Seminario Ciencias Sociales Computacionales | Graph-Based Ideal Points

Lugar: Salas Seminario Doctorado Ciencias Sociales El Instituto de Sociología UC les invita a participar de la próxima sesión del Seminario de Ciencias Sociales Computacionales el que se realizará el próximo miércoles 18 de noviembre entre las 15:30 y las 17:00 hrs La sesión estará a cargo del investigador Naim Bro (Escuela de Gobierno, UAI) quien presentará Graph-Based Ideal Points (con Andrés Abeliuk y Jorge Fábrega) A continuación un resumen de la presentación: Ideal-point models mostly place political actors from traces they produce themselves: roll-call votes, or a corpus of their own words. Actors who neither vote nor write at length—ministers, party leaders, challengers—fall outside the frame. We introduce graph-based ideal points, inferred from signed relations among actors as reported by third parties. The observer supplies evidence that one actor supports or attacks another; position is recovered only from the aggregate signed actor-to-actor matrix, which any signed-graph method can scale. We demonstrate with a signed-Laplacian spectral axis and a neural graph-attention embedding. Validating the extension directly, positions estimated for Chilean deputies before they cast a single vote correlate with their later W-NOMINATE scores at r = .41–.54, rising to .53–.79 in the most recent transition. Among sitting legislators both estimators recover W-NOMINATE at r ≈ .84(Chamber) and .72–.74 (Senate).