Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. ======================================================================== # Friends in Higher Places: Social Fit and University Choice Authors: Nagisa Tadjfar, Kartik Vira Discussant: None Video: https://www.youtube.com/watch?v=7LDwpGoPNFc&t=327s ## Talk (00:05:27 – 00:46:01) [00:05:27] So I'm really excited to present my joint work with my co-author Cortic Vera who's now at the IFS uh in London and this is actually his job market paper. [00:05:36] So um with that I'll get started. Um hopefully this clicker is going to work. [00:05:41] It was working. Okay. No, the screen is not working either. [00:05:47] Okay, I'll just No, no, I just have to keep clicking, I think. Okay. Um, so elite universities are widely regarded as vehicles for social mobility, but at the same time in a number of different countries, including the US, uh, Sweden, [00:06:02] Chile, and the UK, it's been shown that low SCS students are less likely to attend elite universities, even conditional on their high school grades. [00:06:11] And of course, there are many uh potential explanations for this. The first one being simply gaps in academic preparation. Uh there's been some evidence on information friction about admissions chances in some contexts as well as uh lack of transparency on the [00:06:25] final price tag of universities which is a pretty big force in the US. Uh immutable preferences. Lows students might simply prefer to live closer to home for instance. And finally, the potential for uh lows students not reaping the same sort of labor market [00:06:40] returns and other sort of social returns from elite universities that their highs counterparts could have access to. But what's puzzling is that in the UK, socioeconomic gaps in enrollment at elite universities persist despite the [00:06:53] presence of uniform tuition fees across universities and uh widespread policies that these universities have adopted towards income-based affirmative action in an effort to recruit talented uh and socio economically disadvantaged [00:07:06] students. And at the same time, uh elite universities have gotten some bad PR uh if you will. And there's been some anecdotal and qualitative evidence documenting some challenges with the social environment that these students might face. So on the left here is an [00:07:21] article from the Guardian uh talking about how workingclass students can constantly feel like they don't belong if they don't wear the right clothes for instance. And there's some really great work by sociologist Anthony Jack who's done a series of qualitative interviews [00:07:34] with the IB League students who are um from poorer backgrounds talking about you know kind of the day-to-day experiences and small slice that they can experience in these environments. [00:07:45] And that brings us to these three facts that I want to show you from our data in the UK. So first on the left here I'm showing you if you look at standardized test scores from the 90th percentile to the 100th percentile. So these are high achieving students and you split them [00:08:00] into high income in the light blue and low income in the darker color and you just look at the whether or not they attend one of the top 10 most selective universities and you can see that there's actually a persistent gap that actually appears to even grow as you get towards the most talented end of the [00:08:14] distribution. And what's interesting is that the second figure will show you that this gap uh and starts at the application stage. Right? It's not that these students are uh applying at similar rates and simply not being admitted. They're actually choosing to [00:08:28] not apply in the first place. And finally, in this right graph, it's one measure. It's a pretty extreme measure of social exposure to elite universities as we've defined it. So, this is the whether or not someone encounters within [00:08:41] their high school an individual who attended any of the top 10 uh universities over the last three years. [00:08:49] And so, you know, if you if you're zero for that variable, it means there's literally no one in your high school in the last, you know, basically all the time you've been there who's gone to any of the 10 most selective universities. [00:08:59] This is a pretty stringent condition. Essentially, all the high-income students have that exposure, but you can still see that among low-income students, there's there's a a good number of them who aren't uh encountering these individuals. And this really brings us to the heart of this [00:09:14] paper where we're interested in understanding the relationship between these three facts. [00:09:20] So why now why might students respond to the university choices of the individuals around them? So there's there's actually a pretty large literature showing that students are highly responsive to the university choices of family members, neighbors, [00:09:35] schoolmates. Um I just saw a recent paper looking at the effect of neighbors even on career choice. Um at the same time we know relatively little about why uh individuals are kind of responding to uh what people around them do. Now one [00:09:50] possibility is that people simply want to be colllocated around their friends and family members and so this can happen even in the case of no information frictions. Um but another possibility is that there are [00:10:04] information frictions that are being alleviated through this social exposure and individuals are learning something about the university process or otherwise when you know their sibling or uh one of their schoolmates goes to the university. And you can even think about [00:10:17] this information