 01:29:17.005 it's pretty obvious once you think about it 01:29:17.005 --> 01:29:19.380 that most people are actually quite happy about receiving 01:29:19.380 --> 01:29:19.980 such letters. 01:29:19.980 --> 01:29:21.930 It's kind of nice if somebody tells you, look. 01:29:21.930 --> 01:29:25.200 You did something really nice for me some time 01:29:25.200 --> 01:29:28.560 a long time ago, and this helped me a lot in my career 01:29:28.560 --> 01:29:32.700 in whatever way, or getting into school, into college 01:29:32.700 --> 01:29:33.900 or whatever it might be. 01:29:33.900 --> 01:29:35.880 It's really nice to sort of hear from somebody 01:29:35.880 --> 01:29:38.190 that you did something nice in their life 01:29:38.190 --> 01:29:42.270 that people are quite happy about. 01:29:42.270 --> 01:29:46.830 Now, what's sort of some evidence, some summary of that? 01:29:46.830 --> 01:29:49.230 Well, so we tend to systematically underestimate 01:29:49.230 --> 01:29:53.010 other's appreciated and expressions of gratitude. 01:29:53.010 --> 01:29:56.220 That's also true for some other types of effect. 01:29:56.220 --> 01:29:59.640 That tends to be also true for random acts of kindness. 01:29:59.640 --> 01:30:04.050 People tend to be surprisingly more happy about those 01:30:04.050 --> 01:30:05.520 compared to predicted. 01:30:05.520 --> 01:30:07.770 It's also true-- so Epley and Schroeder 01:30:07.770 --> 01:30:10.360 argue for social connections. 01:30:10.360 --> 01:30:13.200 So people also underestimate on average 01:30:13.200 --> 01:30:14.880 how they themselves and others feel 01:30:14.880 --> 01:30:16.047 when starting conversations. 01:30:16.047 --> 01:30:18.570 So what this experiment does is this gets people to-- it 01:30:18.570 --> 01:30:22.620 randomized people to start conversations during commuting 01:30:22.620 --> 01:30:24.190 on buses or trains. 01:30:24.190 --> 01:30:26.430 And once people start doing that-- again, 01:30:26.430 --> 01:30:29.640 before, when you ask them, like, how is the other person going 01:30:29.640 --> 01:30:31.140 to feel, how are you going to feel, 01:30:31.140 --> 01:30:34.920 and so on, people will say, well, it's going to be awkward. 01:30:34.920 --> 01:30:36.540 And what are we going to talk about? 01:30:36.540 --> 01:30:38.207 And is the other person even interested? 01:30:38.207 --> 01:30:39.348 And so on. 01:30:39.348 --> 01:30:40.890 But when you actually do that, people 01:30:40.890 --> 01:30:43.260 seem to be quite happy to have started conversations 01:30:43.260 --> 01:30:45.600 and making human connection. 01:30:45.600 --> 01:30:47.195 To be clear, not everybody is happy. 01:30:47.195 --> 01:30:48.570 Some people might be also grumpy. 01:30:48.570 --> 01:30:51.360 But the vast majority, at least in these types of study, 01:30:51.360 --> 01:30:54.480 seem to be quite happy about initiating 01:30:54.480 --> 01:30:57.180 social contact, about expressions, 01:30:57.180 --> 01:31:00.750 letters of gratitude, or things like random acts of kindness. 01:31:00.750 --> 01:31:02.250 It's really nice if somebody happens 01:31:02.250 --> 01:31:04.830 to be something to do something nice towards you. 01:31:04.830 --> 01:31:07.980 Could be like a random person who you've never seen before 01:31:07.980 --> 01:31:09.030 and you never see again. 01:31:09.030 --> 01:31:11.530 Just some random person on the street is really nice to you. 01:31:11.530 --> 01:31:12.970 That might make your day. 01:31:12.970 --> 01:31:15.690 It could be also somebody who quite well, 01:31:15.690 --> 01:31:18.990 and he's a good friend, and who just 01:31:18.990 --> 01:31:20.910 wants to do something nice for you 01:31:20.910 --> 01:31:23.080 for no good, apparent reasons. 