Optimal timing of cross-sectional network samples in longitudinal network studies

TL;DR

This paper develops a statistical framework to determine the optimal timing of cross-sectional network snapshots in longitudinal social network studies, aiming to improve the precision of parameter estimates.

Contribution

It introduces a novel analytical approach to guide the timing of network observations based on network process durations and dynamics.

Findings

Optimal timing depends on short-duration network processes.

Derived simple approximations for timing based on network event rates.

Application to a dynamic network example demonstrates practical utility.

Abstract

When choosing the timing of cross-sectional network snapshots in longitudinal social network studies, the effect on the precision of parameter estimates generally plays a minor role. Often the timing is opportunistic or determined by a variety of considerations such as organizational constraints, funding, and availability of study participants. Theory to guide the timing of network snapshots is also missing. We use a statistical framework to relate the timing to the precision of the parameter estimates, specifically, the sum of the relative widths of their confidence intervals. We illustrate this computationally using the STERGM suite of the statnet package to estimate the parameters of the network dynamics. Analytically, we derive simple approximations for the optimal timing when the parameters correspond to the rates for different network events. We find that the optimal time depends the most on the network processes with short expected durations and few expected events such as the dissolution of ties. We apply our approximations to a simple example of a dynamic network with formation and dissolution of iid ties.