friction as being one of two potential types. The first would be a general information friction. You can think of this as simply not knowing how to navigate the application process or the admissions criteria and the grades needed to go to an elite [00:10:31] university or being misinformed about the labor market returns. Uh and another possibility is something that's much more specific to a demographic group and a specific university. It's something about that match between the two. For example, it could be whether someone [00:10:45] like them would fit in at a particular university. And I think distinguishing between these three channels actually is important because they have vastly different policy implications and ways in which we can think about closing these gaps. [00:10:58] So what we do in this paper is exactly trying to understand which one of those uh worlds we're in and how social exposure can shape elite university applications. And in order to answer this, we use two complimentary studies [00:11:12] uh based in the UK. The first is we're going to use administrative data that links together school records uh with university applications and enrollment data and labor market outcomes. And we use quasi random variation in exposure [00:11:26] that students in high schools have to elite universities through what we call breakthroughs, which is the first time a given high school sends a student to a specific elite university. And then we're going to look at the effects of that on subsequent cohorts, university [00:11:41] application, enrollment, and earnings. And in the second part of this project, we conduct a field experiment with 640 university applicants across 21 high schools in England and Wales. And in [00:11:54] this uh field experiment, we randomize social exposure through videos and one-on-one mentorship. And what the field experiment allows us to do is elicit beliefs about different aspects of the university experience and process [00:12:06] such as admissions chances, social fit, and graduation. And we collect realized application outcomes directly from schools through data use agreements. [00:12:17] Now, what do we find? Well, first we find that those breakthroughs, so when a student at a given high school goes to an elite university for the first time, that persistently raises applications to that particular elite university in [00:12:30] subsequent cohorts. But interestingly, we find no spillover to other elite universities. So, for example, if a high school sends a student to Cambridge University for the first time, we see applications to Cambridge University go up, but not to Oxford. And we find that [00:12:44] these increases in applications are largest among students in the subsequent cohort who are demographically similar to that first student who broke through. [00:12:53] And we also find that these induced applicants, they're able to earn £4,000 per year more relative to match controls by age 27. So they seem to be seeing labor market returns. Experimental exposure similarly raises applications [00:13:07] uh to elite universities. And we find that our effects in our field experiment are driven primarily by low SCEs students which we define as either lowincome or firstgen. Those students become 17 percentage points more likely to apply to a treatment university. Uh [00:13:22] we find no effect on high SCES students. And what's interesting is that we find that even something like watching video clips of elite university students talking about their experience it is able to shift applications which really points to the presence of an information [00:13:36] friction. Right? they simply can't be about the desire to be colllocated with a sibling or with a friend. And finally, consistent with their administrative data results, we find that the effects are driven by students who are most demographically similar to the mentors [00:13:50] and the video speakers. And finally, we find that the information frictions are primarily around social fit and sense of belonging at an elite university rather than about admissions chances or labor market returns. So in our baseline [00:14:04] surveys, we find that the low SCES students are relatively pessimistic or have believe have perceive a lower probability of fitting in at elite universities. Uh and these low SCS students, however, are very well calibrated on the typical grades at [00:14:18] elite universities and their admissions chances. We don't find any gaps at baseline there. And our exposure treatment shifts beliefs about social fit at elite universities among will assist students rightward but has no effect on beliefs about other aspects of [00:14:32] the process such as admissions chances and graduation probability. And this paper contributes to three strands of literature. Uh the first is the uh literature studying the effects of family members, neighbors and uh schoolmates on university choice. Uh we bridge that literature with the [00:14:47] undermatching literature in return university. And this paper also speaks to the literature on social learning and hurting. And our contribution here is really providing novel experimental evidence on the underlying mechanisms and in particular highlighting [00:15:01] understanding whether someone like them would fit in at an elite university as an overlooked mechanism behind undermatching. [00:15:09] So we think that the UK is actually part uniquely well suited to tackle this question for several reasons. So first uh university applications in the UK are centralized. They're all running through UKCast. This has data advantages, but it [00:15:23] also reduces application frictions. You know, these students don't have to navigate the University