Lecture 13: Social Preferences IV

• Download English-US transcript (PDF)

[SQUEAKING]

[RUSTLING]

[CLICKING]

FRANK SCHILBACH: All right. Welcome, everyone, to lecture 13 of 14.13. This is my last lecture on social preferences. I hope all of you are doing fine remotely. Hang in there. I know this is a very difficult time. This lecture is a bit of a more uplifting lecture than the previous ones. While the previous ones have looked at social preferences, how we can measure social preferences, and how can we perhaps isolate pure altruism-- do people really want others to be better for their own sake, even if nobody knows about it, either others or whether there's no feedback-- we sort of found out that like it seems like what looks a lot like altruism-- people are nice to each other-- is not pure altruism in the sense that like they really want others to do well.

Instead, they want to either look good in front of others, or they're worried about reciprocity or sort of any other negative feedback from others, or it might be they might want to protect their self-image. They want to look good in front of themselves and they try to figure out ways to deceive themselves that they're nice. And given the opportunity to not be nice, people actually seem to be not as nice as one might think from pure dictator games or ultimatum games that we observe.

This lecture is a bit more uplifting in the sense that it talks about field evidence on social preferences, and in particular you can ask the question about can policies increase pro-sociology in certain ways. We're going to talk about three broad sets of studies. First, we're going to talk about social preferences at the workplace. I'm going to talk about the impact of relative pay on productivity.

We talked about this last time at the end of the lecture a little bit, but I'm going to just restart from scratch to sort of explain that a little bit better. Then we're going to talk about the morale effects of pay inequality. What happens to worker inequality when people are paid unequally, and is that a good idea? There's also a very nice paper in ethnic divisions and production in firms. We're going to talk about this a little bit in recitation.

Number two-- we're going to talk about whether policies can induce pro-sociality. In particular we're going to focus on mixing. This is sort of called the contact hypothesis. When you mix people from different backgrounds, do they become nicer to each other in various ways?

You'll discuss a very nice paper by Gautam Rao when mixing rich and poor children in school. And I talk very briefly about mixing cricket players in India to reduce tensions across castes or make people from different castes perhaps more prosocial to each other. And then I'm going to also briefly mention a study on mixing roommates in college by Corno et al.

Finally, we're going to talk about whether people perhaps underestimate the benefits of prosociality. In particular, I'm going to show you a study on people undervaluing gratitude. So when people write letters of gratitude, do they understand how others might react, and is there perhaps some evidence that people are not prosocial enough because they misunderstand the effects that that might have on others?

OK, so let me start with social preferences at the workplace. So Bandiera et al. is a very nice study that looks at the impact of relative pay on productivity. This is field evidence from the fruit farm in the UK. It's literally field evidence because it's from fruit-picking farms from fruit fields. The study looks at two types of payment schemes. These are piece rates-- workers are paid per unit of output. And the second scheme is relative pay-- workers are paid relative to others.

We might think that relative pay is a good way to incentivize workers because you want to do better than your neighbor, and there's a computation of who does best. And in principle, it could be that that, really, if you only care about yourself, that really gets people to work very hard.

Now, what Bandiera et al. look at is, how does the introduction of relative pay affect workers' effort and output? Now, one thing to notice is that while relative incentives might be good because you might induce people to want to be better than their neighbors or their co-workers, there's also a negative externality of relative pay.

That is the idea that increasing your own pay comes at the costs of others pay. That is to say, if there's relative pay and I work really hard, I'm paid how I do relative to my friends or others that I'm working with. Well, if I'm working really hard and do better than my friends, I'm going to be paid more. But at the same time, for any effort that I put in, my friends are going to be paid less.

And so workers anticipating these impacts on others might reduce their effort if they care about others. If I'm really worried about if there's relative pay and I work really hard, my friends are going to look bad. They're going to be relatively unproductive, and I worry that they might not be paid enough. Then I might reduce my effort. And then everybody might do that, and then the introduction of relative pay or the relative pay schemes might actually not be particularly effective or an effective way of incentivizing and motivating workers. So there's sort of this tension that relative pay in principle could work quite well. But if people really care a lot about others, then it might not work at all. It might actually backfire in some ways.

