Marginal effects for the stochastic actor-oriented model
Abstract
Parameter estimates of stochastic actor-oriented models for network evolution are notoriously hard to interpret. Especially a comparison of parameter estimates is problematic, both between different models for the same data set and between the same model for different data sets. Effect size measures based on the calculation of marginal effects are proposed as an alternative to standardize the scale of interpretation but can and should not universally erase the effects of differences in scale and heterogeneity. Simulation experiments show how an omitted effect displaces coefficients and marginal effects differently, and an analysis of empirical friendship network data finds that reciprocity and triadic closure reinforce one another when ties are created but partially substitute for one another when ties are maintained, a contrast the coefficients do not reveal.
Related
- Viscosity-Informed Generative Actor-Critic for High-Dimensional Stochastic Optimal Control
- An analysis of Mirror-Descent Soft Actor-Critic
- Neural Actor--Critic Learning with Minimax Critics and Reflected Langevin Sampling
- Smartphone behavior and loneliness among older couples in China: Actor-partner interdependence modeling
- Schrödinger--Föllmer Actor--Critic: Diffusion Policy Improvement with Finite-Sample Analysis
BibTeX
@misc{gotthardt2026marginal,
title = {{Marginal effects for the stochastic actor-oriented model}},
author = {Daniel Gotthardt and Christian Steglich},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.41122},
eprint = {2610.41122},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.41122},
}
RIS
TY - UNPB
TI - Marginal effects for the stochastic actor-oriented model
AU - Daniel Gotthardt
AU - Christian Steglich
PY - 2026
DA - 2026/10/30
PB - arXiv
DO - 10.48550/arxiv.2610.41122
UR - https://arxiv.org/abs/2610.41122
AB - Parameter estimates of stochastic actor-oriented models for network evolution are notoriously hard to interpret. Especially a comparison of parameter estimates is problematic, both between different models for the same data set and between the same model for different data sets. Effect size measures based on the calculation of marginal effects are proposed as an alternative to standardize the scale of interpretation but can and should not universally erase the effects of differences in scale and heterogeneity. Simulation experiments show how an omitted effect displaces coefficients and marginal effects differently, and an analysis of empirical friendship network data finds that reciprocity and triadic closure reinforce one another when ties are created but partially substitute for one another when ties are maintained, a contrast the coefficients do not reveal.
ER -