of California system, the University of Texas system, Common App, and places like MIT that aren't even on the Common App, right? [00:15:33] It's all in one place. And admissions are largely test based. Uh not entirely, but largely. And so there's limited room for things like extracurriculars mattering. And it's a very high stakes decision. So the number of applications [00:15:46] that students can submit is capped at five applications. So you know you can't just kind of throw in another application to Oxford. It's actually quite costly within your portfolio. [00:15:57] And as I mentioned earlier, tuitions are uniform across all the universities in the UK, at least within England, uh for English students. And there's also pretty standardized loan processes. It's deducted automatically uh from your paycheck based on how much you earn as [00:16:11] you know bundled with your taxes. It's a very different process than what students will have to navigate in the US. And finally, uh we see dense schoolto university pipelines that we think can restrict the kind of social [00:16:24] exposure to elite universities. And that on this uh figure on the right here, which shows a map of England shaded by um the share of students in each of these uh regional areas uh who attend schools that again have sent no students [00:16:37] to any of the 10 most selective universities in the past three years. [00:16:41] And you can see quite a bit of heterogeneity in the darkest uh shaded regions. It's at least 30% of the students who just aren't encountering these people at school. [00:16:51] And we use the UK administrative data that allows us to link together uh data from UKcast which covers the University of University applicants to the NPD which has all the high school records to the HEISA which lets us see whether an [00:17:04] individual drops out or graduates with a degree and when they do so and finally to link those to the HMRC which is um uh earnings data coming from the tax records. So it's it's really great data. [00:17:14] you can only access it in person in the UK. Uh but my co-authors and I uh think it's quite worth it. So uh yes, >> I mean like curiously why it seems like that's the that's the the [00:17:29] crazy friction here, right? So people and they don't know what the chances are. So seems like it might not be a random place to >> It's true. True, but I mean the [00:17:43] undermatching has been documented in the US and Chile in in a variety of different contexts and so we think there's maybe something else at play as well and actually it used to be capped at six before 2008 and then they they kind of made it fine even further. Um [00:17:57] you know it's possible that's suboptimal but you know that that's kind of the world we're in. Uh okay and then finally we use uh survey and experimental data uh from 640 uh applicants at 21 schools. Um, one of [00:18:11] our schools is in Wales, so there that's not included in this map here, but on the on the yellow dots are kind of showing you we have a pretty good uh geographic spread in our experimental sample. Uh, and if you just compare our RCT sample to the kind of universal [00:18:24] university applicants, you'll find that our sample is more socioeconomically disadvantaged and more academically uh prepared, which you might think is exactly uh the type of students we're interested in. [00:18:37] Okay, now this brings me to the breakthroughs. What happens upon a breakthrough? The empirical design for this part of the project is actually very simple. So, we're we're interested in looking at the sets of high schools that haven't sent a single student to [00:18:50] some given elite university, for the last 3 years. Uh the results are pretty robust. If you change that window to five, four, seven, and so forth. Um and then what we're really using is this quasi random variation in the exact [00:19:04] timing of the breakthrough when that first student goes to that university. [00:19:08] And so we run this um event study specification uh both you know for particular universities and we also will show you some results from a stack version that pulls pulls together uh 10 elite universities um and we also we exclude [00:19:22] the breakthrough student themselves and we're going to look at out um outcomes such as uh where an individual applies in rules and so forth. And what we find here is that elite um that the effects are very university specific and have no [00:19:36] spillovers to other universities but it does induce a breakthrough. Yes. Can >> I ask this Cambridge ox you can you only apply too it's kind of mechanical and be nice to see it for the other >> I'll show you for the stack version for [00:19:51] the 10. You also see it for LSC Imperial UCL. Yeah, you'll see it for a lot of things. It's true, it's mechanical, but at the same time, I think what we can learn from this is, you know, I think it's suggestive, at least at this stage, that, you know, if you think that [00:20:05] students think that, oh, my A's aren't good enough cuz I need to be really special. But if they now think that they're good enough to apply to Cambridge, which is what you see once a school sends someone to Cambridge, applications to Cambridge go up, you might think that a student would also think that that would apply to Oxford, [00:20:20] right? I think it's quite surprising that you see no effect to Oxford, but it's also robust to other universities. [00:20:25] Yes. What do you observe the high schools for instance? Do you know if the head of the high school that is connected to Cambridge and that's why >> I'll I'll get to that shortly. We can't observe that in the data right now. Um you know that's that's one possible explanation. Uh