Notice that it could be either that people care about others for their own sake in the sense that you might care about them because you want them to have high earnings, or you might care about others' earnings, not because you actually care about the outcomes per se, but rather because you're worried about negative repercussions. If these are your friends and you work really hard, well, at nights, your friends might really not be happy, and there might be retribution against you socially or in some other ways if you work too hard, if you caused them their earnings to go down.

OK, so now, what does this paper do? It looks at how does switching to piece rates affect people's productivity? So the study has personnel data from a food farm in the UK. They measure productivity as a function of their compensation scheme. This is a quasi-field experiment.

The timeline is as follows. For the first eight weeks of the 2002 picking season, fruit pickers were compensated on a relative performance scheme. So the per fruit piece rate is decreasing in average productivity. So if everybody works really hard and I don't, if I'm relatively bad compared to others, I get less than others. This is an incentive, as I discussed, to keep the productivity low if you care about others.

And then in the next eight weeks, compensation will switch to a flat piece rate per fruit. So that's essentially just you're paid by however how much you produce, how many kilograms of fruit you collect. And the payment is entirely disjoint or independent of what other people do. That's a classic piece rate payment.

So the externalities are entirely shut down. If I work really hard, I'm going to be paid. There's no effect whatsoever on my fellow workers. The switch was announced on the day the change took place, so it came as a surprise to workers. So we should not be worried about sort of pre-trends or productivity going up over time anyway.

Now, what the paper finds is a dramatic increase in productivity with the introduction of piece rates. That is to say what you see here in the figure is the average worker productivity for two of the fields-- so that's essentially like broad, big fields where people work on, and I'm showing you the data for two of those fields, for which, essentially, on most days, there was actually production workers were picking fruit.

What you see on the left side of the graph is when people were working under relative pay, there you see, essentially, relatively flat productivity. People picked about 5 kilograms per hour. There doesn't seem to be any trend upwards over time. So before the new policy was introduced, essentially, workers were producing or picking their fruit at a relatively constant rate over time, about 5 kilograms per hour. There was no trends, as I said.

Now, you see the vertical line in the middle. This is when piece rates were, in fact, introduced. So there was no relative pay anymore, and now people were paid by piece rate. And that increased productivity by over 50%. So workers became a lot more productive after that. That's true for both of those fields.

It is not the case that the average payment per unit of output per kilogram of fruit paid was actually was going up. So you might say, well, if the piece rate was actually higher under the piece rate payments on the right side of the screen, of the graph, then maybe workers would just be more productive. But in fact, if you look at the piece rate over time, if anything, the piece rates actually went down over time what the company was paying.

So that is to say, the introduction of those piece rates, the productivity effect is not explained by workers being paid more per unit of output. Instead it seems to be really coming from the different incentives people workers have when it comes to how their productivity affects their fellow workers on the field.

Now, you might have two potential explanations here for this evidence. One is social preferences. So you might work less to help others under relative incentives, even less when your friends benefit. You have many friends on a certain field that work with you. Well, then you might be particularly inclined to not work very hard because you care a lot more for them. And this is exactly what they find in this paper. These effects are stronger when there are more friends on the field for a particular worker.

Second, however, it's also like a repeated game. That is to say, there's a low effort equilibrium, where, essentially, if there relative pay, we can just all agree everybody on the field might agree that not working particularly hard is a good idea because we are essentially paid relatively anyway.

So if everybody doubles their effort, nothing is going to happen to a worker's compensation. So you might as well just decide everybody could just decide work half as hard, and you're going to be paid the same. And as long as we can sustain that equilibrium, that's a good idea. Of course, in any given period, any given worker might have the incentive to deviate from this equilibrium because on any given day, if you work really hard, you're going to be paid a lot more for that day.

And so how might we be able to disentangle that? Well, what Bandiera et al. have is they have also some variation in the types of fruits that were collected. What do I mean by that? Well, if you think about what you need to be able to disentangle these, what you might want to have is differences in the observability of effort, right?

