Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
========================================================================

# How Far Can Inclusion Go? The Long-term Impacts of Preferential College Admissions

Authors: Michela Carlana, Sara Chiuri, Enrico Miglino, Michela M. Tincani  
Discussant: None  
Video: https://www.youtube.com/watch?v=7LDwpGoPNFc&t=5159s


## Talk (01:25:59 – 01:44:47)

[01:25:59] Okay. Great. Thank you so much. So, let me start by saying that it's a bit slow. It's true.  
[01:26:09] Okay. Let me start by saying that this project looking forward Um >> I just was using buttons on keyboard.  
[01:26:18] >> Okay. Okay. Great. It's co-authored with like um Sarah Cury that is here in the audience. Sarah with she's a PhD student um at Princeton Rico Mino.  
[01:26:33] So here we're going to think about like how the so if you think like more broadly soio economic gaps in higher education begin like way before we can student actually apply to university and actually we know that this family  
[01:26:46] disadvantage is way larger for boys compared to girls that have higher dropout rates. They have like lower u probability of like applying and being admitted like and completing like university uh degrees. So now one of the  
[01:27:00] action that different selective universities have been taken all over the world is thinking about affirmative action policies or form of preferential admissions uh in order to try to achieve diversity and enhance the chances of  
[01:27:14] disadvantaged individuals to get into university and with the idea that we may try to get like more talent out of like this um this choice. From the literature, we actually know that there  
[01:27:27] are positive and longlasting benefit from these type of policies for those students that are extremely wellprepared from underrepresented groups. So what we are going to ask in this project is thinking like how far can this type of  
[01:27:42] policy go while still delivering on the promises of leading this positive outcome uh on the students. So to do that we're going to focus on the context of Chile where there is pass that is like a preferential admission policy is  
[01:27:56] a top percent plan but that targets like the top 50% of students in extremely disadvantaged schools which implies that we end up like getting to selective universities student that are far below  
[01:28:09] uh the um the standard like admission uh cut off that we get for the top percent plans that have been studied the US context for example.  
[01:28:19] And we're going to try to provide some direct test of the mismatch hypothesis trying to exploit these like long-term outcomes on top of like precise information on uh beliefs about uh decision- making frictions that are  
[01:28:32] elicited at the time of like high school before they enter into universities. So we're going to do that exploiting two different empirical methodologies. So we have a randomized control trial where in  
[01:28:44] one year in 2016 out of 221 potentially eligible schools for their low high level of vulnerability only 64 got admitted into this plan. And we're going to compare the student that at the point  
[01:28:57] of the announcement of the policies were in the top 15% of the ability distribution in the school with student that end up in the top 15% is schools that end up in the control group. So the second identification strategy we are  
[01:29:11] exploiting is NDIA pretty standard RDA results because we we have this like admission cut off at the top 15% and so we can try to go a little bit below and start to think exactly the margin as to the how these type of policies are  
[01:29:24] affecting their specific outcomes. There is one part of the paper that I will not have time to cover but like I encourage you to look at the paper that it will be like this link between the RCT and the RDD uh results and we're going to think about like exactly at which point we  
[01:29:39] still have like a only positive effect for all the student and when we start to see at which point of like the ability distribution we start to see this negative effect that start to emerge in a direction of like a mismatch.  
[01:29:51] So a few other details about the context of the program. Um so as I said top percent plan top top 15% of these selected like extremely disadvantaged high school get a preferential admission slots but on top of that there are a few  
[01:30:05] other criterias that are important. So first of all in order to get one of these preferential admission slot you need to be attending one of these passes school for both uh the last two years of like high school and you need to sit for  
[01:30:18] the PSU this like admission exam that is done uh to all student even if your PSU or admission exam uh grade will not matter for your ranking and you need to submit this application for the preferential admission slots and you may  
[01:30:32] get in the years we are using for this uh uh paper right now You may get two separate offers from the preferential admission and from the regular uh channel. This is super numerary. So they are not taking slots. They are adding  
[01:30:45] like some position on top specific for uh this program. On top of that there are orientation sessions that are done for all student in past schools and this includes information on college application, tuition, financial aid etc.  