yeah and I'll I'll kind [00:20:40] have a slide that kind of talks about that more explicitly matter that it's driven by uh I've hired a new teacher at Cambridge in my high school and that's why our our for what [00:20:54] you want to say it doesn't matter why the breakthrough happens. Um I mean I think on the one hand yeah we we think that you know once a breakthrough happens the point is that there's you know through our RCT we show that there's this kind of mechanism where students learn from that student about [00:21:08] what it's like to go there and it can draw them in and so in some sense that might not matter. It's this idea that some breakthrough caused by whatever force can have these dynamic effects. Uh but at the same time I think the underlying um mechanism is also informative like of whether you know [00:21:22] it's really coming from the teachers. I think that our experimental evidence kind of suggests that it's not coming from teachers. Um, now there's a lag in this. There's another figure that's somehow not popping up anymore. Um, that [00:21:36] I don't know why it's not popping. Oh, there it is. We got it back. Um, this is just showing you symmetrically if you look at the sets of schools that send someone to Oxford for the first time. [00:21:44] Similarly, you'll see increases in applications to Oxford but kind of no effects in um in Cambridge. Okay. And as I mentioned, yes. Do you know colleges review? [00:21:57] >> We don't observe that data. You'd have to get that directly from Cambridge and Oxburg. Yeah. Um and you know this is just to say um where are these applications being drawn from? Kind of consistent with the figure I showed you earlier. Well, first in blue here, this is generalized. This is stacked across [00:22:11] the 10 universities. So you'll see that it applications to one of those to the UN breakthrough university itself increases. we don't see spillovers to the same tier um universities that are kind of equally selective um and it's really pulling applications away from [00:22:25] lower ranked universities. So that's kind of >> does increase that >> we've done versions of this excluding it as well. you'll you'll find um the same [00:22:37] effects and um crucially this kind of effect of being induced into apply uh is amplified by individuals who are either demographically similar to the breakthrough student or uh we call these kind of some uh proxies for like social [00:22:51] proximity. So whether or not they were in the same high school for several years and the same you know school even before that whether or not they took the same subject um and that kind of we think can speak to some teacher based mechanisms uh which I'll get to uh [00:23:05] momentarily but all this is to say you know it seems like there's also some suggestive evidence that it's demographic similarity. Now of course this could simply be capturing the fact that individuals are more likely to be friends with people who are demographically similar to them. Right? [00:23:19] we don't know if they're learning something that's demographically specific about that university yet based on the administered data results. Now I want to really briefly touch on kind of uh threats to vidity and potential confounders which I think like we [00:23:32] discussed. So the first is the potential for a common school level shock. This could be like the school suddenly got a bunch of new funding or they got a very inspirational high school counselor something like that. Well our results are pretty robust to controlling for uh test scores and other you know [00:23:47] composition effects. we don't see any abrupt jumps that coincide with the breakthroughs in test scores and also the fact that these are university specific kind of rules out that they just got a great high school counselor a more challenging confounder is a university specific school level shock [00:24:01] which I think is uh what one of the questions was addressing what if a teacher comes who has a really good connection or a lo a university started doing an outreach program um we have a few uh pieces of evidence that we think are kind of help rule those out one [00:24:16] university outreach which typically happens uh geographically nearby and we find that the effects are not concentrated among schools that are close to the breakthrough university. [00:24:25] And second, we find um you know it's a smaller effect but we still find a statistically significant effect for students who never even took the same subject. So that means you know they wouldn't have even interacted with the same teacher. um you know in a few [00:24:39] months my co-author uh will have access to teacher level data that's not um accessible to public researchers but he'll have access through a seconment at the IFS which would let us in principle also track teachers moving across but all of this is to say I think our [00:24:54] experiment is able to kind of address these as well yes >> I think on the common school level shots uh I mean like if you're just looking at this results robust for test scores if you're looking at averages when we're focusing on the like narrow right tail [00:25:07] Then I think it's like you're able to have some other measure that micers whether that's changing that would be good or >> like whether there was someone with a bunch of A stars or something like that on the ambitious teacher uh avenue. I I [00:25:20] don't think that you know if if a teacher sends a student to MIT for the next four years they're going to be thinking ah I sent a student to MIT. So I think that would I would still sort of expect to see some university specific effects there [00:25:34] >> and for an effect to last through the