So if I can see what my co-worker does, I can punish them very effectively. I can just know whenever Frank is working really hard, my co-workers might want to punish me. And then it's very hard for me to work really hard because I'm getting in trouble at nights after work with my co-workers. Or in the next day, everybody might also work really hard, and I don't want that.

On the other hand, if I can hide how hard I'm working, well then, I might secretly work really hard, and my co-workers might not even find out. So then, if that's the case, in the repeated game equilibrium might just not be enforceable, as in I might just tell my friends, oh, I wasn't really working hard. But in fact, I was. And they have no way of finding out because I was unobservable.

If, instead, I really care about them, if I'm really interested in their well-being overall and their payment, it doesn't matter whether my effort is observable. Regardless of how observable my effort is, I will not work very hard because not working hard makes them better off. So if you just had some differences in observability in effort, how hard people work, we could just disentangle these two explanations.

So Bandiera et al. have this very nice variation. They have two fruits there, fruit type 1, which are strawberries, and they have fruit of type 2, which are raspberries. If you ever have picked strawberries, you would know that strawberries is very easy.

These are essentially very flat and low fruit that when somebody picks strawberries, it's very easy to see how fast somebody works, how much anybody picks and so on because you can just see across the entire field. You can see essentially what all of your coworkers are doing.

In contrast, if you look at raspberries, these are bushes. So you can essentially literally hide behind the bushes and work really hard to pick really hard and fast without others noticing. So essentially, productivity is at least more unobserved than for strawberries.

Now, what Bandiera et al. then find, in fact, is no impact of the piece rate on fruits of type 2, which is raspberries. So that suggests there is no evidence of pure altruism, and the effects, perhaps, could be driven by reciprocity. This is consistent with what we had found before.

When you remember the dictator ultimatum games that we showed you, it looked a lot like people are altruistic. They care about others. But when you look at their actual motives, when you dig a little deeper, we find out that, in fact, this is driven by reciprocity, or by worries about repeated game effects, or worries about essentially just retribution, either on the field that other people will work also really hard if I work really hard right now, or socially after work, my coworkers might be really mean to me. They might even beat me up, or they might not be my friends anymore.

Here is regression tables that show you this. The first column, you see fruit type 2. Again, these are the raspberries. This is when things are unobserved. There's no effective introducing the piece rate. In contrast, if you look at column 2, this is fruit type 1. These are strawberries, and there's large productivity effects of strawberries there.

So these results, again, highlight the importance of setting the incentives carefully. Small details matter, and you really want to be very careful in how you said relative incentives that in principle seemed like a good idea because they provide steep incentives to workers, but in practice might actually not work particularly well.

OK, so the second paper we're going to look at is Breza et al. These are the morale effects of pay inequality. This is a very nice randomized field experiment in rural India with low-skill manufacturing workers. The question the authors ask is do workers care about relative pay? And perhaps why do we see so little variation in pay across settings?

So what we often see in many companies in India, but also in other places, also in certain villages in India, what we see is wages tend to be very compressed. So workers essentially earn the exact same wage, regardless of their underlying productivity, experience, and so on. Good and bad workers tend to earn exactly the same amount of money for per day or per hour and the like.

Economists think that's inefficient because if you want to hire somebody who's really, really productive, you might want to pay them more. If you want to hire somebody who is not that productive, you might want to pay them less. You might actually want to pay them less than the prevailing wage, but you might not be able to do so because the norm is that you have to pay them the prevailing wage. And then you might end up not hiring that person, and that person might end up being unemployed, even though you would be very happy to pay them at a lower wage.

This is a really relevant question. Sorry, the second question the authors ask is what is the notion of the underlying notion of fairness? What is fair and what's not fair? As in if there is wage dispersion across workers within teams or within a certain firm, under some circumstances under which that's OK to do?

Are there some good justification that workers are OK with, or is it just always bad to do so? So can we find some situation under which it's justified for workers or not, or is it just like we cannot pay have pay inequality under any circumstances in those kinds of firms?

