Outline

Softwares:

Tikz
DAGitty
ggdag
dagR

Presentation Structure:

Overview and resources
Pros and cons
Three examples of DAGs (confounding, mediation, more complex)

We will conclude with a comparison chart, honorable mentions, and our general recommendations depending on your needs.

Tikz

This is a LaTeX package for creating DAGs.

\usepackage{tikz}
\tikzset{> = stealth} % nicer looking arrows

Overview:

Developers: Till Tantau, Christian Feuersänger
Depends: TeX

Resources:

Tikz

Pros:

Very flexible, able to easily adjust size, shape, and color of nodes and lines
Allows for curved lines, subscripts and math symbols
Lots of online help and resources
Very intuitive code

Cons:

Must specify node locations using coordinates or relative to other nodes, no automatic placement
No built-in exposure, outcome covariates framework (this may be good for more complicated DAGs)
Steeper learning curve if unfamiliar with latex
So many options for customization, can be overwhelming

Tikz

Code

\begin{center}
\begin{tikzpicture}
% nodes
%generally: create node, circle shape, 
%named C with (X = 0, Y = 2) coordinates, 
%labeled C
\node[circle, draw] (C) at (0, 2) {C}; 
\node[circle, draw]  (A) at (1, 0) {A};
\node[circle, draw]  (Y) at (3, 0) {Y};
% edges
\draw[->, thick]
(C) edge (A)
(C) edge (Y)
(A) edge (Y);

\end{tikzpicture}
\end{center}

DAG 1

Tikz

Code

\begin{center}
\begin{tikzpicture}
% nodes
\node[circle, draw] (C) at (0, 0)  {C};
\node[circle, draw]  (A) at (1.5, 0) {A};
\node[circle, draw]  (Y) at (4.5, 0) {Y};
\node[circle, draw] (M) at (3, 0) {M}; 
% edges
\draw[->, thick]
(C) edge (A)
(C) edge [bend right=40] (Y) % creating curve 
(C) edge [bend right=30] (M)
(A) edge [bend left=30] (Y)
(A) edge (M)
(M) edge (Y);

\end{tikzpicture}
\end{center}

DAG 2

Tikz

Code

\begin{center}
\begin{tikzpicture}
% nodes
% note we can use math mode to write labels!
\node[circle, draw] (C1) at (0, 2)  {$C_1$}; 
\node[circle, draw] (C2) at (0, 0)  {$C_2$}; 
\node[circle, draw] (C3) at (2, 2)  {$C_3$}; 
\node[circle, draw]  (A) at (0, 4) {$A$};
\node[circle, draw]  (Y) at (4, 4) {$Y$};
\node[circle, draw] (M) at (2, 4) {$M$}; 
% edges
\draw[->, thick]
(C1) edge (A)
(C1) edge (M)
(C1) edge (Y)
(C1) edge (C2)
(C2) edge [bend left=40] (A)  
(C2) edge (M) 
(C2) edge [bend right=30] (Y)
(C3) edge (M) 
(C3) edge (Y) 
(A) edge [bend left=30] (Y)
(A) edge (M)
(M) edge (Y);

\end{tikzpicture}
\end{center}

DAG 3

DAGitty

This is a browser-based environment for creating, editing and analyzing DAGs. It also an available R package on CRAN.

library(dagitty)

Overview:

Author: Johannes Textor, Benito van der Zander, Ankur Ankan
Published: Jan 20, 2021
Depends: R(>=3.0.0)

Resources:

DAGitty

Pros:

Built-in exposure, outcome covariate framework
Can auto-generate or specify location
Easy learning curve when using the browser application
Can incorporate curved lines (not well)
Easy to get adjustment sets

Cons:

Can not easily handle Subscripts and color or math notation
Can look cluttered if you auto-generate location
Doesn’t have circles around text in R
To make DAG look “pretty” google resources say to use ggdag.

DAGitty

Code

dag1 <- dagitty( "dag {
                  A [exposure]
                  Y [outcome]
                  A -> Y
                  C -> A 
                  C -> Y}" )
coordinates(dag1 ) <-  list(
  x = c(C = 0, A = 0, Y = 1),
  y = c(C = 0, A = 1, Y = 1)
)

plot(dag1)

DAG 1

DAGitty

Code

dag2 <- dagitty( 'dag {
  bb="-1.956,-1.956,1.956,1.956"
  A [exposure,pos="-1.000,0.000"]
  C [pos="-1.500,0.000"]
  M [pos="-0.500,0.000"]
  Y [outcome,pos="0.000,0.000"]
  A -> M
  A -> Y [pos="-0.500,-0.250"]
  C -> A
  C -> M [pos="-1.00,0.250"]
  C -> Y [pos="-1.00,0.500"]
  M -> Y}' )

plot(dag2)

