Explanatory Item Response Models for Dyadic Data in Multiple Groups

• In Event: 4351 - Measurement Issues in Social Statistical Models

Abstract

Like other quantitative social scientists, network researchers can benefit from pooling information on multiple variables to infer underlying (latent) attributes or social processes. Appropriate data for this task is increasingly available. The inherent dependencies in relational data, however, pose unique challenges. This is especially true for those involved in the ascendant tasks of cross-network comparisons or multilevel network analysis. The author draws on item response theory and multilevel (mixed effects) modeling to propose a flexible framework that accounts for these dependencies and allows the analyst to model variation of latent dyadic traits across relations, actors, and groups precisely and parsimoniously. Three example applications demonstrate the approach’s utility for important research areas: tie strength in adolescent friendships, group differences in the meaning of network name interpreters, and the analysis of multiplex relations. Possible extensions for ordered manifest variables and simultaneous modeling with tie formation are discussed.

Author

• James Murphy, University of Chicago