This is potentially quite relevant for many features of the labor market. It might explain wage compression. That's what I'm saying, like wages tend to be fairly compressed, even if productivity is fairly dispersed across workers. There tends to be lots of wage rigidity. Wages tend to not move very much. Employers have trouble adjusting wages, or they are reluctant to adjust people's wages much, in particular when it comes to like downward rigidity. We talked about this a little bit already when we talked about reference dependence.

This can help us think about like sorting of workers into firms and inequality. We can think about firm boundaries in a sense and say should they be really large firms or a number of small firms? If you think the relative comparison is across workers within firms, that would say, well, we should probably have a lot of small firms.

And then each of these small firms could have different wages, as opposed to one large firm where essentially everybody has to be paid more or less the same. And we might think that's inefficient because as returns to scale of having a large firm. Think about it is also relevant for some HR policies in terms of how do you pay workers, how do you set wages, and so on.

So as motivation in the study, the authors ask people the following questions. The questions are the following. This is rural India in Rissa, a relatively poor area with lots of small-scale manufacturing. The question is three people from a village get hired to work on a construction site together.

The prevailing wage is 250 rupees. That's the standard wage in the village. Each village tends to have like a prevailing wage, which is how much people are paid usually. The contractor pays them 250 rupees per day. How well will they work together?

And what you see is what respondents tend to say is 80% of respondents say people worked very well together. 20% say as well as usual, and 0% say there will be conflict. So that's seems like paying everybody the same is the socially acceptable thing to do. Everybody seems happy with that.

In contrast, if you ask the following, which is the contractor pays them different wages. If the contractor pays 250 per day, 270 per day, and 290 per day, how well will they work together? And so now, what you see here is when there is differences across pay, 94% of workers say there will be conflict.

Listen, in this version of the question, it doesn't specify whether that's due with underlying productivity differences. So why is it that some workers are paid different than others? However, it's very strong suggestive evidence that workers might be really unhappy, and there will be conflict across workers if they're paid different wages in the same kind of team, or at least on the same construction site.

And then, of course, if you're an employer, you might be quite worried about paying people different wages because of this conflict. Presumably, workers will not work or collaborate as well as they could potentially.

OK, so now the authors set up a very nice field experiment that looks like this. So there's 10 production units of 3 workers in each factory. So think of factories like work sites, factories in quotation marks because it's really not just a factory as you might imagine it. It's really like a small work site, an office where people are working in different areas. There's 10 units of 3 workers each, so it's not a huge office.

Each unit of three workers produces different products. So each unit has one product. For example, unit 1 makes brooms. Unit 2 makes incense sticks. Unit 3 makes leaf plates, and so on. So each unit has one item of production.

Now, why are there different units? It's essentially to separate workers into different lines of work. So what we're going to expect is that workers are going to compare each other within unit, but not so much across units because if I make brooms and you make incense sticks, that's very different work in some ways. But if you make the exact same thing, the natural comparison now for workers to compare yourself with another worker.

All unit members, so everybody in a given unit, makes the exact same product. So everybody in unit 1, who is three workers, makes brooms. Everybody in unit 2 makes incense sticks, and so on. Now the key experimental variation in the experiment is weight dispersion across workers within teams. Finally, there's also-- and I'll get back to that at the end-- production tasks vary in their observability of performance. So that is to say for some tasks it just happens to be that it's much easier to understand how productive your co-worker is than in other tasks. OK, so now what is the variation in relative pay.

So here you see people are paid depending on their worker rank. There's four different regimes of pay in the different columns, heterogeneous, compressed L, compressed M, and compressed H. Workers are sorted. So there's a baseline period where workers who work for a few days, and the few days are used to assess their baseline productivity.

They're ranked into three turnstyles, so there's like the low productive workers, median productivity, and high productivity workers. They're ranked essentially based on how well they did at baseline. OK, so if you're like really bad at the start, you would be a low productivity worker. If you're like an average worker, you would be median productivity. And if you're really like highly productive worker, you would be classified as high productivity workers.