01:31:23.080 --> 01:31:25.230 So one caveat to these kinds of experiments 01:31:25.230 --> 01:31:28.265 is these are very much like one-short, short 01:31:28.265 --> 01:31:28.890 interreactions. 01:31:28.890 --> 01:31:31.200 These are one-time interactions, and then the effects 01:31:31.200 --> 01:31:32.080 are measured. 01:31:32.080 --> 01:31:33.780 So there are some questions about, 01:31:33.780 --> 01:31:36.430 do these effects persist for repeat interactions? 01:31:36.430 --> 01:31:40.190 That is to say, like, maybe if you do that once, people 01:31:40.190 --> 01:31:41.340 are quite happy. 01:31:41.340 --> 01:31:44.320 But when you try it more often, these effects tend to go away. 01:31:44.320 --> 01:31:46.820 So there's a bit of a question kind of like, how persistent, 01:31:46.820 --> 01:31:48.450 how important are these in practice 01:31:48.450 --> 01:31:52.160 if you do that more often in particular in the long run? 01:31:52.160 --> 01:31:55.010 And then another important question 01:31:55.010 --> 01:31:57.380 here is then the question about always 01:31:57.380 --> 01:32:00.960 under-investing in prosociology in terms of our behavior. 01:32:00.960 --> 01:32:04.760 So lots of people who tend to be quite selfish 01:32:04.760 --> 01:32:07.220 and do good things for themselves and not so much 01:32:07.220 --> 01:32:08.288 others-- 01:32:08.288 --> 01:32:09.830 presumably, they do that because they 01:32:09.830 --> 01:32:11.930 want to make themselves happy. 01:32:11.930 --> 01:32:14.600 Presumably, people who do nice things towards others 01:32:14.600 --> 01:32:17.120 to some degree do that because they want 01:32:17.120 --> 01:32:18.830 to make the other person happy. 01:32:18.830 --> 01:32:23.960 In part perhaps because of social image 01:32:23.960 --> 01:32:26.130 and self-image concerns. 01:32:26.130 --> 01:32:30.170 But one important hypothesis that Epley and others raises 01:32:30.170 --> 01:32:34.640 the question of, like, well, are we underestimating how good it 01:32:34.640 --> 01:32:37.070 might not only make others feel but also 01:32:37.070 --> 01:32:39.350 ourselves feel from being nice? 01:32:39.350 --> 01:32:43.190 That is to say, perhaps, one easy way of making ourselves 01:32:43.190 --> 01:32:47.840 happy is not just by being selfish and maximizing whatever 01:32:47.840 --> 01:32:51.980 outcomes but really, being nice towards others, in part 01:32:51.980 --> 01:32:54.170 perhaps because it just makes us happy to see 01:32:54.170 --> 01:32:55.400 when others are happy. 01:32:55.400 --> 01:32:57.770 Perhaps it makes us happy because others 01:32:57.770 --> 01:32:59.600 will be then nicer to us. 01:32:59.600 --> 01:33:01.100 And there's a question about, can we 01:33:01.100 --> 01:33:04.070 make others and ourselves happier 01:33:04.070 --> 01:33:08.030 by being more prosocial, perhaps because we 01:33:08.030 --> 01:33:13.113 underestimate to start with what these effects might be? 01:33:13.113 --> 01:33:14.780 I don't think there's that much evidence 01:33:14.780 --> 01:33:17.430 on this specific question, but I'd love to learn more. 01:33:17.430 --> 01:33:19.170 And when you think about your own life, 01:33:19.170 --> 01:33:21.860 you might want to experiment for a while 01:33:21.860 --> 01:33:24.800 and seeing trying to be nice or trying 01:33:24.800 --> 01:33:28.910 to engage in random acts of kindness, letters of gratitude 01:33:28.910 --> 01:33:31.760 and so on, that might be a nice habit to acquire. 01:33:31.760 --> 01:33:35.990 And you might see it may or might not make you happier. 01:33:35.990 --> 01:33:39.230 Surely, it will make the other person on the receiving end 01:33:39.230 --> 01:33:40.710 happier. 01:33:40.710 --> 01:33:41.210 OK. 01:33:41.210 --> 01:33:44.090 So let me sort of summarize what we learned 01:33:44.090 --> 01:33:46.220 or what we studied on social preferences. 