 Last of today, so I'm presenting affirmative action and demand for schooling. Evidence from a nationwide policy. This is joint work with Ana Paula Mello from Howard University. It's very early stages, so comments are very welcome. So affirmative action in higher education has been adopted worldwide to mitigate inequality in access, in performance, in graduation. The economic literature has shown that these policies are indeed very effective to decrease inequality in college access, especially to elite universities and to high-return majors. But a much smaller strand of the literature analyzes how these policies adopted at higher education institutions impact students earlier on. In their educational trajectories, for example, how those university policies affect students, high school students' effort and human capital accumulation. A couple of theoretical papers show that the impact of affirmative action on pre-college human capital depends on how they change admission probabilities. So we know that if you create an affirmative action for a targeted group, it will increase admission probabilities for these targeted groups and you'll decrease admission probabilities for non-targeted groups. But the impact on high school students' behavior will depend to the extent of how affirmative action will change these admission probabilities and how students will then perceive the changes in their returns to pre-college human capital. On the empirical side, the literature has found mixed evidence on the impact of affirmative action on pre-college human capital decisions. Some has found positive effect, some has found negative effect, and it shows us that the design of the policy matters, the aggressiveness of the policy matters, the context matters, and the way we contribute to this literature is then by asking, well, does affirmative action in higher education affect high school persistence and demand for college? And we do so by analyzing a very large affirmative action reform that took place in Brazil in 2012. I think this is maybe the most aggressive quota reform in the world. I'm going to explain later. And this is targeted at students that graduated, that studied three years of their high school in a public school as a proxy for social, for low socioeconomic status. And we look at how this policy affects high school dropout, high school graduation, and college demand. So we use the take up of a national college entrance exam that we take in Brazil as this proxy for college demand. And to identify the causal effects, we explored both geographical and time variation in treatment intensity. So we explored the proportion, like the expansion of the proportion of college seats that are located to affirmative action quotas by municipality. I will get into details soon. And we also exploit, we look at how these effects are heterogeneous by type of school and by school socioeconomic status. So our context is Brazil. So affirmative action policies in Brazil started in the early 2000s. They, those are policies that were adopted by public institutions, both federal and state colleges. Those colleges, they comprise around 28% of total enrollment. But importantly, they are considered of being the better universities in Brazil. They are on average of superior quality than the private universities. Plus they are free tuition. So there is high competition for these spots of those highly selective institutions. And these affirmative action policies, they most often, they primarily, the group that these policies chose to target are those students that graduate from public high schools. It's really weird when I talk about this in Europe. They don't understand this, but I think for an audience of mostly Latin American and developing countries, we understand this better. So the government chose to target public schools because it was the easiest proxy for low socioeconomic status that they could find. And there are also always sub quotas destined to known white individuals, black, mixed and indigenous, because those groups are historically under representative in public colleges, especially in selective majors. So until 2012, these policies, they were a choice of institutions. So institutions decided to adopt by the way, they adopted different types of policies, different intensities, or there were some state laws that mandated state universities to adopt these policies, but there was no national unified policy in place. Until 2012, when the federal government approved what is called the quota law, Leiji Cortes, and it established that 50% of all spots in each major and in all federal higher education institutions in Brazil have to be reserved to students that started the three years of high school in a public school. And then there are also sub quotas for known white and for low income students. But here I'm going to focus on the quotas for this national, the broader group that is the public school students. So the policy was approved in August of 2012 and it started, universities had to start to adopt this national policy in their admission process up 2013. They had around four years to reach full adoption, to reach 50%, but they had yearly minimum quota requirements. So there is a large expansion. So this is the variation at the extensive margin that we're going to explore in this paper. So if you see, there were 94 federal higher education institutions in Brazil in 2012. So in 2012, a little over 50% of these 94 institutions already had some sort of quotas for public school students. But then in 2013, 100%, all univeral institutions adopted these quotas because of the federal law. And if you see, this line is for the state universities. They are not subject to this national federal law. So there's not much of a change here. And this is the total variation. So here is the share of college seats that are allocated to quotas for public school students in the federal system. So in 2012, there were already 23% of seats that were already allocated to these quotas for public school students. And this expands reaching in 2015, 48%, but like full adoption, it was in 2016 when they reached 50%. But then our data spans this time period. So basically we explore this shift here, which was mostly primarily due to the 2012 federal law. All right, so we know, we gathered the data, so we know how each university in Brazil was exposed to this national law. But here we want to understand how this policy adopted at the university affect high school students, high school students persistence, high school students take up of this college exam. So we don't know how each high school student is matched to each university because you can apply to any university in the country. So we have to create a measure of exposure. So what we do here is first, we create this measure of affirmative action exposure for municipality M and ERT, which is the percentage of total seats in municipality M and ERT that are allocated to affirmative action or to quotas for public school students. These measures normalize to vary from zero to one. So one is 50% of reserved quotas or full adoption of the national policy. And we also know from, we gather also, we have a administrative data of pre-reformed flows of students between high school and college. So we observe this theta. So for example, theta and M is the percentage of students that lived in municipality M during high school and then keep the students that stay in municipality M during college in a pre-reformed year. So this is the percentage of stayers. While theta M D is the percentage of students from municipality M that were attending high school in municipality M and that moved to college, to municipality D for college, also in a pre-reformed year. And we use this pre-reformed flows to weight this affirmative action measure of local exposure. So we create for each, for students that were attending high school in municipality M, the treatment exposure of these students is defined by this expression where the treatment for municipality M at time T is weighted by the percentage of high school students that stayed in that municipality for college. And then the treatment at any municipality D in the country is weighted by the percentage of students from municipality M that moved to municipality D. So basically we just use this pre-reformed flows to weight treatment happening at