Interesting paper on peer effects

I'm not sure this is the ideal methodology to approach this problem (I'd like to see this combined with something that captures the actual interactions in the classroom), but, based on what I've seen, this is a definite improvement.

From VoxEU (via Thoma)

Our recent research uncovers peer effects in education as distinct from the contextual and other correlated influences. Our econometric strategy uses the topological structure of social networks as well as network fixed effects to identify each of these effects separately.

Our analysis is made possible by the use of a unique database on friendship networks from the National Longitudinal Survey of Adolescent Health (AddHealth). The AddHealth database has been designed to study the impact of the social environment (i.e. friends, family, neighbourhood and school) on adolescents' behaviour in the US by collecting data on students in grades 7-12 from a nationally representative sample of roughly 130 private and public schools in years 1994-95 (wave I). A subset of adolescents selected from the rosters of the sampled schools, was then interviewed again in 1995-96 (wave II), in 2001-2 (wave III), and again in 2007-2008 (wave IV). For our purposes, the most interesting aspect of the AddHealth data is the friendship information, which is based upon actual friend nominations. It is collected at wave I, i.e. when individuals were at school. Indeed, pupils were asked to identify their best friends from a school roster (up to five males and five females). As a result, we can reconstruct the whole geometric structure of the friendship networks.

In Calvó-Armengol et al. (2008), we exploit such information to test a peer-effect model which relates analytically equilibrium behaviour to network location. This analysis shows that the structure of friendships ties is an important, and so far unnoticed, determinant of a pupil performance at school. In Patacchini et al. (2011), we follow this line of research by exploiting the other AddHealth waves in order to investigate whether such effect is carried over time. Indeed, the longitudinal structure of the survey provides information on both respondents and friends during the adulthood. In particular, the questionnaire of wave IV contains detailed information on the highest education qualification achieved.

We analyse the impact of the friends' educational attainment on an individual's educational attainment where they are identified as friends during school and in to adulthood. We find that peer effects in education are not only strong but also persistent over time. We find that the most relevant peers are the friends people make in grade 10-12, from when they are around 15 years old. This suggests that individuals are more likely to work towards and apply to college if this choice is popular among their peers, especially in the last years at school. This could represent the effect of contagion and collective socialisation and mean that any education policy targeting specific individuals will have multiplier effects.

The heroin's still doing the heavy lifting -- why Ivy League legacies work

From Christopher Shea's Boston Globe column:

Richard D. Kahlenberg, editor of the forthcoming book "Affirmative Action for the Rich: Legacy Preferences in College Admissions," points out that universities in other countries don't give so-called legacy preferences to sons and daughters of their alumni. (Even Oxbridge colleges don't, despite the class-bound history of British education.) So, he asks, why on earth do we do it in America?

Broadly speaking, students go to college in search of four things: certification; instruction; reputation; and connections.

In terms of certification, any well-accredited school would do. In terms of undergraduate instruction, the best deal for the money (and perhaps the best deal period) is the small four-year school. (I'm leaving this as an assertion but I'm fairly confident I can argue the point if anyone wants to debate.)

In the next two categories, however, the Ivy League cannot be surpassed, in part because of the legacy system.

Without loss of generality, look at Harvard. The student population of the school consists entirely of two overlapping groups: people who can get into Harvard; people whose parents can get them into Harvard.

The first group is hard-working, ambitious and academically gifted. Assuming the number of need-based legacies is trivial, the second group comes from families that are wealthy (they're paying for a Harvard education) and well-connected (at least one parent went to Harvard).

Putting aside luck, you can put the drivers of success into three general categories: attitude, drive and work habits; talent, intelligence and creativity; reputation and connections. It is possible to succeed with just one of these (hell, I can think of people who made it with none), but there is a strong synergistic effect. A moderate talent who works hard and has connections will generally go farther than a spectacular talent who's lazy and isolated.

Connections are governed by the laws of graph theory. I'm not going to delve too deeply into the subject (since that would require research and possibly actual work on my part), but as anyone who has read even the cover blurbs on Linked or Small Worlds can tell you, adding a few highly connected nodes (let's call them senator's sons) can greatly increase the connectivity of a system.

It would be interesting to model the trade off between picking a well connected legacy over a smarter, harder-working applicant given the objective of producing the greatest aggregate success. Because of the network properties mentioned above, it wouldn't be surprising if the optimal number of legacies turned out to be the 10% to 15% we generally see.

Optimized or not, this mixture is almost guaranteed to churn out fantastically successful graduates regardless of what the schools do after the students are admitted. I'm certain the quality of instruction on the Ivy League schools is very good, but, like most education success stories, the secret here is mostly selection and peer effects.

Update: For a different interpretation (this time with actual data), check out this post at Gene Expression.

Updated update: Why doesn't spell check work in the title field?

"Ignore the parts about crystal meth and pancakes"

The education reform movement relies heavily on anecdotes of remarkable, odds-beating schools, but when you take a close look at those schools that had significantly superior performance, some if not most of the difference in scores could be explained by selection and peer effects.

This isn't to say these schools weren't benefiting their students. Regardless of the reason, these kids were better off. Nor is this to say that these schools weren't doing something right. I can tell you that many are well-run and highly innovative.

But even taking all of that into account, selection and peer effects are huge and can swamp almost any other factor you can think of. These effects are seldom if ever adequately accounted for (And before anyone says the word 'lottery,' please take a look at this). This makes it all but impossible to accurately measure the impact of these schools but people like Jonathan Chait continue to cite them without any caveats.

I came across a segment of This American Life that beautifully captured my feelings on the subject. Just play the clip below and every time you hear 'heroin,' substitute in 'selection and peer effects' (you can just ignore the parts about crystal meth and pancakes).

From Kumail Nanjiani:



So remember, selection and peer effects are doing the heavy lifting.