Notice that's always done for all of the different treatment groups. The difference now is across these different regimes how workers are paid. In their heterogeneous pay essentially, there's differences in cross workers in the sense of like the low productivity workers are paid low wage, the median productivity paid workers are paid the median weight, and the high productivity workers are paid the high wage.

As you might expect, that's, sort of, the natural thing to do in terms if wanted more productivity, the most productive workers you're going to pay the most. Notice that these wage differences are modest. Even like the difference between W high and W low are like only about like up to 10%. So really these are not huge differences. It's not that the other guy if you are like low productivity worker if you get W low, it's not the other guy gets like twice as much. It really is like modestly more, but it is more money that they get.

Now then if you look at the other three regimes, compressed L, compressed M, and compressed H, low productivity-- so in compressed L, everybody gets to low wage, in compressed M, everybody gets the minimum wage, and in compressed H, everybody gets the high wage. And now the study now lets the authors compare the different columns for holding exactly constant baseline productivity and wage levels. That is to say, for instance, we can compare the heterogeneous-- the low productivity workers that happen to be randomized into like the heterogeneous treatment, they receive W low or compare them with like low productivity workers at baseline who happen to be randomized into a compressed L. So I should have said more clearly people are randomized into any of these four groups, and then depending on what your baseline productivity is you get the wages as I show here in the table.

So for example, if you are like a low productivity worker, you might be randomized into the heterogeneous treatment group. You might receive W low, or you might be randomized into the low-- compressed, low treatment where you also receive W low. Notice that in both cases, that worker is a low productivity worker. In both cases, the worker receives W low. However, what's different here now is that his co-worker-- his or her-- in this case, these are all men-- co-workers are receiving either the same, which is in compressed L, or they receive higher wages in the heterogeneous weight treatment, right.

So now we can look at workers who have the exact same baseline productivity on average at least and have the exact same weight, but what's being varied is like how much other workers are earning. We can do this for low productivity workers. We can also do it for median productivity workers or also for high productivity workers. I skip the median productivity worker, but that's exactly the same. For high productivity workers, we can look at the workers who had been randomized into the heterogeneous treatment group or heterogeneous worker group-- sorry, heterogeneous wage group where the worker gets a WH but everybody else gets like a lower wage, or we can compare that to compressed age where everybody gets WH in that group.

OK, so now what do you what do the authors find? Let's start with the low productivity workers at baseline. Remember the collinear comparison here is between pay disparity-- this is a heterogeneous group-- pay disparity and compressed L, which is like the group where everybody is paid the same. Now what the authors find-- and you see this in the graph fairly nicely, which is like the productivity-- these are like the red line and the blue line-- the productivity on the left side- this is before the treatment starts for about over 10 days-- the productivity looks pretty much the same.

But then when you go to the right side of the graph, not immediately but after a few days they seem to be a gap or a gap emerges between the red line and the blue line, which is exactly the gap as you would expect if the disparity makes workers less productive, which is to say the workers who receive a low wage but others in that group are receiving higher wages are becoming less productive compared to workers that receive a low wage where others have the exact same productivity.

And you see this also in the regression tables, you see about a 22% reduction in mean output and a 9% reduction in earnings, which are pretty large effects. So these are like large effects compared to like other interventions that people have tried. Interesting, you see a little bit like while the treatment effect when you look at the graph initially looks not particularly large, if anything it might not even be there often like a few days, that treatment effect increases over time. It becomes larger so over time people become less and less productive compared to the compressed wage treatment.

Interestingly, we find-- the authors find similar effects for high ranked, high productivity workers. So notice that these are not workers who either who are high productivity workers to receive the same wage in the pay disparity treatment and the compressed H treatment that have the same wage. In one case, they are paid exactly the same. In the other case, they're paid more compared to their coworkers. And what happens what seems to be the case is that the compressed high paid workers are, in fact, more productive compared to others.