any municipality. We could do this by distance or anything like that. We try different things. But basically results are robust even to using only treatment happening at the municipality where the student resided for high school because 87% of students in Brazil attend university in the same municipality where they reside. So basically we do this, but this is robust to also not doing this. And in principle we can, when we use this definition, we can define treatment for students of any municipality in the country. But in practice, when you look at the data, there are many municipalities that do not send any students to higher education, to further higher education. So if there are zero students from high school at a certain municipality that go to further higher education, then these theaters are going to be zero. And then we don't define exposure for this municipality and then we cannot, those municipalities are not in our sample. So then our final sample is restricted to municipalities for which in the pre-reform year, students actually go to further higher education, okay? So then we estimate this equation where outcomes, so the outcomes that I'm gonna show you today are high school dropout and take up of this college entrance exam. So then why is the outcome for a school S at municipality M in time T? So our time period spans from 2010 to 2015, some pre-reform years and some post-reform years. So this AAMT is this measure of exposure that I just defined. DZ is a vector of municipality time varying control. So basically we have to control for the adoption or the expansion of the centralized admission system, CISU, the resilience note that also is being expanded during these years. We try also some other time varying controls like GDP and things like that, but our results are robust to not controlling for this as well, but the control for CISU is important. I can explain later. We have controls for school fixed effects and for time fixed effects. And our standard errors are clustered at the municipality level. And we weight this equation by school size. So our outcomes are at the school level and yeah. All right. Okay, so really briefly on the identification strategy. Our identification here relies both on the geographical and time variation in treatment intensity. So in the proportion of college seats, they are located to affirmative action. The way that institutions expand this policy in these years depend largely on already on the share of affirmative action that they had before the national policy. So basically we are relying here on a parallel trends assumptions that the dynamics and the outcomes between treatment and controls would be parallel in absence of the treatment. So in a previous paper of mine, that was my previous job market paper, what I showed, I also in this paper, I look at the effect of this quote expansion on the demographic composition of universities. And what I show in this paper is that the pattern of adoption of affirmative action by the federal institutions during this period is not correlated with pre-trends in the enrollments of low socioeconomic status students at this public universities. This means that these federal universities, the way in which they adopt, they expand these quotas in this period is not correlated to changes in their student body. So this is because universities are expanding the quotas because of the national law that is adopted in 2012 that externally mandates that they do so. So our identification strategy here relies on the fact that if universities are not adopting quotas due to what they observe in their student's body, it's very unlikely that they are adopting their expanding these quotas because of what they observe in the behavior of high school students in their municipality. All right, so, yeah, thank you. So, yeah, we show it more formally through a placebo experiment in the paper, but I didn't include it here. All right, so I have only five minutes. So let me show you some results. So this first, we have some results for high school dropout. So this is the targeted students, so students from public schools and these are non-targeted students, students from private high schools. So a full adoption of affirmative action of the quota law decreases high school dropouts for public high school students in 1.9 percentage points from a baseline of 18%. And it doesn't change dropout rates for students from private schools, which is important because in theory, we could also expect that the non-targeted groups that this group would have lower incentives to study, lower incentives because now they have lower chances to get to university. So in theory, we could expect this to goes both ways and it's important that we find zero effects in this group. So there is a positive effect on the targeted group but no negative effect in a non-targeted group. So there is a decrease in the baseline social economic gap in dropout rates by 11.45%. Then we look at this demand for public colleges that we measure through the take up of the national exam. So what we see is that for public school students, the targeted group, there is an increase in 4.2 percentage points in the take up for this national exam. But for non-targeted group for the private school students, there is a decrease in the take up for this exam in 6.9 percentage points. I'm gonna show you or tell you about our hypothesis about that soon. And then think my last slides of results. So this is just some of the heterogeneity, the most interesting one that I think. So what I show here is that for the results on the college exam take up, this increase for demand for college for the targeted group comes from students from low socioeconomic status schools. So these students before they didn't even take the exam to go to college and now, I mean, there is a large increase in this group. So now there is, we interpret this as an increase in aspiration. So now these students aspire to go to college and before they didn't even take this exam. While this decrease for the non-targeted students is concentrated among private school students from high SES private schools. And yeah. So what we show in this paper is that affirmative action in higher education can affect pre-college human capital investment decisions. So affirmative action in Brazil contributed to the narrowing of the socioeconomic gap in high school persistence and in demand for college for public college. So regarding these negative effects on demand for public college among the high SES private school students, we are still investigating this more. But we have in mind two positive effects. One is that these students are delaying college entrance now it's getting more competitive for them. So this is this demand for college right after conclusion of high school. It might be that they are going to preparatory courses. They're very common in Brazil and they are taking this exam. Anna, my co-author thinks that this is likely, but for me what's going on is this thing. Those students are being displaced to the private university. So now they don't even want to take this exam to attend federal universities anymore. I mean now those very rich private school students, they are just moving to the private universities right away. And for those universities, you don't really need to take this national exam. But we still have to connect data with the higher education data to really prove that this is what is going on. And what is also important is that we find economically significant effects that are induced by marginal short-term changes in policy intensity. So what we explore here are really year by year changes in the intensity of affirmative action during this period and we already find effects. And also we are working now on a dynamic specification in which we explore the timing of exposure of different cohorts to this law and then we expect to find even larger results. But what is important is that, well affirmative action is a very controversial policy in Brazil to today. But then any policy debate that ignore these facts, that ignore this positive effect on human capital accumulation may understate the benefits of these policies, also understand the costs. And so it's really important to have a full picture of how these policies affect different outcomes to really understand their importance and that's it. Thank you.