Help

Keyboard & mouse

Action Shortcut
Create a node Shift + Left Click
Select a node Left Click on node
Add / remove node from selection ⌘ / Ctrl + Left Click
Edit selected node (name & role) E
Toggle unobserved (latent) U
Toggle adjusted A
Toggle selection node S
Delete selected node X
Cycle edge type Shift + Left Click on another node (while one is selected: none → directed → bidirected → undirected → delete)
Search canvas (pan to node) Search box, bottom left
Pan canvas Right Click (hold) + Drag
Zoom Mouse Wheel
Move node Left Click (hold) on node + Drag

Node statuses

Status Purpose
Exposure / outcome / intermediate Mutually exclusive role. Causal identification needs ≥1 exposure and ≥1 outcome.
Unobserved (U) Latent / not in the data. Dashed outline; SEM ellipse. Never in adjustment sets.
Adjusted (A) Already conditioned on. Pale fill; must-include in identification.
Selection (S) Sampling node. Rectangle even in circle mode. Required for the causal odds ratio.

Canvas tools

Tool What it does
Layout Spring layout (DAGitty). Writes new positions — structural, same as dragging nodes.
SEM Labels become the node bodies (DAGitty style)
Ego Dim nodes outside the hovered node's k-hop neighborhood. Radius stepper appears when Ego is on.
Cycles Highlight directed cycles in rose
Paths Grey the graph; color directed paths between selected nodes (⌘/Ctrl+click to add nodes to the selection)
Highlight Hover an edge to paint it and its two nodes in pathway green, drawn on top of other arrows

Causal identification

Left panel, Causal mode. Roles drive this; Paths uses selection instead. With Paths off and X/Y set, lime = causal paths, magenta = biasing paths.

Identification Purpose
Adjustment (total effect) Minimal covariates that block back-doors from exposure to outcome.
Adjustment (direct effect) Same, plus block indirect causal paths. No selection nodes.
Instrumental variable Z, or Z given W, for a unique exposure and unique outcome.
Causal odds ratio Unique exposure, outcome, and selection node. Z such that X ⊥ S | Y ∪ Z and Z identifies the total effect.

View mode

Causal panel radios. Display only — the authored DAG is unchanged.

Mode What you see
normal The model you drew
moral graph Ancestor moral graph of the back-door graph (needs exposure and outcome)
correlation graph Undirected pairs that can be dependent given the empty set
equivalence class CPDAG: unprotected arrows become undirected
atomic direct effects Thicken arrows that have no mediator

Graph settings

Rename this graph and choose which snapshot is canonical. Maintainers only. Private graphs can be deleted only by their creator.

Members

Snapshot settings

Rename this snapshot. Maintainers can mark milestones. Creators can share by email and delete non-canonical snapshots.

Shared with

Create Snapshot

Save the current graph state as a new snapshot.

Go live

Shares your current working copy. Nothing is saved until someone creates a snapshot.

Members

Live session invite

End live session

Ending disconnects everyone. The last working copy stays on each canvas so anyone can still create a snapshot.

Edit Node

Role

Create Graph

Name your graph. Example DAGgity code is provided below — keep it, edit it, or replace it with your own. Clear the field to start from a minimal two-node template instead.

Export

DAGgity code for the current graph. Copy it into another tool or a new Pallium graph. Export as PNG captures the canvas view at publication resolution.

Loading...

Select a graph from the dropdown to begin.

For instructions on how to use the graph editor, click the Help button in the bottom right.

Select a snapshot from the dropdown to load.

For instructions on how to use the graph editor, click the Help button in the bottom right.

Network Stats

Nodes
—
Edges
—
Isolates
—
Diameter
—
Radius
—
Median Eccentricity
—
Density
—
Transitivity
—
Degree Assortativity
—
Open Triples
—
Mediated
—
Confounding
—
Colliders
—
Closed Triples
—
Power Law α
—
Scale Free
—

Node Analytics

—

In-Degree
—
Out-Degree
—
Total Degree
—
PageRank
—
Inverse α
—
Betweenness
—
Closeness
—
Eccentricity
—

Conditional Independencies

The model implies the following conditional independences:

    Minimal Sufficient Adjustment Sets

      View mode

      Bayesian Network

      View this graph as a Bayesian network — joint factorization, Markov blankets, and related structure.

      Coming soon.

      Analytics Mode

      Welcome to Pallium

      Create a graph to get started.

      For instructions on how to use the graph editor, click the Help button in the bottom right.

      1
      Scale
      X
      Y
      Size
      Nodes
      Edges