Published on Development Impact

With great help from my friends: spillovers from college aspirations. Guest post by Jessica Gagete-Miranda

This page in:

This is the 12th in our series of posts by PhD students on the job market this year.

Take a moment to think about your friends from school. Remember that friend anxious about the exams, who would always invite you to study? Or maybe that friend who would always play it cool to avoid being seen as a geek? Did you have friends that were worried about getting a job soon so they would have money to buy nice clothes? How much do you think these friends influenced your trajectory after school? Did they motivate you to keep studying? To follow your dreams? To go to college? Or, perhaps – given that you are the one reading this World Bank blog post – how much did you influence your friends' future trajectories? 

When writing my job market paper, I was interested in these questions. More specifically, I wanted to understand how much students' aspirations spillover to their friends. The literature on aspirations shows how crucial such a trait is in determining human capital accumulation. It also shows the importance of peers, or similar others, in determining one's aspirations (see, for instance, Genicot and Ray, 2017). Most of the literature on aspirations formation investigates how peers' socioeconomic status influences individuals' aspirations (Janzen et al., 2017). My paper asks a different question: how much do peers' aspirations influence individuals' aspirations, above and beyond socioeconomic considerations? That is, if two individuals are from the same socioeconomic status, but one of them aspires to have more, how does the interaction between them shape their aspirations?  

In my paper, I exploit novel data from middle schools in Brazil to investigate how friends' aspirations toward going to college influence students' aspirations toward it. College aspirations are quite a relevant measure of educational aspirations in developing countries. On the one hand, tertiary education in these countries carries a high earnings premium compared to other OECD and partner countries. On the other hand, low percentages of adults in these countries attain such a level of education (see, for instance, Figure A6.2. of this OECD report).


Challenges in measuring peer effects 


Identifying causal peer effects is quite a challenging exercise. Two problems, in particular, have received significant attention from the literature. The first is the reflection problem (Manski, 1993), which is a simultaneity that emerges when investigating how my friends' outcomes influence my own outcomes. The main problem here is that I influence my friends while my friends influence me back. So, it is hard to disentangle who is influencing whom. The second challenge is the endogenous formation of friendship. We all know that we do not just pick our friends at random: there is a particular reason for choosing each one of them. Moreover, homophily – people's characteristic to click with similar orders -- is a primary driver of friendship formation. Hence, if we see two friends similar to each other, it is hard to know if they are similar because they are friends or friends because they are similar.


Addressing the challenges


My identification strategy leverages network structures to address these challenges. The data comes from a survey conducted on students enrolled in state-operated middle schools in Sao Paulo (Brazil), combined with administrative data. One block of questions in the survey mapped students' social networks. They were asked to name their four closest friends or colleagues in their grade. Another block of questions sought to understand students' educational aspirations. One specific question asked how many years students would like to study if this choice were entirely up to them. I use this question to identify students' aspirations toward pursuing a college degree.

Figure 1: Example of friendship links

Figure 1

My identification strategy builds on the works of Bramoullé et al., 2009, De Giorgi et al., 2010, and König et al., 2019. To address the reflection problem, I exploit the fact that not everybody is connected to everybody else in the network, and I use the characteristics of second-order connections – that is, friends' friends -- as instrumental variables for friends' aspirations. To understand this strategy, think of three students in a school: Lisa, Ana, and Tina, as shown in Figure 1. Imagine that Lisa is friends with Ana, and Ana is friends with Tina, but Tina and Lisa are not friends. Hence, Tina will influence Lisa only through the influence that she has on Ana. If this is the case, one can use Tina's characteristics – such as her race, parental education, etc. -- as instruments for Ana's aspirations and, as such, disentangle the impact that Ana's aspirations have on Lisa's aspirations. [1]  

This strategy would suffice if we did not have the endogenous formation of friendship. However, as said before, friends are not chosen at random, and there might be a particular reason why Lisa and Ana are friends, but Lisa and Tina are not. This would make the instruments themselves endogenous. To address this last challenge, I implement a step before building the instruments: I model friendship formation based on homophily in pre-determined characteristics and students' random interactions. Essentially, I estimate the likelihood of two students in the same school being friends depending on how similar they are in terms of gender and race and whether their first name starts with the same letter -- classroom allocation in the first grade of middle school is based on students' first names' alphabetical order, and so is within-classroom allocation of seats at each grade of middle school.  I then get the predicted friends’ friends coming from the model.

Hence, instead of using friends' friends' characteristics as instruments for friends' aspirations, I use the characteristics of predicted friends' friends as instruments for friends’ aspirations. The main idea behind this substitution is that I'll use as instruments only second-order connections that emerged due to exogenous circumstances. 


Friends influence not only your aspirations but also your school trajectory


My results show that friends indeed influence students' aspirations: An extra aspiring friend – that is, an additional friend that aspires to go to college -- increases, on average, the likelihood that a student will also aspire to it by 11.25 percent. Such an impact is quite sizable. For instance, it is about 30 percent greater than the impact of moving students up by one standard-deviation in the reading performance distribution.  

I investigate several mechanisms through which such an influence could be happening. I find that students' who comply with some harmful social norms -- such as the fear of being seen as nerds or the urge to start working soon -- also influence their friends to abide by these norms. Likewise, students' that put more effort into studying also influence their friends to do so. 

Finally, I look at students' future outcomes and find evidence that an extra aspiring friend decreases an average of 28.27 percent the likelihood of students' dropping out of school before middle school graduation and, therefore, directly impact a student's chances to actually attend college. This finding is particularly important for developing countries where school completion rates are significantly low (see, for instance, Figure 2.2 in this World Bank report). 

Several works show how some educational interventions increase students' aspirations (see, for instance, Carlana et al., 2018). My results highlight positive unintended consequences from these interventions, since they might not only increase the aspirations of non-targeted students, but also directly influence their outcomes in school.  

Jessica Gagete-Miranda is a Ph.D. candidate at Bocconi University




[1] It is important to highlight that friends’ friends’ aspirations should not be used as an IV; otherwise, we would again incur the reflection problem.

Join the Conversation

The content of this field is kept private and will not be shown publicly
Remaining characters: 1000