DAG 2

DAGitty

Code

dag3 <- dagitty('dag {
  bb="-3,-3,3,3"
  A [exposure,pos="0.250,0.150"]
  C_3 [pos="1.025,0.600"]
  C_2 [pos="0.500,1.000"]
  C_1 [pos="0.500,0.600"]
  M [pos="0.900,0.150"]
  Y [outcome,pos="1.500,0.150"]
  A -> M
  A -> Y [pos="1.000,0.090"]
  C_1 -> C_2
  C_3 -> M
  C_3 -> Y
  C_2 -> A [pos="0.150,0.500"]
  C_2 -> M
  C_2 -> Y [pos="1.500,0.800"]
  C_1 -> A
  C_1 -> M
  C_1 -> Y
  M -> Y
}')

plot(dag3)

DAG 3

ggdag

This is an R package on CRAN that is built on top of dagitty and used ggplot2 and ggplot to

library(ggdag)

Overview:

Author: Malcolm Barrett
Published: Oct 29, 2022
Depends: R(>=3.4.0)
Built on top of dagitty
uses ggplot2 and ggplot

Resources:

CRAN Documentation

Vignettes/Tutorials:

ggdag

Pros:

Built-in exposure, outcome covariate framework
Can auto generate or specify location
Can incorporate curved lines
Can handle subscripts and color
Lots of examples online
Easy to get adjustment sets
tidy_dagitty function

Cons:

Not clear when to use ggplot vs ggdag functions
Location specification takes trial and error
Not clear if math symbols can be incorporated
Can looked cluttered if you auto generate location
Steep learning curve

ggdag

Code

dag1 <- 
  dagify(
    # relationship for each node
    A ~ C,
    Y ~ A + C,
    exposure = "A",
    outcome = "Y",
    # location for each node
    coords = list(
      x = c(Y = 4, A = 2, C = 2),
      y = c(Y = 2, A = 2, C = 3)
    )
  )

# could use ggdag or ggplot
ggdag(dag1, layout = "circle", 
      node_size = 20) +
  theme_void() +
  labs(title = "DAG1")

DAG 1

ggdag

Code

dag2 <- 
  dagify(
    # relationship for each node
    A ~ C,
    M ~ A + C,
    Y ~ M + A + C,
    exposure = "A",
    outcome = "Y" ,
    coords = list(
      x = c(Y = 5, M = 4, A = 3, C = 2),
      y = c(Y = 2, M = 2, A = 2, C = 2)
    )
  )

# plotting with curves 
dag2 %>% 
  ggplot(aes(x = x, y = y, 
             xend = xend, yend = yend)) +
  geom_dag_point() +
  geom_dag_edges_arc(
    curvature = c(0, 0.1, 0,-0.25,-0.25,0)
  )+
  geom_dag_text() +
  theme_dag() + 
  labs(title = "DAG2")

DAG 2

ggdag

Code

dag3 <- 
  dagify(
    # relationship for each node
    A ~ C_1 + C_2,
    C_2 ~ C_1,
    M ~ A + C_1 + C_2 + C_3,
    Y ~ M + A + C_1 + C_2 + C_3,
    exposure = "A",
    outcome = "Y"
  )

ggdag(dag3, layout = "circle") +
  theme_void() +
  labs(title = "DAG3")

DAG 3

ggdag

Code

dag3 <- dagify(
   # relationship for each node
    A ~ C1 + C2,
    C2 ~ C1,
    M ~ A + C1 + C2 + C3,
    Y ~ M + A + C1 + C2 + C3,
    # specify exposure and outcome 
    exposure = "A",
    outcome = "Y",
    # Location of each node
    coords = list(
      x = c(Y = 3, M = 2, A = 1, 
            C1 = 1, C2 = 1, C3 = 2),
      y = c(Y = 2, M = 2, A = 2,
            C1 = 1, C2 = 0, C3 = 1)
    )
  ) 

dag3 %>% 
  ggplot(aes(x = x, y = y, 
             xend = xend, 
             yend = yend)) +
  geom_dag_point() + # plot nodes
  #geom_dag_edges() + # straight edges 
  geom_dag_edges_arc( 
    curvature = c(0, 0.25, 0, 0, 
                  0, 0, 0.25, 0, 
                  -0.25, 0, 0, 0)
    # straight + curved edges
  ) +
  theme_void() +
  geom_dag_text(
    parse = TRUE, 
    label = c( # include subscripts
      "A", expression(C[1]), 
       expression(C[2]), 
       expression(C[3]), "M", "Y")
  ) +
  labs(title = "DAG3")