social connection just between students but that to last 24 years you really have to have this like chain of student that's going through >> right because like two students that are [00:25:47] next to each other in grade will maybe but so the persistence of that I think kind of speaks more from a school level shock. [00:25:57] >> Okay. Um I think your experiment speaks to like the one mechanism of the student and I think that's really nice. Uh but I but I wouldn't rule out that there's uh sort of this happening as well. [00:26:06] >> Yeah. No, I think that's that's a really great point and I think understanding more the roles in which teachers can also kind of interact with students differently in response to breakthroughs is is a really uh important avenue for future research. And so now we're interested in understanding what happens [00:26:20] to these students who are induced into applying to these elite universities. [00:26:23] Um, it's not that they just apply and these applications are wasted because they don't get in and they don't go. [00:26:28] They actually get offers at pretty high rates. They enroll and you can see if you look at the green square in the black diamond, essentially all of the students who enroll graduate. So, we don't really find a kind of evidence that these students were mismatched uh [00:26:41] in any way. Um, this isn't this isn't perfect, but we've done a few different uh uh approaches to understanding the labor market returns. So we did um match controls in two different ways. Uh in on [00:26:54] the left we matched u these kind of students who were induced in subsequent years to uh students from control schools and we match them based on demographics, test scores, major choice and in the same year. On the right here we do another type of matching. We match [00:27:08] them to prior cohorts uh within the same school again based on test scores and demographics. What's nice is that the results look kind of similar across both of these ways of doing matching. And what we really see is pretty large uh returns. It's around £4,000 a year by [00:27:23] age 27. Yes. >> What is the count of actual school that they built in the sense? I know the applications to low tier schools went down, but my guess is the school that they the contractual school that they [00:27:35] will graduate from is probably not that much worse than Cambridge or better. [00:27:40] >> Yeah, it's it's a good question. Um I mean I think this suggests that it is. [00:27:45] Um but we haven't done we haven't looked at like the ranking of the kind of average uh matched uh control student and that's something that I think uh we we plan on doing. Um we've also done a Dale and Krueger analysis which is kind [00:27:58] of like an additional robustness to this uh just looking within our data and what we find is that it's actually quite similar. we find there's like a large return after you control for application and offer set fixed effects to the top [00:28:11] seven or eight universities and then it's flat after that. And so that kind of gives us a sense too what the counterfactual might mean. We find that the average returns to age 27 for those top seven is around 4600. And so that that's actually really consistent with [00:28:24] this as well. Um and I think that suggests that the counterfactual is maybe in the 10 to 20 or 20 to 30 range. [00:28:31] good schools, but there seems to be a really big uh right tail elite university premium at least in the UK. [00:28:36] Okay. And so this brings us to the question, well, okay, why did these students forego these labor market benefits and you know being the ability to even graduate from one of these elite universities before the breakthrough? [00:28:48] And what changed after? Of course, one possibility is the comfort of attending a university where they already have a friend. Um, another possibility is new information. But if that's the case, what's the new information? Is it about admissions requirements? Is it about [00:29:03] applications processes, returns, or something that's demographically specific? And second, why are the effects stronger for students who are demographically similar to the breakthrough student? Uh were those effects and interaction effects simply capturing the fact that you're more likely to be friends with someone who's [00:29:18] demographically similar, or is it something about a demographic specific uh information friction that's being alleviated? [00:29:26] And we answer kind of those questions through our field experiment and which randomizes social exposure to elite universities. So beginning in January 2025, we reached out to um a bunch of high schools and then these schools kind of opted in. We ended up with a final [00:29:41] sample of 21 high schools and these schools distributed baseline surveys to their students which elicited things like you know their beliefs about um specific universities, where they want to apply to and so forth. And when they take that survey, students are [00:29:54] randomized into either treatment or control. Um uh and um then at the end of this um baseline survey, there's also an embedded video clip that they're going to watch. I'll get into each of the treatments uh in more detail shortly. [00:30:07] Then all students both in treatment and control attend an inschool workshop uh led by um you know, one of our our volunteers from our nonprofit organizations. And in that at the end of that uh in-person workshop, they also do [00:30:19] a midline survey. And then the students in the treatment arm are further randomized into two treatment subarms. [00:30:26] Subarm T1 is going to match students to demographically dissimilar mentors. And these mentors are, you