Bayesian Ego Network Analysis
Bayesian Ego Network Analysis (Probabilistic Inference on Personal Networks) · Also known as: Bayesian personal network analysis, Bayesian egocentric network analysis, probabilistic ego network modeling, Bayesian egonet
Bayesian ego network analysis applies probabilistic inference to ego-centered (personal) network data, combining a likelihood model for the ego's local network with prior distributions over network parameters. The result is a full posterior distribution that quantifies uncertainty about structural features such as alter composition, tie density, and network size — rather than producing point estimates alone.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Bayesian ego network analysis when ego-centered survey data carry substantial measurement uncertainty — small alter lists, partial tie reports, or sensitive topics where respondents may omit alters — and honest uncertainty quantification matters. It is well suited for comparing personal network structures across demographic or experimental groups while accounting for small group sizes, and for incorporating prior knowledge from validated personal network surveys. Avoid it when a simple descriptive summary of ego network composition suffices (classical mean and SD are adequate), when no reasonable prior can be specified, when software and Bayesian expertise are unavailable to reviewers or collaborators, or when the ego network data are very large and MCMC convergence becomes prohibitively slow.
Strengths & limitations
- Produces full posterior distributions over network parameters, making uncertainty explicit rather than hidden in standard errors.
- Hierarchical extensions borrow strength across egos, stabilising estimates for respondents with few or ambiguous alters.
- Incorporates prior knowledge from validated personal network norms (e.g., typical active network sizes) in a principled way.
- Enables direct probabilistic comparisons between groups without relying on asymptotic approximations that fail at small samples.
- Naturally accommodates missing or partially observed alter ties through marginalisation over the missing data.
- Compatible with rich structural models (ego-level ERGM, latent position models) within the Bayesian framework.
- Requires careful prior specification; poorly chosen priors can dominate results when alter lists are short.
- MCMC-based inference is computationally intensive and may require expert tuning for structural likelihood models.
- Results can be difficult to communicate to audiences unfamiliar with posterior distributions and credible intervals.
- Bayesian ego network software is less mature than classical ego network packages (e.g., egonet, EgoWeb); custom coding is often needed.
- Hierarchical models assume exchangeability across egos, which may not hold in heterogeneous populations.
Frequently asked
How is Bayesian ego network analysis different from classical ego network analysis?
Classical ego network analysis computes point estimates of density, composition, and structural holes for each ego. Bayesian analysis replaces those with posterior distributions, so every estimate comes with a credible interval. This matters most when alter lists are short (fewer than 10 alters), tie reports are uncertain, or group sizes are small.
Do I need a hierarchical model, or can I run Bayesian inference ego by ego?
Running inference separately per ego is valid and simpler. A hierarchical (multilevel) model adds value when you want population-level summaries or when individual egos have very few alters, because it borrows strength across respondents to stabilise estimates.
What prior should I use for ego network size or tie probability?
Empirical personal network research provides useful calibration: active network sizes typically range from 5 to 25 alters, and tie probabilities among alters are usually 0.3–0.6. A weakly informative beta prior for tie probability and a Poisson or negative-binomial prior for size are reasonable starting points.
Which software supports Bayesian ego network analysis?
There is no single dedicated package. Researchers typically combine ego network data-preparation tools (R packages egor, egonet) with general Bayesian inference platforms (Stan, JAGS, brms, PyMC). Bayesian ERGMs within ego networks can use Bergm or hand-coded Stan models.
Can this method handle missing alter-to-alter ties?
Yes. One of the clearest advantages of the Bayesian framework is that unobserved alter-to-alter ties can be treated as latent variables and marginalised over, rather than imputed with a fixed rule or dropped, which can bias classical density estimates.
Sources
- Krivitsky, P. N., & Kolaczyk, E. D. (2015). On the question of effective sample size in network modeling: An asymptotic inquiry. Statistical Science, 30(2), 184–198. DOI: 10.1214/14-STS502 ↗
- Ego network. Wikipedia. link ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Ego Network Analysis (Probabilistic Inference on Personal Networks). ScholarGate. https://scholargate.app/en/network-analysis/bayesian-ego-network-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bayesian Exponential Random Graph ModelNetwork analysis↔ compare
- Bayesian Social Network AnalysisNetwork analysis↔ compare
- Bayesian Stochastic Block ModelNetwork analysis↔ compare
- Ego Network AnalysisNetwork analysis↔ compare
- Social Network AnalysisNetwork analysis↔ compare