You might have expected in some ways like if one guy gets paid high payment but then others get paid less, that makes the high pay worker more productive, because maybe he feels good about himself that he's a high paid workers. And maybe he feels he can prove himself or the like, but instead what seems to be the case the group that works in his team that does not work well together or they just becomes uncomfortable to work with somebody else who is like really mad at you for earning more than they do. And so now the high pay workers in the compressed treatment, in fact, earn more or produce more than the high paid workers in the pay disparity treatment.

So that is to say the pay disparity, pay inequality does not only reduce worker's productivity for the low pay workers, so not only the people-- the workers who are earning less compared to their co-workers, and that's in some sense to be expected. You might just be mad at everybody else or other people in your group are earning more. Then you might be unhappy and mad about that and just then produce less. Instead it seems to be the case even the high productivity workers, the workers who earn more than others in their group are producing less compared to the control group where everybody earns the same.

So what did we learn from that? Well, pay disparity lowers worker performance for all team members, and so the interpretation of that is that pay disparity undermines workers ability to cooperate in their own self-interest. The paper has some additional evidence with some cooperative tasks where essentially workers are worse at cooperating with each other. It seems to really be that workers are not happy to work in the same place with somebody else who earns less.

Importantly, the perceived justification is essential in mediating these effects. That is to say I told you previously that for some task it was easier to see who was more productive than others, especially at baseline just because in the task the difference is across highly and less highly productive. Or high and low productivity workers was just wider, and it's just easy to see that Frank is really bad at this and somebody else is really good. And for some task, it's really easy to see that, for others not.

So if it's really easy-- if there's a perceived justification for workers, if workers, sort of, thought, OK, it's really easy to see who is more productive than others, the effects are much weaker than when there is no receive justification. That is to say workers themselves are saying, like, if I'm saying, look, there's this other worker who is really productive and if I'm sort of saying, hey, look, this is really obvious that they're more productive than I am, it's only fair that they're earning more, then the affects are much damage-- then there's much less than an effect that's accepted, and there's much less of a productivity effect, if any, if there's pay inequality.

However, if it seems to be the case that they're, kind of, like producing the same thing, if we like similarly producing the same thing and we're equally good at it or approximately equally good and one person is paid more than I am for no good reason, so it appears that workers are really not happy about that, then work performance, in particular output falls.

So what are the implications? Well, this evidence suggests that weight compression may be more likely in some settings than in others, right. That's to say like if it's easy to justify, if the production function or like the production process is such that all aspects of the production and the output or the performance are easy, measurable, and observable for workers, then workers might say it's OK if some workers are earning more than others. However, if that's not the case, if there's a bunch of stuff in the work process that's really not observed-- maybe it's verbal or maybe just really hard to justify or hard to sort of assess in some tangible way-- then workers might be really unhappy and that might lead to weight compression, because the employer might anticipate that.

So overall, this, sort of, says that like relative comparisons matter quite a bit. Fairness matters quite a bit, and what's really key for an employer and anybody who sets incentives is to take into account these fairness considerations, and one wants to really understand what is fair and what is not. And upsetting workers by violating these fairness considerations or norms might be really, really costly for an employer. So the good news here is that if one understands that well, one can really sort of produce increased productivity quite a bit, and in some cases weight inequalities actually fine. One just has to be careful in figuring out when exactly is that the case, OK.

As I said, there's a very nice paper by Jonas Hjort on ethnic divisions and production in firms. You're going to talk about that in recitation.

OK, the second part of this lecture we'll talk about policies to increase pro-sociality. And so the first of these papers is very nice recent paper by Gautam Rao that looks at the question are social preferences malleable. Like, what are the origins of social preferences? Why is it that some people appear nicer than others? And then once we understand that, perhaps, we can also understand what policies, if anything, can affect social preferences.

The main question that this paper asks is how does being mixed with poor students in school affect the social preferences of rich students. Now in a lot of the, kind of, policies where poor students are mixed with rich students, you might ask the question, how does it affect the poor student. If a poor student for whatever reason would not be able to afford like a rich school or rich student's school and you, sort of, allow them by giving them scholarships or other policies, often the question that people ask is like what are the benefits of doing that. Is the poor student doing better in school, or do they have different types of friends, different networks, and does it lead to better jobs and so on and so forth?