know, current university students. Um and then subarm T2 will match students to demographically similar mentors, right? [00:30:39] And the control students, you know, continue to have uh neither mentors, no videos. And then in October 2025, uh we had students fill out an endline survey. [00:30:48] And in March 2026, we collected realized university applications data from our high schools. Okay. Now, first, as I mentioned, all students are going to do this in-person workshop at school. So, it's an active control. It's around 1 [00:31:02] hour in length. It's led by a current university student preparing material, presenting material prepared by the research team. It covers things like application processes and admissions. [00:31:11] Part of why we did this is to convince the schools that every student's going to get something out of it. But we were actually pretty intentional about what we put in there. So this is just two. [00:31:20] It's it was a much longer deck than this, but kind of showing you two kind of slides that are informative. One, you know, we showed them what the kind of average earnings um 3 years after graduation at various universities are. [00:31:31] And we al also showed them kind of the statistics on the various test scores that students typically have at universities. We told them also what a 25th percentile means and just trying to kind of give them information. And the reason for this is that we wanted our [00:31:44] RCT treatment effects to capture the impact of social exposure of over and above that of simply acquiring uh information about admissions and and earnings. Okay. And so what are the video clip treatments? So participants are matched to videos based on their [00:31:58] test scores. So we didn't show every treated student a video clip from University of Cambridge. We thought that that would if anything that might be discouraging. Uh so there were kind of four tiers of video clips from our pool pools of videos and we made the match [00:32:11] kind of aspirational but within reach. Uh each student will watch two video clips um that are around four minutes uh in length each and it'll discuss the speaker's experience at university and application process. Uh we told all the video speakers to kind of tackle a [00:32:25] certain set of points. They'll they'll introduce their themselves. They'll say where they're from, where, which high school they went to, what classes they took, grades they got, where they applied, got offers, and so forth. Um, and this is meant to really replicate at scale what a student could learn by [00:32:40] talking to a schoolmate, right? And so you can see here, this is kind of an example of two of the video students will see Yua from um, Imperial College London, Peter from University of Cambridge. But this exposure is passive rather than relational, and it helps us isolate the role of information, right? [00:32:55] These these video speakers aren't actually friends with the students. they have no way of kind of contacting them. [00:33:00] And at the same time, it's information though that's different than the information in the previous slide which is just like giving you labor market returns. You get to hear a person's accent. You get to see them. It's a real person. You get to see how they're talking about their university. [00:33:19] >> Yeah. So, >> it's a good question. Um, we kind of recruited these through a number of different nonprofit organizations. A lot of them are actually from more disadvantaged backgrounds just because those are the types of students that [00:33:33] volunteer with like the sudden trust, right? Um, yeah, that's right. Uh, so I think it was like the majority of them are firstg for example. Uh, okay. And then the the mentor treatment in arms T1 and T2. So students were matched with up [00:33:47] to two university mentors from our pool of 245 mentors. So that allowed us to kind of do this demographic match versus not match. This is an example. It's a pretty intense treatment. Like this mentor reached out to a student being like, "It was great meeting you on Wednesday. Um, I'll email you to [00:34:01] schedule another meeting. In the meantime, keep the questions coming." It's a pretty strong treatment and it's also relational, right? This is different than the videos. You're actually interacting with this person. [00:34:11] It's plausible that you can become friends with them. So this is kind of what I wanted to highlight about that. [00:34:15] Okay, now we're going backwards. Okay, so what do we find? What we find is that experimental exposure raises applications to the treatment universities. Uh but and so here on the left I'm going to show you the ITT. The [00:34:28] toot is just kind of sc it's multiplied by a factor of two essentially because our take up was around 50%. Um but on the left here you see it's around a 14 percentage point increase overall students in whether or not you apply to one of the mentor or video universities. [00:34:42] The mentors and videos were different universities. So we're kind of able to um estimate those effects separately. [00:34:48] Um, you see it's about 10 percentage points in the video, 10 percentage 11 percentage points to the mentor university pooling across T1 and T2. [00:34:57] What happens if you split up T1 and T2? The all of the effects are coming from the demographically matched mentor arm. [00:35:04] And this is actually quite surprising because in the previous slide I showed you how engaged and intense that mentor interaction was. And so really this really points to there's something about the demographic match that's really important for the students. [00:35:18] >> What are the demographics you use? >> The demographics