Bayesian Analysis of Multilevel Probit Models for Data With Friendship Dependencies

Johan Koskinen, Sten Åke Stenberg

Research output: Contribution to journalArticlepeer-review

Abstract

When studying educational aspirations of adolescents, it is unrealistic to assume that the aspirations of pupils are independent of those of their friends. Considerable attention has also been given to the study of peer influence in the educational and behavioral literature. Typically, in empirical studies, the friendship networks have either been reduced to some summary measure or, when the complexity of the social structure has been used to its full extent, the studies have been limited to single networks, thereby effectively neglecting the context of the studied network. The authors propose a multilevel probit model with correlated responses and a Markov chain Monte Carlo-based inference scheme for analyzing multilevel data where there may be dependencies not only within levels but also between peers that are directly relationally tied. The authors describe the procedures in the context of the illustrative case of the Stockholm Birth Cohort, and the decision to proceed to higher secondary education for 12,660 pupils in 619 school classes in Stockholm. © American Educational Research Association 2012.
Original languageEnglish
Pages (from-to)203-230
Number of pages27
JournalJournal of Educational and Behavioral Statistics
Volume37
Issue number2
DOIs
Publication statusPublished - Apr 2012

Keywords

  • crossed random effects
  • data augmentation
  • ERGM
  • mixed membership models
  • multilevel networks
  • peer effects

Fingerprint

Dive into the research topics of 'Bayesian Analysis of Multilevel Probit Models for Data With Friendship Dependencies'. Together they form a unique fingerprint.

Cite this