Peer effects and endogenous network formation

Informations projet

PEERS

N° de convention de subvention: 657332

DOI

10.3030/657332

Projet clôturé

Date de signature de la CE 8 Avril 2015

Date de début 1 Juin 2015

Date de fin 31 Mai 2017

Financé au titre de

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Coût total

€ 158 121,60

Contribution de l’UE

€ 158 121,60

Coordonné par

FUNDACIÓ MARKETS, ORGANIZATIONS AND VOTES IN ECONOMICS
Spain

Periodic Reporting for period 1 - PEERS (Peer effects and endogenous network formation)

Période du rapport: 2015-06-01 au 2017-05-31

Résumé du contexte et des objectifs généraux du projet

Strong correlations between the behavior of individuals and their peers have been observed by research groups throughout the world in a variety of contexts. In particular such correlations have been repeatedly observed in academic settings, pertaining a wide range of personal characteristics and behaviors, including academic achievement. One major challenge encountered by this literature from the outset is that it is very difficult to disentangle the part of this correlation that stems from assortativity or selection (the tendency of individuals to build relations with similar others) from the part that stems from actual peer influence.
The main objective of the project was to construct a structural model and estimate it using a very rich dataset related to academic achievement, and study networks in a European higher education institution. Understanding how much individuals influence each other in academic contexts is fundamental for informing policies on academic tracking and optimal group creation.

Travail effectué depuis le début du projet jusqu’à la fin de la période considérée dans le rapport et principaux résultats atteints jusqu’à présent

We developed a structural model of network formation and behaviour with the aim of decomposing the observed positive correlation of academic performance of individuals and that of their peers. The model is a general purpose one in the sense that with appropriate modifications it can be applied to many of the several contexts in which such correlations have been observed. Having said this, several modelling choices were strongly influenced by special features of the two main settings in which we conducted our empirical work : A European elite university (with Giacomo De Giorgi, Isabel Melguizo and Michele Pellizzari) (Dataset 2) and an elite University in Bogotá, Colombia, (with Juan Camilo Cárdenas) (Dataset 1).

The main findings based on Dataset 1 was a positive correlation between how trusting the subjects were, as revealed by the trust experiment, and their closeness centralities in various of the networks of relations that we uncovered via surveys. This finding prompted us (with Davide Pietrobon and Danisz Okulicz) to propose a simple modification of a model of network formation introduced by Jackson and Rogers in 2007, to allow for heterogeneity in agents’ mean willingness to form a link conditional on any given meeting opportunity. Provided that this propensity to form a link is related to agents’ willingness to trust, the model offers a variety of simple novel implications on the relationships between agents’ levels of trust, their degrees in their networks of relationships, the degree distributions of their friends and the distribution of trust among them, their positions in their networks, and in particular their centralities.

Most of the work throughout the action focused on (Datset 2). Using this Dataset we obtained estimates of the structural parameters of our theoretical model,
and in particular were able to decompose the observed correlation in the academic achievement of individuals and that of their peers between assortativity and influence under reasonable assumptions. In addition to estimating the parameters of the model related to this decomposition, which was the main objective of this part of the project, we were able to estimate parameters reflecting the role played in network formation of individual characteristics other than academic skills and performance, such as gender and region of origin.

Progrès au-delà de l’état des connaissances et impact potentiel prévu (y compris l’impact socio-économique et les conséquences sociétales plus larges du projet jusqu’à présent)

The main dataset with which we have been working throughout our project has offered us a unique opportunity to carry out such a decomposition in a specific case. The reason is that a significant subsample of the students in our dataset share the following characteristics (1) They all take the same classes and (2) They are initially randomly assigned to one of several class sections, and randomly reassigned after their first year of study. This exogenous variation in socialization opportunities allowed us to write a model in which the parameters corresponding to peer influence and those corresponding to individual’s preferences for building links with other similar agents are identifies.

Our analysis suggests that most of the observed correlations in academic performance among individuals and their peers is attributable to assortativity rather than peer influence.
Having estimated this mode will enable us in the near future (outside the scope of the action) to compute some valuable counterfactuals, involving alternative distributions of incoming students into class sections according to their academic skills and/or regions of origin.

The tendency of networks to be formed assortatively as well as the strength of peer influence arguably depend on variables that are specific to each context, and thus our findings only have very limited external validity. However, there is no way to augment or knowledge on the subject of peer effects other than carrying out thorough specific analysis. In the near future, and once our findings are published, along with many other studies attempting to answer similar questions in other contexts and using other techniques and natural experiments, meta-analytical efforts will be able to yield more general conclusions.

bydprogram.png