[01:31:00] And there is some optional tutoring in college that unfortunately we see from the data uh the uh information we have that has a very limited like take up um uh from student side. So this is like a map of Chile with like the colorcoded  
[01:31:14] like areas of like disadvantage with the light blue being the most disadvantaged area. So you see in 2014 you were targeting like extremely disadvantaged areas and then it increased reaching  
[01:31:27] like 456 high school uh by the uh 20 uh 2016.  
[01:31:34] So the empirical framework is very straightforward. First design uh we're going to take at the point of the announcement of the policy in 2016 the 64 schools out of the 22 uh one that were eligible to be like a  
[01:31:49] pass school and this was randomly selected by the government in collaboration uh with like the research team and so they got these like and so what we do is that at the point of the the announcement we take the student that in the baseline GPA rank they were  
[01:32:03] in the top 15% of their school and we compare them between treatment and control uh schools. Here it's very important to keep in mind is that the treatment is a combined effect of being in a passive school versus like a  
[01:32:17] control schools which implies also the orientation classes for example. Um in the de composition at the end we try to go a little bit in depth in separating that out but for now keep in mind that this is like a overall treatment of like  
[01:32:29] the components we discussed together. Um so this is like a to give you a sense of like the GPA rank and the probability of a student getting a preferential co college admission offer and you see that  
[01:32:41] those student in the top 15% uh of uh uh the GPA baseline GPA uh rank they are not sure of getting like an offer but their probability is 55% because they need to meet all the other criteria and  
[01:32:54] there are some student that do not but so essentially we're going to compare these top 15% in treated and control schools for the presentation of of today. Second research design very straightforward RDD where we're going to compare student around the admission  
[01:33:09] cutoff. Let me show you right away the first stage. So where you see that if you are at uh the just like above the admission cutff uh you're going to get like a probability of 80% of receiving a preferential admission uh offer. This is  
[01:33:23] a fuzzy RDD because you need also in this case to satisfy all the other criteria. So this is not this is why it's not 100%.  
[01:33:32] The data that we use Chile has like an amazing set of data as uh you all know we are I'm going to focus on the data from the ministry of education on uh the um enrollment type of institutions. we  
[01:33:45] have like a labor market data on employment insurance uh uh for now and on top of that I'm going to exploit some survey data in the RCT sample of like high school student in 2017 that are  
[01:33:58] going to be helpful for us to get some more information uh about like or some try to pin down exactly what is happening uh for the mismatch hypothesis.  
[01:34:09] So uh a bit of like a conceptual framework for from this type of preferential admission policies we may expect two uh set of results both in terms of like intensive and extensive margin. So in terms of like intensive margin you may expect that the student  
[01:34:24] are induced to select to attend more selective uh program because they are offered a slot in a selective university instead of like going to less selective one or it could happen that they were not planning they were planning to go out on the labor market and now they are  
[01:34:37] planning to attend university. Now in terms of like uh the the impact it can have it can have actually both some positive and negative effect. So some positive effect could come from a quality uh effect and this is actually  
[01:34:52] what has been mainly shown from these type of policies with like a if you go to a better university with like more larger monetary input higher quality of professor and peers you may get like some positive effect. On the other hand  
[01:35:05] if you are an extremely unprepared like student and you get um admitted to one of these selective university the exam can be very hard for you to pass. It could be like fast pacing and it could have like a negative like um friction  
[01:35:19] due to the fact of you not fitting in with the environment you are in as we saw also in the presentation early this morning. Um okay so let me move quickly quickly to the results. So this is like a uh the type of outcomes I'm going to  
[01:35:32] show you. So the first four columns are the effect on selective college. The other four columns are the effect of any institution any tertiary institution.  