This question, this paper ask a different question. It asks the question about what is the effect of rich students if they randomly or quasi randomly are exposed to being in class with additional or some more poor students. The paper exploits a policy change that introduced an admissions quota of 20% for poor students in primary schools in Delhi. In Delhi, these are rich primary schools.

The paper looks at two sources of variation. There's variation across classrooms that allows the author to look at the overall effects. So you can look at, like, within schools there's going to be treated and controlled cohorts, because the policy was introduced at some point. So for some students, they enjoyed the benefits of being in class with poor students or the costs. We'll see about that.

And for other schools, they're already too far advanced. They were like essentially before the-- they went to school before the policy was enacted, and therefore they were not exposed to having poor students in school. For these treated schools there is essentially variation within school across cohorts that we can look at.

Second within cohorts. We can look at treated and control schools. So some schools were getting additional poor kids and others did not. I'll tell you about this in a second. And then there's variation within classrooms to allow us to look at the role of personal interactions. There are, sort of, idiosyncratic assignments to study groups, so some students happen to be in study groups with a poor kid, and others were not.

Now what does this study measure. The study measures three broad set of outcomes. It measures pro-social behavior and generosity. This is very much what we have already looked at.

In particular, it looks at dictator games and volunteering for charities at school. The paper has a very nice mix of laboratory outcomes, sort of, measured in the field, which is, sort of, the dictator game that you're all familiar with. And field outcomes, were just, sort of, trying to collect real world outcomes in the sense of like things that are, perhaps, somewhat less contrived from the perspective of kids in school.

And what's very nice about the study is that it seems to be that the results are very much aligned between, sort of, these live outcomes, the dictator game type outcomes, and the field outcomes, which is volunteering for charity at school in this case. Second, the author looks at discrimination in social interactions. In particular, he does it all sports contests and looks at teammates selection among these students and then willingness to attend play dates with poor students.

Finally, the author also looks at academic outcomes, in particular test scores and disciplinary infractions. Now why might one want to look at academic outcomes, and why we're interested in social preferences, there's several reasons for that. In particular, an important reason here is that if you're against this type of policy, I might, sort of, show you that, oh, adding poor children might affect the social preference of rich children and so on in various ways. Maybe discretization goes down, but perhaps it's the case that all comes at the cost of academic performance, right.

If you're trying to, sort of, implement this policy and persuade policymakers, maybe teachers or parents of like, OK, let's have more poor students in your school. Well, parents might, sort of, be OK fine. There's going to be like some change in social preferences, some change in discrimination, but really what we care about is test scores or discipline at school, and these poor kids might not be-- might be worse in terms of test scores, and they might have negative peer effects in terms of test scores but also in terms of disciplinary infractions. And, sort of, as a policy question, then it's really important to understand if there are some benefits in terms of poor sociology or reduced discrimination, do these benefits come at the cost of reduced academic performance?

So now what is, in fact, the policy innovation in Delhi in 2007. There was a 20% admissions quota in private schools introduced for poor students in some of these private schools. There's a household income cutoff of $2,000 per year. So these are not, sort of, the poorest of the pole of the students who are qualified. Schools which received subsidized land from the government were essentially included in this policy change. That's over 90% of elite private schools. So think of these elite private schools.

These are really, sort of, like very rich kids or parents of these kids. There's no fees for the poor kids, because they would not be able to afford these fees anyway. There was importantly also no tracking. So it wasn't that high ability or high performance kids would do very well where they like tracked into the high tracks and the good tracks verses lower tracks. Instead everybody was mixed together, poor and rich, high performing, low performing students.

And the poor kids were selected using lotteries, which in principle also allows the author or others to look at the effective on those poor kids, the students who are selected versus not. To be clear this is not the subject of this paper. Just to give you a sense of the magnitude of like the mixing, you see is that the average beneficiary was a, sort of, at the 25th percentile of the income distribution in Delhi, and wealthy students are very much on the right tail-- sorry on the right tail of the distribution, something like the 90th percentile, 95th, which is about-- the US equivalent of that would be about $200,000 per year.

