dag {cfid}

Directed Acyclic Graph

Description

Define a directed acyclic graph (DAG) describing the causal model.

Usage

dag(x, u = character(0L))

## S3 method for class 'dag'
print(x, ...)

Arguments

x

A dag object

u

A character vector of variable names which should be considered unobserved (besides those defined by bidirected edges). These variables are subsequently removed via latent projection. Variable names not found in the graph are ignored.

...

Not used

Details

The syntax for x follows the dagitty package closely for compatibility. However, not all features of dagitty graphs are supported. The resulting adjacency matrix of the definition is checked for cycles.

Directed edges are defined as X -> Y meaning that there is an edge from X to Y in the graph. Edges can be combined in sequence to create paths for concise descriptions, for example X -> Y <- Z -> W.

Unobserved latent confounders are defined using bidirected edges as ⁠X <-> Y⁠ which means that there is an additional variable U[X,Y] in the graph, and the edges X <- U[X,Y] -> Y, respectively.

Different statements in x can be distinguished from one another using either semicolons, line breaks, or spaces.

Subgraphs can be defined by enclosing the definition within curly braces. For example ⁠X -> {Y Z}⁠ defines an edge from X to both Y and Z. Individual statements within a subgraph can be separated by a comma or semicolon, but this is optional. Edges can also be defined within subgraphs, and subgraphs can be nested. For example, X -> {Z -> Y} is the same definition as ⁠X -> Z; X -> Y; Z -> Y⁠. Similarly ⁠X <-> {Z -> {A B}} -> Y⁠ is the same as ⁠X <-> {Z A B} -> Y; Z -> {A B}⁠.

Note that in the context of this package, vertex labels will always be converted into upper case, meaning that typing Z or z will always represent the same variable. This is done to enforce the notation of counterfactual variables, where capital letters denote variables and small letters denote their value assignments.

Value

An object of class dag, which is a square adjacency matrix with the following attributes:

• labels
  A character vector (or a list) of vertex labels.

• latent
  A logical vector indicating latent variables.

• order
  An integer vector giving a topological order for the vertices.

• text
  A character string giving representing the DAG. .

Examples

dag("X -> {Y Z} <- W <-> G")

# Subgraphs can appear on both sides of an edge
dag("{X Z} -> {Y W}")

# Semicolons can be used to distinguish individual statements
dag("X -> Z -> Y; X <-> Y")

# Commas can be used to distinguish variables within groups if there
# are no edges within the group
dag("{X, Y, Z} -> W")

# Edges within subgraphs is supported
dag("{X -> Z} -> {Y <-> W}")

# Nested subgraphs are supported
dag("{X -> {Z -> {Y <-> W}}}")

# Line breaks are also supported for statement separation
dag("
  Z -> W
  X -> Y
")