[01:35:41] Okay. So what we see here is and I'm going to show you the fact on every enrollment graduation still enrolled by seven year after uh the um the end of like high school and dropout. So what we  
[01:35:54] see and the main takeaway overall from uh the RCT uh results is that if you compare the student that were randomly they were in a school that was randomly assigned to this program compared to  
[01:36:07] students in the top 15% of the ability distribution at baseline um in a school that was in the control group we see that there is like a pretty substantial like a um intensive margin effect but no extensive margin effect. Substantially  
[01:36:21] we are not taking bringing to university student that were not going um on the um labor market but we are substantially um pushing uh people to attend the more selective university that was the goal of the policy maker uh when implementing  
[01:36:36] this specific policy and here the effect are actually huge. you get this student to attend university and they persist into actually graduating with an increase of like 24% in actual graduation from a selective like a  
[01:36:49] college um and like no effect overall even like in general dropout. So you see the last column is showing you dropout from any institution there doesn't seem to be any like overall like negative  
[01:37:01] effect. Um so let me move quickly and show you uh one interesting factor that uh we have here in terms of education by gender. All the positive effect that I was discussing is fully coming from  
[01:37:15] women. Women when offered this opportunity are the one actually taking up the opportunity and attending the selective college and you see that the probability for women to attend like a selective attend and graduate within the seven years to a selective college  
[01:37:29] increases of 9 percentage point with a mean of the control group here that is 22% of students. So this is like a huge effect for women but actually in terms of graduation for men there is like a  
[01:37:42] very precisely estimated zero effect uh uh from the policy. On the other side we start to see some negative effect for for men. Uh and in particular if you see in terms of like a graduation from any institution there is this like a  
[01:37:56] negative 66 percentage point of student that end up like not graduating from any institutions. And in part it seems to be matching the fact that they attend a little bit more the selective but they end up dropping out from the selective  
[01:38:10] um trying like the lower level but then like ending up like dropping out or like not graduating actually from that university as well. So what's the implication for the labor market? And so here I'm showing you uh two main  
[01:38:24] outcomes that we have. We have way more in the paper but here we I'm showing you on the left hand side the average yearly earnings and the average monthly income.  
[01:38:33] The years one to seven are observed data. Uh and they are years in which student could like if they go to university they are in university is actually quite common in this context to work a little bit even when you are uh  
[01:38:46] at university. So we do observe uh these uh uh this information uh there but like so for women we actually observe a substantially short-term loss uh in the years in which they are uh at university they are more likely to attend university they are in difficult  
[01:39:00] university they earn less in the um in the years uh one to five so it's around around 15% 14.4 four uh percent uh earning loss in the short term, but they make up uh for that in the long term.  
[01:39:14] And so if you we check the average monthly income um conditional uh on working is like a 4 uh 9% increase in years 12 to uh 15 estimated with like a surrogate index. uh for men we have  
[01:39:28] actually some um no significant short-term uh earnings impact and some negative effect uh in the long term that are coming from the fact that they are less likely uh to graduate from uh university. So this is like overall  
[01:39:42] effect for the RCT sample. So the main takeaway that I want you to keep in mind for me this is actually actually this program led like to very clear positive effect on education and income for women  
[01:39:55] but like losses uh for uh for men that were offered this opportunity. Now the uh second part of the paper goes more in depth into the RDD um results and so thinking there about the marginal students. So those that uh are  
[01:40:09] potentially more likely to be mismatched uh in this uh in this setting. But here like in the overall results of the of education we actually still find this extremely positive effect in terms of a graduation with a 7.8 percentage point  
[01:40:23] increase in graduation probability also for the marginal student. But now we start to see that the dropout from any institution. So the last column is increasing with a non percentage point a pretty large uh effect in uh this like  
[01:40:37] um uh dropout uh effect with like a 44% increase in the dropout probability from any institutions. Let me jump in to the results by gender and we see the what we were seeing in the RCT sample is  