that we match on are gender, ethnicity, and region. [00:35:25] >> But there's no E which I guess >> there's no income. Yeah. Yeah. We didn't we didn't m >> sounds like it's like a role model effect more than maybe an employee like sorry. So like students could have [00:35:39] perfect information >> but role model effects might increase their aspirations and that could backfire if you only have five places you're allowed to do and you're like you over you're like oh someone got in I can [00:35:53] get into Cambridge too means I take off another from my list but you don't end up getting because you're actually not that good >> if it's a role model effect things could go all right but >> if it was just p it was a pure [00:36:06] information effect then usually that's like I'm lowering frictions and >> exactly >> we're better off. So is there a way to distinguish how much of it is a role model thing rather than what you think >> I think yeah so I mean I'll show you kind of how beliefs evolve um so that kind of points to something going on [00:36:21] with information acquisition to in in the experiment and kind of to speak to your point they think the administrative data results really show that kind of even at these schools that previously nobody was going to like Cambridge once they start going they they graduate and they see labor market returns so we're [00:36:35] not really seeing evidence for like students just kind of throwing in applications and then losing a safety school and going on match. We don't really see that. Yeah. [00:36:42] >> If I understand well in the UK you also select the subject you apply. [00:36:45] >> That's right. Yeah. >> Do you have results whether they select subject the mentor also of study that would be like the aspirational one? [00:36:52] >> We've collected that. We haven't actually looked but I think you know because students can only they would usually taking three A levels like you know if you didn't take calculus in high school you're not going to suddenly become an engineer. I mean that would be very impressive. But I think you know for the most part what we see is that in [00:37:07] our in our baseline survey students aren't changing their their majors. So I mean we haven't actually estimated the treatment effect but we don't find a lot of shifts. You know they might shift from political science to econ or something. [00:37:19] >> Yeah that I think at that granularity we could look at it but we're not going to see huge we don't expect to see huge shifts. Yeah correctly. So you're only finding effects for the mentor of video [00:37:33] university but like not for general application to all universities. [00:37:37] >> Yeah. Well, because these are um we kind of screened out students who weren't interested in applying to university. [00:37:42] Right. This is this is targeting students who want to apply to university and we're trying to understand which like elite university they're choosing um and whether or not they're going to apply to an elite university not so much the extensive margin of >> measure like I I'm not familiar with the [00:37:56] UK system like uh application intensity like number of places they apply >> and even in national sample like I think over 85% of the students apply exhaust the full five so we're not seeing a lot [00:38:09] of um yeah But they also die as opposed to match to Cambridge first week they also now start planning per what does the mentor university [00:38:23] >> yeah we haven't we haven't looked at that explicitly um but I mean I think the in the breakthrough results and the administrator suggests that it's not going to spill over in that way right um and I will show you I mean so we have belief effects for example so none of [00:38:37] the video universities were Oxford and so that's why we elicit beliefs about Oxford as like a bench mark and we don't see any effects in beliefs about Oxford shifting. So, but you know students were watching videos from Cambridge. I I might hold on defer on that question um just in the interest of time. Sorry, [00:38:51] there's always a question in the back over there. Um and so splitting these effects um by SCES you see there's there's you know the the point estimate is positive for high SS students as well but we do see that you know it's it's much larger and statistically [00:39:06] distinguishable from zero among the low SCES students right and so that brings us to the mechanisms so what's going on here right I think a role model effect is definitely at play here but I'm going to show you that there's some evidence for information frictions being [00:39:20] alleviated as well so at baseline Um we elicited from students as I mentioned beliefs about um so the questions specifically were phrased as what is the probability from 0 to 100 that you would socially fit in and make friends at [00:39:34] university X. The probability that you would receive an offer from university X if you applied and that you would graduate successfully if you enrolled. [00:39:40] And we asked this for Oxford as a bench as a benchmark elite university as well as for the treatment universities for each student. At baseline, after you control for test scores, what you see is that low SCEs students defined as either low-inccome or firstg, but again, those [00:39:53] those are all kind of similar. Um, they kind of perceive lower probabilities of fitting in and making friends, but not of receiving an offer or graduating successfully, right? So, that there there's something going on there. Um, and to kind of be very explicit about [00:40:07] this, we even asked them because there's a ground truth measure of what the test scores are at a place like Oxford. We ask them what