Bayesian Network Analysis

BETA

A network over the modeled factors. Lines show conditional dependencies in the fitted network; each arrow points to the factor (or the outcome) that depends on the one at its tail — so up vs. down is just the layout, what matters is which way the head points. The % on each line (and its thickness) is that link's confidence — how consistently it holds up across resamples. Factors are ranked by the model-implied difference in the outcome when a factor is set to different displayed values and the rest of the fitted network is allowed to respond. These projections follow the fitted network and its assumptions; they are not identified causal effects. Treat thin (low-confidence) links as unsettled. These are the factors in the active Model Builder specification — change the set or refit there.

Reading it against the logit: the two models provide different views of conditional structure. The logit's incremental measure asks what a predictor adds after the others are included; the network distinguishes direct and indirect paths in the fitted graph. Agreement can strengthen confidence that a pattern is not unique to one specification, while disagreement is diagnostic. Neither comparison establishes causal or temporal direction.

Outcome

Subgroup

Level

Choose an outcome and any filters above, then click Run analysis to build the network.

Cross-check summary

Does each variable add incremental information and also connect to the outcome in the fitted network?

—

Do the two model views tell the same story about this factor? A factor can look important in the logit because it adds information after the other predictors are included, yet have no path to the outcome in the fitted Bayesian network. This cross-check compares those two properties. A factor supported by both is labeled Network-linked; one with incremental association but no modeled path is labeled Logit-only. These are properties of the fitted models, not proof that changing a factor would causally change the outcome.

What each verdict means

Network-linked Supported by both views. It adds incremental information in the logit and has a modeled path to the outcome in the fitted network.

Logit-only Logit signal without a network path. It adds incremental information in the logit, but the fitted network does not place a route from it to the outcome.

Minor Limited support in both views. It adds little incremental information and has little model-implied outcome change.

Chip colour shows the factor's network path: direct indirect no-path

What-if Simulator

Project the outcome if a factor were set for the whole sample.

1Set one or more factors to a level — the rest stay as observed.
2Click Run simulation.
3Read the projected outcome and its change vs the observed base.

Only factors with a modeled path to the outcome are included here. Factors with no path in the fitted network are left out because this network would not propagate a change from them to the outcome. Each scenario uses a do() setting within the fitted network, so downstream nodes respond. The result is a model-based projection under that structure, not an identified intervention effect.

Projected outcome

—

interval diagnostics: waiting for a run

How Are the Selected Factors Conditionally Related?