Simulate Data from a DAG and Associated Node Information
simDAG.RdWhat is this package about?
This package aims to give a comprehensive framework to simulate static and longitudinal data given a directed acyclic graph and some information about each node. Our goal is to make this package as user-friendly and intuitive as possible, while allowing extreme flexibility and while keeping the underlying code as fast and RAM efficient as possible.
What features are included in this package?
This package includes two main simulation functions: the sim_from_dag function, which can be used to simulate data from a previously defined causal DAG and node information and the sim_discrete_time function, which implements a framework to conduct discrete-time simulations. The former is very easy to use, but cannot deal with time-varying variable easily. The latter is a little more difficult to use (usually requiring the user to write some functions himself), but allows the simulation of arbitrarily complex longitudinal data.
Through a collection of implemented node types, this package allows the user to generate data with a mix of binary, categorical, count and time-to-event data. The sim_discrete_time function additionally enables the user to generate time-to-event data with, if desired, a mix of competing events, recurrent events, time-varying variables that influence each other and any types of censoring.
The package also includes a few functions to transform resulting data into multiple formats, to augment existing DAGs, to plot DAGs and to plot a flow-chart of the data generation process.
What does a typical workflow using this package look like?
Users should start by defining a DAG object using the empty_dag and node functions. This DAG can then be passed to one of the two simulation functions included in this package. More information on how to do this can be found in the respective documentation pages and the three vignettes of this package.
When should I use sim_from_dag and when sim_discrete_time?
If you want to simulate data that is easily described using a standard DAG without time-varying variables, you should use the sim_from_dag function. If the DAG includes time-varying variables, but you only want to consider a few points in time and can easily describe the relations between those manually, you can still use the sim_from_dag function. If you want more complex data with time-varying variables, particularly with time-to-event outcomes, you should consider using the sim_discrete_time function.
What features are missing from this package?
The package currently only implements some possible child nodes. In the future we would like to implement more child node types, such as nodes with generalized mixed linear models or more complex survival time models.
Why should I use this package instead of the simCausal package?
The simCausal package was a big inspiration for this package. In contrast to it, however, it allows quite a bit more flexibility. A big difference is that this package includes a comprehensive framework for discrete-time simulations and the simCausal package does not.
Where can I get more information?
The documentation pages contain a lot of information, relevant examples and some literature references. Additional examples can be found in the vignettes of this package, which can be accessed using:
We are also working on a separate article on this package that is going to be published in a peer-reviewed journal.
I have a problem using the sim_discrete_time function
The sim_discrete_time function can become difficult to use depending on what kind of data the user wants to generate. For this reason we put in extra effort to make the documentation and examples as clear and helpful as possible. Please consult the relevant documentation pages and the vignettes before contacting the authors directly with programming related questions that are not clearly bugs in the code.
I want to suggest a new feature / I want to report a bug. Where can I do this?
Bug reports, suggestions and feature requests are highly welcome. Please file an issue on the official github page or contact the author directly using the supplied e-mail address.