A1735
Title: Latent population size estimation in directed networks via microfounded random graph models
Authors: Yuki Iwanaga - Kobe University (Japan) [presenting]
Teruyoshi Kobayashi - Kobe University (Japan)
Abstract: Estimating the true size of a population from network data is challenging when nodes with no recorded connections are systematically absent from observation. While link formation probabilities admit various specifications, the generalized random graph model offers a particularly tractable and economically interpretable foundation. A microfounded model is proposed for directed networks, where each link probability is derived from a discrete choice framework in which nodes differ in attractiveness and link formation cost. Under Gaussian heterogeneity in node-level parameters, the probability that a node goes unobserved is characterized as a function of the distributional hyperparameters and the latent population size. This isolation probability is used to identify the latent population size via moment conditions. A moment-matching algorithm is developed to jointly estimate the population size and hyperparameters by minimizing discrepancies between observed and theoretically implied degree sequence moments, without requiring node-level parameter inference. Simulation studies confirm the accuracy of the proposed estimator across a range of network configurations, and applications to real-world directed networks demonstrate its practical utility.