01:33:46.220 --> 01:33:49.940 So first, others' outcomes and utility matter for people's 01:33:49.940 --> 01:33:51.960 choices quite a bit. 01:33:51.960 --> 01:33:54.380 So in various situations, essentially, people 01:33:54.380 --> 01:33:56.600 are willing to give to others, and they're influenced 01:33:56.600 --> 01:33:58.760 by others in their choices. 01:33:58.760 --> 01:34:01.010 Now, upon closer look, there's not 01:34:01.010 --> 01:34:02.900 much evidence of pure altruism. 01:34:02.900 --> 01:34:05.810 Rarely does it seem that people just do stuff for others 01:34:05.810 --> 01:34:08.520 just for the sake of others doing well. 01:34:08.520 --> 01:34:11.990 That is to say, if people get the chance of hiding 01:34:11.990 --> 01:34:14.630 they're not so nice actions, they will often 01:34:14.630 --> 01:34:16.770 take advantage of that. 01:34:16.770 --> 01:34:19.190 The motivation there often is instead 01:34:19.190 --> 01:34:21.650 motivation to give to others or be nice to others 01:34:21.650 --> 01:34:25.470 to be prosocial is often saving face in front of others-- 01:34:25.470 --> 01:34:26.690 this is social image-- 01:34:26.690 --> 01:34:30.330 or themselves, which is self-image. 01:34:30.330 --> 01:34:34.310 So here-- and because of that, situational circumstances 01:34:34.310 --> 01:34:35.090 matter greatly. 01:34:35.090 --> 01:34:38.160 Societal norms are really important to consider. 01:34:38.160 --> 01:34:39.830 So when you think about incentives 01:34:39.830 --> 01:34:42.750 or any sort of types of structures and organizations, 01:34:42.750 --> 01:34:44.030 how might you-- 01:34:44.030 --> 01:34:48.500 incentives, or how might be sort of a structure certain groups 01:34:48.500 --> 01:34:51.830 of people working together, understanding these norms 01:34:51.830 --> 01:34:55.170 and circumstances is key for fostering prosociology. 01:34:55.170 --> 01:34:57.090 If you want people to be nice to each other, 01:34:57.090 --> 01:35:00.710 you have to set it up in a way that it's maybe observable 01:35:00.710 --> 01:35:02.280 what people do. 01:35:02.280 --> 01:35:03.110 It's encouraged. 01:35:03.110 --> 01:35:06.660 There's opportunities for reciprocity, and so on. 01:35:06.660 --> 01:35:08.720 So a lot of the design of society, 01:35:08.720 --> 01:35:10.880 of a firm, of a group that you work in 01:35:10.880 --> 01:35:14.180 or a team that you work in really seems to matter. 01:35:14.180 --> 01:35:16.050 So on the one hand, as I said before, 01:35:16.050 --> 01:35:18.050 it's a little bit disappointing that there's not 01:35:18.050 --> 01:35:19.280 much pure altruism. 01:35:19.280 --> 01:35:21.890 But on the other hand, if you sort of understand 01:35:21.890 --> 01:35:24.950 the motivations for people being to engage 01:35:24.950 --> 01:35:28.640 in acts that are good for others, 01:35:28.640 --> 01:35:31.430 you can sort of design incentives and structures 01:35:31.430 --> 01:35:36.470 that people work in accordingly, which will then create people 01:35:36.470 --> 01:35:40.110 being friendly and cooperative to each other. 01:35:40.110 --> 01:35:42.020 Second, I showed you that social preferences 01:35:42.020 --> 01:35:44.990 matter at workplaces. 01:35:44.990 --> 01:35:47.720 Relative pay can depress incentives to work. 01:35:47.720 --> 01:35:49.100 This is evidenced by Bandiera. 01:35:49.100 --> 01:35:50.930 The fruit pickers that I showed you. 01:35:50.930 --> 01:35:53.990 Pay inequality can lower performance via reduced morale. 01:35:53.990 --> 01:35:56.030 So in particular, if pay inequality 01:35:56.030 --> 01:36:01.250 is seemingly unjustified, that's really bad for worker morale 01:36:01.250 --> 01:36:05.180 and might lower our worker outputs and sort 01:36:05.180 --> 01:36:08.460 of attendance and so on. 01:36:08.460 --> 01:36:12.200 Third, social preferences appear to be malleable and shaped 01:36:12.200 --> 01:36:14.810 by external factors. 