Bayesian Event Tree Analysis — Probabilistic Risk Modeling with Prior Updating
Bayesian Event Tree Analysis · Also known as: Bayesian ETA, B-ETA, Probabilistic Event Tree Analysis, Bayesian Inductive Risk Model
Bayesian Event Tree Analysis (B-ETA) is a quantitative risk assessment method that extends classical event tree analysis by incorporating Bayesian inference to assign and update branch probabilities. Starting from an initiating event, it maps sequences of successes and failures through safety barriers, using prior distributions and observed evidence to produce posterior outcome probabilities. Widely used in nuclear safety, process industries, and system reliability engineering.
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 Event Tree Analysis when you need to quantify the risk of specific accident sequences following a known initiating event, especially when branch probabilities are uncertain or must be updated as new operational data accrue. It is particularly well suited to industries with rigorous safety cases — nuclear power, oil and gas, chemical processing, aviation — where regulators expect probabilistic risk assessment with explicit uncertainty quantification. It is appropriate when historical failure data exist but are sparse, making a Bayesian prior-to-posterior update more honest than a fixed point estimate. Do not use it when the causal structure of the accident sequence is unknown (fault tree analysis or a causal model is needed first), when data are entirely absent with no defensible prior (the result will be dominated by assumptions), or when a qualitative hazard checklist suffices for the decision at hand.
Strengths & limitations
- Explicitly represents and propagates uncertainty in branch probabilities through full posterior distributions rather than single-point estimates.
- Supports sequential updating — as new incident or test data become available, branch probabilities improve without restarting the analysis from scratch.
- Integrates naturally with expert elicitation when historical failure data are sparse or unavailable.
- Provides a structured, auditable accident sequence model that satisfies safety-case and regulatory documentation requirements.
- Monte Carlo propagation yields uncertainty bounds on outcome frequencies, enabling informed risk-tolerability decisions.
- Building a credible tree and eliciting defensible priors is time-intensive and requires domain expertise in both the system and Bayesian statistics.
- The method models linear, binary (success/failure) sequences; common-cause failures, dependent barriers, and feedback loops are difficult to represent without converting to a Bayesian network.
- Prior choices can dominate results when data are very sparse, and different experts may elicit substantially different priors, leading to wide credible intervals.
- Computational cost rises quickly with the number of branches when Monte Carlo sampling is used for full uncertainty propagation.
Frequently asked
How does Bayesian ETA differ from classical ETA?
Classical ETA assigns fixed point probabilities to each branch, typically from historical databases. Bayesian ETA treats each branch probability as a probability distribution, elicited from prior knowledge and updated with observed data via Bayes' theorem. The result is a full distribution of outcome frequencies rather than a single number, allowing explicit uncertainty quantification and sequential updating as evidence accumulates.
When should I use a Bayesian network instead of Bayesian ETA?
Bayesian ETA is best when the accident sequence is linear and the event tree structure (initiating event followed by sequential barrier successes or failures) is well established. Bayesian networks are more appropriate when barriers are interdependent, common-cause failures matter, or feedback loops exist, because a network can represent arbitrary conditional dependence structures that a tree cannot.
How do I choose a prior distribution for a branch probability?
A conjugate Beta(alpha, beta) prior is standard for a probability. Set alpha and beta to reflect your prior belief: if you believe the failure probability is around 0.01 with moderate confidence, you might use Beta(1, 99). If you have historical records of n trials with k failures, a natural non-informative starting point is Beta(1,1) updated to Beta(k+1, n-k+1). Expert elicitation methods such as SHELF or the percentile method can translate expert judgment into Beta parameters.
Can Bayesian ETA handle common-cause failures?
Only in a limited way. You can introduce a common-cause branch that collapses multiple barriers simultaneously, but modeling rich dependence structures requires moving to a Bayesian network or a bow-tie model. If common-cause failures are a significant contributor to risk, a full Bayesian network representation is recommended.
What software supports Bayesian event tree analysis?
Specialized probabilistic risk assessment tools such as SAPHIRE, RiskSpectrum, and OpenPSA support classical event trees and some Bayesian updating. General Bayesian network software (Netica, Hugin, AgenaRisk) can implement B-ETA as a network. Monte Carlo platforms such as OpenTURNS or custom Python scripts using PyMC or Stan are used for full uncertainty propagation in research settings.
Sources
How to cite this page
ScholarGate. (2026, June 3). Bayesian Event Tree Analysis. ScholarGate. https://scholargate.app/en/experimental-design/bayesian-event-tree-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 failure mode and effects analysisExperimental design↔ compare
- Bayesian Fault Tree AnalysisExperimental design↔ compare
- Bayesian Reliability AnalysisBayesian↔ compare
- Event Tree AnalysisReliability↔ compare
- Failure Mode and Effects AnalysisExperimental design↔ compare
- Fault Tree AnalysisReliability↔ compare