The average beneficiary had like the US equivalent income of like $23,000. Again, these are not the poorest of the pole but relatively poor as about like an order of magnitude poorer than the average person in the actual school. Now this policy then induced large variation across classrooms. If you look at poor students in the fraction of the numbers of poor students in by grade in 2011, I told you the policy innovation was in 2007.

So if you look at 2011-- that's four years later-- anybody who is in grade four in 2011 has essentially no poor student in that class in those rich schools. So for students who were in 2011 in grade four, five, or six. That policy came too late. They did not have any poor classmates.

In contrast, if you look at the lower grades, grades 3, 2, 1, and 0, which essentially preschool or minus 1, which is preschool grades, there's lots of additional poor students now in those rich schools. Now again, there's variation within schools and across schools. The variation within schools is across classrooms. That's, kind of, what I'm showing you here.

But it's also a variation across schools. So just to be clear, the variation within schools across classrooms is like in the treated schools there's some classes like the fourth grade have like essentially zero poor kids in their classes, and the third grade, in contrast, have lots of poor kids in the classroom. They can do that comparison across classrooms within school.

In addition, has also variation across schools. There is, in particular, there's treatment schools, which comply in 2007 as they were supposed to, but then there's also delayed treatment schools, which complied in 2008. So the jump shifted by one cohort. So the cohort, the jump that I showed you here, this is for the treatment schools that complied or this for everybody that complied in 2007.

The delayed one complied a year later, so you'll see them essentially, somewhat later increasing-- or one year later exactly increasing the fraction of poor kids in their classroom. And then there is the control schools that were not subject to the policy, at least until 2013. So they had received-- they had received land from federal government or private foundations. So essentially there was no treatment. That is all students are rich in these types of classes.

So that is-- the land receiving is essentially just-- that was like a fairly arbitrary rule in some ways. So that, sort of, leads to at least quasi random variation. And some schools received the treatment, and others did not.

Now let me show you some outcomes. So one outcome is our friend the dictator game which you have seen quite a bit and played already as well is students are allowed with 10 rupees, which is not very much-- that's about like $0.15-- and they choose to share-- notice that these are pretty young kids, so 10 rupees is quite a bit of money for them. They choose to share an amount as before between 0 and 10. And Gautam sets this up such that students could-- it was not just about the money, but they also can exchange the money for candy later, add 1 rupee per piece. So that was a pretty good deal.

Then the dictator games for the order was randomized. The kids would play two games. Again, so all rich kids would do that. They play first game one where the recipient is a poor student in a school for poor children. And in game 2, it's a rich student in a private control school. Notice it's a control school to be clear, so it's not a kid that they might know.

So the kids are chosen such that the kids were given the names and the photographs of the school shown to the subject, to the kids. So you could sort of see-- and this is verified in the briefing by the author-- is that the children understand very well who is a rich kid and who's a poor kid, but these are not kids that they would actually know. And something that's a less interesting question if you have gone to school with somebody for a long time, you might have seen them around in your school yard, you might be nicer to them or less nice to them.

That's not what the study is looking at. The study is looking at other kids in other schools that look rich or look poor both in terms of the way they're dressed and so on, but in particular the way their school looks like. So essentially these are-- now we're looking at social preferences towards other people, people that these kids don't know.

OK, so now what do we see. So this is a generosity to the pole. So like this is essentially the results from game number one. Again, remember these are all rich kids who are giving now to poor kids, and the question now is being exposed to or does being exposed to poor children for several years, does that affect giving in the form of dictator games towards poor children to start with. And we're going to talk about rich students after that.

So here you see the generosity to the poor in control schools. You see this is the percent to the poor recipient. It's remarkably similar to what we've seen previously, which is about 20% to 30%. About 25% is given to the pole kid in the control groups in the control schools. This fraction seems to trend up a little bit over time, but it's also quite constant over time across grades. You see here by grades two, three, four, and five on the x-axis. On the y-axis, it's the percent given to the poor recipients.