[01:40:50] actually uh even more extreme here in the RDD sample where you we have that the entire effect on graduation is a positive effect on graduation is again still coming from women. So even the marginal women are the one that are  
[01:41:04] still benefiting from this opportunity. they go and they graduate more. Even if now we start to see also for women some negative effect and the last column with the dropout you see that see that also some women when it comes to the marginal  
[01:41:18] women they are actually uh dropping out where for men there seem to be like this more clear negative effect with no uh selective college graduation and some negative effect in terms of like both dropout but also like a graduation from  
[01:41:32] any institution. And here the uh what is happening like in terms of like a labor market for these uh different groups um where we find substantially so now it's like is a bit small but like it's a statistically significant effect in the  
[01:41:46] long term for positive effect for women and like a marginally significant for negative effect for men in the long term. So there is one uh I I told you that we have this information uh on um  
[01:41:59] the perceived graduation probability in selected colleges at the time of high school before they actually get into the selected college. So here you see uh how overoptimistic a student are. So this is all above the 45 degree lines. It seems  
[01:42:13] like all student are extremely overoptimistic about the graduation probability. And we have actually men that are around 0.36 standard deviation more overconfident in their graduation probability from a selective college  
[01:42:26] conditional on admission compared uh to uh to women. And so what we can do is try to understand if uh in the RDD sample the sample is here is a bit small but we can try to think about whether the fact was driven by the highly  
[01:42:39] overconfident or the student that were less overconfident because by the mismatch hypothesis we would expect this to be coming from those that were extremely overconfident and indeed we see especially in the dropout that all the fact that we were finding a dropout  
[01:42:54] was actually coming from the extremely overconfident students.  
[01:42:59] So um I'm running out of time so let me try to wrap up. So the first part of the paper is like the uh the RCT results and on that the uh uh the strong like effect is like this very positive average  
[01:43:13] effect on education coming mainly from women that like spillovers into the labor market effect that are extremely positive for women but for men given that there is no positive education gains we do uh estimate like some  
[01:43:26] negative effect uh in the labor market. This actually helps closing a little bit of the gender gap in labor market for uh men and women. Uh but it's important to keep in mind the different different reaction to being offered these type of  
[01:43:39] policies. For the marginal students, we have a clear tradeoff. So one for one selective college degree more we actually have 1.15 um higher education dropout more. And so  
[01:43:52] here is like the where the trade-off start to emerge. And it emerged especially again with men having the negative side more likely to have the negative side and women like more likely to have the positive side. So in the final part of the paper, we start to uh  
[01:44:06] think about like how um how far can inclusion go um uh bringing together like some ideas from Angloan and exploiting the fact that here we have the RCT plus the RDD in the same sample  
[01:44:20] and we actually estimated that above like the top six% of student that is around the 58% percentile of the um uh the test score by the end of like high school there is like essentially only  
[01:44:34] benefit from these type of policies. Just below that, you start to have a trade-off between the benefit and the cost uh in inducing like higher dropout rate with potentially uh negative effect in the labor market. Thank you.  

## Q&A (01:44:47 – 01:45:55)

[01:44:47] >> And one minute for question uh >> very quickly I found the gender differences striking partly you address my question with the RD but there's  
[01:45:01] another set of results could be that the men are just worse now on everything.  
[01:45:07] >> So how much of the men male female differences come from fact that worse on test scores but then maybe also on other measures that you probably don't have the data.  
[01:45:16] >> Yeah, we actually even if we control we control for test score on that to try to clean that up.  
[01:45:22] >> H is it could be a little bit nonlinear but it's actually surprising even like among the marginal students we have like a composition of both men and women. So is actually for me was uh so we we have a little bit more women. Okay. So among  
[01:45:36] the top 15% and especially the super high end but I this is like not enough to justify uh the fact that we actually do find it seems to be something beyond that.  
[01:45:50] >> Maybe we can do more questions. >> Exactly. Drink the coffee all the presenters next.