DAG 3

dagR

This is an R package for drawing, manipulating and evaluating DAGs, and simulating corresponding data.

library(dagR)

Overview:

Author: Lutz P Breitling
Published: Oct, 09, 2022
Depends: R(>=3.0.0)
Built on top of dagitty

Resources:

dagR

Pros:

Built-in exposure, outcome covariate framework
Has a feature to simulate data from specified DAG
Can provide basic interpretation of DAG (e.g. identify ancestors, backdoor paths)
Can generate autoplacement of nodes

Cons:

Few resources/examples online
Not very intuitive code
No curved arrows
No subscripts
Not nice visuals as other software
Annoying name to google (DAGR is a GPS software)

dagR

Code

dag.dat <-
    dag.init(outcome = NULL, exposure = NULL,
             covs = c(1), #1 per covariate
             arcs = # 0:exposure, -1: outcome # write in pairs
               c(1,0, # C to A
                 1,-1, # C to Y
                 0,-1), # A to Y
             assocs = c(0), # directed arcs, one 0 per cov
             xgap = 0.04, 
             ygap = 0.05, 
             len = 0.1,
             x.name = "Exposure",
             cov.names = c("Covariate"),
             y.name = "Outcome",
             symbols = c("A", "C", "Y"),
             noxy = T
             )

dag.dat$x <- c(0.25, 0, 1) # A, C, Y x-axis positions
dag.dat$y <- c(0, 1, 0)


plot1 <- dag.draw(dag.dat, noxy = T, legend = T)

DAG 1

dagR

Code

dag.dat <-
    dag.init(outcome = NULL, exposure = NULL,
             covs = c(1, 1), #1 per covariate
             arcs = # 0:exposure, -1: outcome # write in pairs
               c(1,0, # C to A
                 1,2, # C to M
                 1,-1, # C to Y
                 0, 2, # A to M
                 0,-1, # A to Y
                 2,-1 # M to Y
             ),
             assocs = c(0, 0), # directed arcs, one 0 per cov
             xgap = 0.04, 
             ygap = 0.05, 
             len = 0.1,
             symbols = c("A", "C", "M", "Y"),
             noxy = T
             )

dag.dat$x <- c(0.33, 0, 0.66, 1) # A, C, M, Y x-axis positions

plot2 <- dag.draw(dag.dat, noxy = T, legend = F)

DAG 2

dagR

Code

dag.dat <-
    dag.init(outcome = NULL, exposure = NULL,
             covs = c(1, 1, 1, 1), #1 per covariate
             arcs = # 0:exposure, -1: outcome # write in pairs
               c(1,0, # C1 to A
                 1,4, # C1 to M
                 1,-1, # C1 to Y
                 1,2, # C1 to C2
                 2,0, # C2 to A
                 2,4, # C2 to M
                 2,-1, # C2 to Y
                 3,4, # C3 to M
                 3,-1, # C3 to Y
                 4,-1, # M to Y
                 0, 4, # A to M
                 0,-1 # A to Y
                 
             ),
             assocs = c(0, 0, 0, 0), # directed arcs, one 0 per cov
             xgap = 0.04, 
             ygap = 0.05, 
             len = 0.05,
             symbols = c("A", "W", "X", "Z","M", "Y"),
             noxy = T
             )

plot3 <- dag.draw(dag.dat, noxy = T, legend = F)

DAG 2

Honorable Mentions

Quiver: Web resource to quickly draw a DAG by clicking and dragging, provides LaTeX code

CausalFusion: Web resource that can generate DAGs and has analysis capabilities, requires an approved account

Tetrad: Free Downloadable tool (with a 30 year history) that creates, simulates data from, estimates, tests and predicts with causal and statistical models.

shinyDAG: Web app that use R and LaTeX to create publication-quality images of DAGs.

igraph: R Package designed to handle large graphs

Conclusion

For paper/presentation- tikZ

For quick drawing- DAGgity or Quiver

For html formal drawing: ggdag

Maybe to simulate data: dagR

Software to Draw DAGs

Outline

Softwares:

Presentation Structure:

Tikz

Overview:

Resources:

Tikz

Pros:

Cons:

Tikz

Tikz

Tikz

DAGitty

Overview:

Resources:

DAGitty

Pros:

Cons:

DAGitty

DAGitty

DAGitty

ggdag

Overview:

Resources:

Vignettes/Tutorials:

ggdag

Pros:

Cons:

ggdag

ggdag

ggdag

ggdag

dagR

Overview:

Resources:

dagR

Pros:

Cons:

dagR

dagR

dagR

Comparison

Honorable Mentions

Conclusion