01:36:14.810 --> 01:36:17.720 In particular, there's evidence in favor 01:36:17.720 --> 01:36:20.030 of the contact hypothesis. 01:36:20.030 --> 01:36:23.600 Being exposed to others from different backgrounds really 01:36:23.600 --> 01:36:26.930 seems to make us more tolerant and more 01:36:26.930 --> 01:36:30.420 understanding, more prosocial towards these other groups. 01:36:30.420 --> 01:36:33.410 But also more prosocial perhaps in general, which 01:36:33.410 --> 01:36:35.630 is the evidence by Gautam Rao. 01:36:35.630 --> 01:36:38.910 And then finally, there's some evidence 01:36:38.910 --> 01:36:43.090 that biased beliefs may lower prosociology in the sense 01:36:43.090 --> 01:36:47.370 that people might under-invest potentially in how nice 01:36:47.370 --> 01:36:50.700 they are to others perhaps because they misunderstand 01:36:50.700 --> 01:36:53.970 what the effective of engaging on such a prosocial 01:36:53.970 --> 01:36:56.080 act might have on others. 01:36:56.080 --> 01:36:59.550 And so potentially correcting these beliefs or experimenting 01:36:59.550 --> 01:37:02.220 might sort of increase prosociology, at least 01:37:02.220 --> 01:37:04.830 in some settings. 01:37:04.830 --> 01:37:06.285 Now, what's coming next? 01:37:06.285 --> 01:37:07.410 What are the next lectures? 01:37:07.410 --> 01:37:10.180 Lecture number 14 is about limited attention. 01:37:10.180 --> 01:37:13.240 15 is about projection and attribution bias. 01:37:13.240 --> 01:37:15.420 This is the idea that people have 01:37:15.420 --> 01:37:18.000 trouble projecting how they might 01:37:18.000 --> 01:37:19.810 feel on different states of the world. 01:37:19.810 --> 01:37:21.900 That is the idea if you're really hungry, 01:37:21.900 --> 01:37:24.630 it's really hard to project how you might feel when you're not 01:37:24.630 --> 01:37:26.220 hungry and vise versa. 01:37:26.220 --> 01:37:27.870 We're going to look at that. 01:37:27.870 --> 01:37:30.960 In lecture 16 and 17, we'll look more specifically 01:37:30.960 --> 01:37:32.460 about beliefs and learning. 01:37:32.460 --> 01:37:34.020 We talked a little bit about beliefs 01:37:34.020 --> 01:37:35.520 already in various ways. 01:37:35.520 --> 01:37:37.320 But now, we're going to talk specifically 01:37:37.320 --> 01:37:40.140 about biases in beliefs and learning, 01:37:40.140 --> 01:37:44.640 as in people sub-optimally learning perhaps because 01:37:44.640 --> 01:37:46.860 of computational issues in the sense that it's 01:37:46.860 --> 01:37:48.990 really hard to learn in some settings, 01:37:48.990 --> 01:37:51.390 in part because of motivated beliefs. 01:37:51.390 --> 01:37:56.250 People might drive utility from their beliefs. 01:37:56.250 --> 01:37:58.080 As in, for example, I might want to think 01:37:58.080 --> 01:38:03.220 that I'm a good-looking, smart, and a good teacher. 01:38:03.220 --> 01:38:06.030 And if I get feedback on one way or the other, 01:38:06.030 --> 01:38:08.820 I might react to positive feedback a lot more. 01:38:08.820 --> 01:38:11.145 I might sort of update positively if somebody says, 01:38:11.145 --> 01:38:12.270 Frank, you're really smart. 01:38:12.270 --> 01:38:14.070 I might sort of update positively 01:38:14.070 --> 01:38:15.960 because I feel really good about it. 01:38:15.960 --> 01:38:18.180 If somebody says instead, Frank, you're not so smart, 01:38:18.180 --> 01:38:20.100 I might mostly ignore that feedback 01:38:20.100 --> 01:38:24.290 because it might make me feel bad about myself. 01:38:24.290 --> 01:38:26.080 So that's all for now. 01:38:26.080 --> 01:38:27.950 Thank you so much.