Relational event models (REMs) are statistical models for the analysis of sequences of relational observations. Estimating these models is straightforward (albeit potentially computationally demanding) when the exact sequence of events is known, such as in the case of phone call records. However, sequences of relational event data may be incomplete or difficult to gather, e.g., when online platforms only provide aggregated data on user behaviors or when networks derive from cognitive perceptions of relationships, like friendship perceptions. Those perceptions are typically collected through survey panels that do not gather instantaneous network changes.

We present an application of the Expectation-Maximization algorithm to estimate the parameters of a REM variant, the Dynamic Network Actor Model (DyNAM), when the sequences are partially observed. We consider the evolution of an incomplete sequence as the result of two conditionally independent DyNAM processes, one governing the event creation and the other the event deletion. The generalized Expectation-Maximization is considered for imputing missing data and generating a sequence of proposals that do not necessarily maximize the objective function. The Expectation step is computationally intractable and approximated by random samples of plausible event sequences aggregated following Multiple Importance Sampling schemes.

Additionally, we show that this implementation is not limited to the analysis of incomplete sequences of relational events but can also be applied to analyze the co-evolution of network panel data and relational event data. We illustrate the method by analyzing data on friendship networks collected in a panel design and online interactions.

The Stochastic Actor-Oriented Model (SAOM) is a model of network evolution, where estimated coefficients represent the timing of decisions and the propensity of actors to form or maintain relationships over time. One notorious challenge about these models is that the coefficients are on a latent rather than probability scale. As a consequence, they suffer from the same scalability issue as many generalized linear models, where the value of coefficients is affected by non-collapsibility due to unobserved heterogeneity. This makes them hard to interpret and to compare across models or datasets.

To address these interpretational challenges, we propose employing effect size measures based on empirical and counterfactual differences in predicted probabilities, thus calculating marginal effects, which avoid scaling issues. This is particularly challenging for network data however, as the interdependency of network statistics necessitates considering higher-order derivatives and discrete differences. We show that averaging over these marginal effects can differ in magnitude and sign compared to the original (latent scale) coefficients.

Since SAOM models unobserved decision sequences between network observations, researchers might either want to interpret effects at observation moments or take the simulated chains into account. We develop a flexible simulation-based method to allow combining either approach with different quantities of interest to be averaged over. An analysis of empirical data demonstrates differences between coefficient-based and marginal effect size interpretations for predicted tie and change probabilities.

These new tools will not only support mediation and meta-analysis for longitudinal network models but facilitate answering new research questions with SAOM.

Brokerage is a central concept in organizational network research, describing the influence actors exert by connecting or separating otherwise disconnected individuals. Traditionally, brokerage has been measured through structural positions in networks, often using Burt’s (1992) constraint measure. More recently, attention has shifted to brokerage orientations, reflecting semi-permanent behavioral tendencies such as tertius iungens (connecting others) and tertius gaudens (maintaining separation). Despite theoretical advancements, methodological approaches in network analysis have lagged. Most management studies still rely on non-network regression models that overlook tie interdependence, potentially leading to flawed interpretations. At the same time, while network models such as stochastic actor-oriented models (SAOMs) offer powerful tools for modeling network dynamics, they provide limited options for capturing complex brokerage behaviors beyond basic betweenness measures.

Our study addresses these gaps by conceptualizing brokerage as a dynamic process based on personal preferences and introducing novel SAOM effects that capture different brokerage orientations and the impact of brokerage positions on actor behavior. Using a unique dataset, we illustrate the application of these effects and offer practical guidance for interpretation. Our contributions include (1) integrating brokerage orientations into a widely-used network modeling framework, enabling researchers to study brokerage as behavioral trajectories without relying on trait-based data, as is necessary when modeling interorganizational relations; (2) providing a method to estimate the impact of brokerage positions while accounting for network interdependencies; and (3) fostering a coevolutionary perspective on brokerage, aligning network dynamics with brokerage outcomes and advancing research on organizational networks.

Social network researchers often assume that mutual nominations between actors are more likely to occur than one-sided nominations. However, this assumption does not always hold, particularly in large social networks. Actors who nominate many others often do not achieve complete reciprocation from all their nominees. In this presentation, we introduce a new parameter termed “overchoosing,” which integrates reciprocity and outdegree popularity effects as their product. This parameter captures an endogenous tie-formation process not adequately represented by existing network configurations. We hypothesize a negative estimate for the overchoosing parameter, indicating that the greater the number of outgoing nominations by an actor, the lower the likelihood that each nomination will be reciprocated. To test this hypothesis, we implemented the overchoosing parameter within stochastic actor-oriented models (SAOM). Analyses of secondary school friendship networks (N > 1000, two waves) demonstrated that the model including the overchoosing parameter provided a better fit to observed data compared to a model without it, as evaluated by the triad census goodness-of-fit indices—particularly for the triadic out-star without reciprocation (021D). We also present and discuss findings from additional analyses of online social networks and bitcoin trust networks (Ns > 500, number of waves > 100) by Bayesian SAOM optimized for parallel computation on supercomputers (Chan et al., 2022). These insights help refine our understanding of how reciprocal relationships form or dissolve in larger social networks.