Author: Robin Denz
simDAG is an R-Package which can be used to generate data from a known directed acyclic graph (DAG) with associated information on distributions and causal coefficients. The root nodes are sampled first and each subsequent child node is generated according to a regression model (linear, logistic, multinomial, cox, …) or other function. The result is a dataset that has the same causal structure as the specified DAG and by expectation the same distributions and coefficients as initially specified. It also implements a comprehensive framework for conducting discrete-time simulations in a similar fashion.
Currently this package is not available on CRAN, but can be installed easily using the
If you encounter any bugs or have any specific feature requests, please file an Issue.
Suppose we want to generate data with the following causal structure:
age is normally distributed with a mean of 50 and a standard deviation of 4 and
sex is bernoulli distributed with
p = 0.5 (equal number of men and women). Both of these “root nodes” (meaning they have no parents - no arrows pointing into them) have a direct causal effect on the
bmi. The causal coefficients are 1.1 and 0.4 respectively, with an intercept of 12 and a sigma standard deviation of 2.
death is modeled as a bernoulli variable, which is caused by both
bmi with causal coefficients of 0.1 and 0.3 respectively. As intercept we use -15.
The following code can be used to generate 10000 samples from these specifications:
dag <- empty_dag() + node("age", type="rnorm", mean=50, sd=4) + node("sex", type="rbernoulli", p=0.5) + node("bmi", type="gaussian", parents=c("age", "sex"), betas=c(1.1, 0.4), intercept=12, error=2) + node("death", type="binomial", parents=c("age", "bmi"), betas=c(0.1, 0.3), intercept=-15) sim_dat <- sim_from_dag(dag, n_sim=10000)
By fitting appropriate regression models, we can check if the data really does approximately conform to our specifications. First, lets look at the
This seems about right. Now we look at
The estimated coefficients are also very close to the ones we specified. More examples can be found in the documentation and the vignette.
© 2023 Robin Denz
The contents of this repository are distributed under the GNU General Public License. You can find the full text of this License in this github repository. Alternatively, see http://www.gnu.org/licenses/.