are the grades, you know, of students who typically go to Oxford. [00:40:16] Students are pretty well calibrated. If anything, the low SCS students are slightly underestimating the grades needed to go to Oxford. But um it's actually perhaps rational given the contextual offers that are being used by elite universities or if you're firstgen [00:40:29] or from a low-income neighborhood. Uh you are they're you're explicitly told you can get in with lower grades. So I mean they seem pretty well informed. [00:40:36] This is baseline before we even showed them any any information at that workshop. And this is I might skip this in the interest of time but we know we if just kind of anecdotally what students are learning from videos. Uh it's interesting to hear what they'll [00:40:50] say like I was surprised by how diverse Cambridge is because of its exclusive reputation but then there was another person who said it confirmed my suspicion that Cambridge is not easy to get into. So you know kind of kind of a sense of what students were saying once [00:41:03] they watched the video and now quantitatively what happened. So on the left here it's a histogram on the right here we are showing you the CDF. These are belief shifts from between baseline to midline among low SS students on that [00:41:16] belief. Um the probability that they would fit in at an elite university and make friends there. Uh and you can see it it there is like a statistically significant distributional shift right words in these beliefs in the treatment [00:41:27] group relative to the control. Uh but interestingly we find no changes in beliefs about admissions probability, no changes in beliefs about graduation probability. um we look at Oxford University as that kind of placebo [00:41:42] university. We find no effects on any of the dimensions of the beliefs about Oxford. So it's not spilling over. Um and then we also find no effects on high SCES students. Okay. And you can also we are in our paper we also have you know [00:41:56] regression table versions of these looking at mean updates, positive updates, negative updates. [00:42:03] And you know this is again kind of more descriptive but we surveyed the mentors after their conversations with students and asked them to kind of select from a list uh which topics came up in the conversations. Um in 103 conversations [00:42:17] um that mentors reported in our survey. You can see that the kind of modal topics that were discussed was student life at university and fitting in. So in 70% of those conversations, mentors indicated that that's what they talked about with students and you know careers [00:42:32] and succeeding academically. It's shocking how how low that is and we were quite surprised by that. And so you know maybe lows students are right to be pessimistic, right? Who's to say? [00:42:43] There's some there's sociological evidence showing that these students might go and in fact have a really bad time there. We don't have a ground truth for that in our experiment. But you know we think the fact that one that the beliefs do become more optimistic in response to treatment suggests that they [00:42:57] were maybe overly pessimistic before and second in our administrative data results we found that those breakthrough effects persist right and so if their schoolmate went and had a terrible time you'd think well you know maybe they would stop updating and so that our preferred interpretation is that [00:43:12] persistent non-attendance at a high school to elite universities can generate pessimism about a demographic group specific component of that experience without distorting beliefs about admissions and returns and that [00:43:23] exposure can change that. So to conclude uh in our paper we use UK administrative data in a field experiment to show how lack of social exposure can deter uh university applications from lows students. We showed you that once a [00:43:37] student breaks through to an elite university, this can induce future cohorts into applying to that particularly university. Experimental exposure through videos and mentors also induce applications to elite universities. Um, the response is larger [00:43:50] among demographically similar students and it seems to operate through information specific to a university group such as would someone like me fit in? And this, you know, from a policy implication standpoint, policies that explicitly lower admissions thresholds [00:44:05] like affirmative action and contextual offers may be insufficient to fully close this gap. And you know if you when you have better peers in your schools there's been evidence that that can also have offset while it can raise aspirations it can also offset that [00:44:19] through relative rank um concerns and teachers can penalize students sometimes and so external mentorship to demographically similar role models offer social connectedness at scale without some of those relative rank uh [00:44:32] you know downsides in some sense. And finally, policies recruiting from disadvantaged schools from the first time can have dynamic spillover effects in future cohorts. And that's something that's important in policy considerations. Um, thank you. Sorry, I [00:44:44] went one minute over >> very much. [00:44:53] Next up, we have Glenn. Um from any notice we as presenters please share your screen uh because there's like going to be a [00:45:05] live stream uh as well. So please share your screen soon. [00:45:25] Sorry, it's just a little complicated to get all the screen sharing done in between every uh speaker. [00:45:50] >> Doesn't look good. There we go. Yeah. [00:45:56] >> Are we all set? Okay. Um, good. Uh, uh, thank you all so much, uh, for coming.