bayesactR provides utilities that allow R users to run simulations using the C package BayesACT, developed by Dr. Jesse Hoey and colleagues, entirely from within R.
Visit the Getting Started page for installation information. For examples of how to set up and run simulations, visit this help page.
Bayesian Affect Control Theory
BayesACT, developed by Drs. Jesse Hoey, Tobias Schröder, Kimberly Rogers, and colleagues, is a theoretical extension of affect control theory. A short summary of affect control theory (ACT) and BayesACT is provided below. See affectcontroltheory.org, Schröder, Hoey, and Rogers (2016), and bayesact.ca for more information.
ACT is a social psychological theory of interaction that models how the affective meanings of people and behaviors relate to what we expect to happen in a situation. Within ACT, words describing people and behaviors are modeled as having locations in a three-dimensional space of affective meaning. These dimensions are evaluation (good/bad), potency (powerful/weak), and activity (active/quiet), and range between approximately -4 and 4. These affective meanings have been measured empirically in a number of data collection efforts across several different countries since the 1960s. For example, in one recent data collection, the word teacher was found to be seen on average as extremely good (E = 2.62), quite powerful (P = 1.82) and slightly active (A = 1.3).
These values can be used to run simulations of interactions and the results of these simulations can tell us what kinds of behaviors we expect particular kinds of people to engage in, how we expect people who do certain things to be labeled, how strange particular social interactions seem to us, and more. These results can and have been verified through comparisons to empirical data.
In the mathematics of the core theory, meanings are treated as points in the three-dimensional meaning space. BayesACT, by contrast, treats them as Gaussian distributions, with a mean and a variance. Simulations in BayesACT are probabilistic. On each run, values are sampled from the distributions for the relevant terms, and results are determined based on those sampled values. The output of a BayesACT simulation is a compilation or summary of the results across a number of individual runs.
Why bayesactR?
Hoey and colleagues have developed and released an implementation of BayesACT that is written in C and designed to be interfaced with via the command line. bayesactR is an R wrapper for this tool.
The goals of bayesactR are (1) to make BayesACT more accessible to social scientists who prefer R-based rather than command-line-based workflows, (2) to make setting up and running multiple simulations at one time simpler, and (3) to facilitate creating analytic workflows that are easily reproducible.
This is a work in progress!
The current version of bayesactR is designed to work with BayesACT C 2.3.8, last modified on June 19, 2021.
This package is currently in an early-stage beta state. Key functionality has been implemented and has worked in my local tests, but testing in other contexts has so far been limited. In particular, all development and testing of this package has so far been done on MacOS (11.2.3 - 12.3.1). Development is ongoing, and I can’t promise that there won’t be breaking changes in future versions. I ask that you bear with me as I work towards the goal of developing a tool that is as flexible, useful, and user-friendly as possible!
Please get in touch with me (ahc26atduke.edu) if you encounter any bugs or confusions or have thoughts about how this might be made a more useful tool. All feedback is helpful and appreciated!
You may also be interested in…
This package was developed in conjunction with two other open-source R packages that may also be of interest to the ACT research community. Together, the goal of these packages is to make ACT research more accessible, and to make it possible to use R to do analysis in a self-contained and completely reproducible and transparent way.
inteRact: Like bayesactR, inteRact, developed and maintained by Em Maloney, is a package that allows users to run affect control theory simulations in R. It was built to be an open source, R-based version of its namesake Java program, INTERACT, which has been used by researchers to run affect control theory simulations since the 1990s. The conceptual difference between inteRact and bayesactR is that inteRact runs simulations using the mathematics of the core theory, rather than those of the Bayesian extension. In a nutshell, this means that it treats EPA values as points, rather than distributions, and simulations are deterministic rather than probabilistic. These simulations are much less computationally intensive than BayesACT simulations, and are easier to set up and run. However, they have fewer adjustable parameters and are less useful for some kinds of research questions.
actdata: actdata, which I develop and maintain, is an R package that serves as a data repository. The ACT research community has a long and commendable history of making their tools and data publicly available. This package provides that data in a standardized format alongside functions that help users search, subset, and export it in a format that works for their analysis program of choice. It makes it unnecessary for researchers to store their own local copies of publicly available data, and greatly simplifies the process of comparing values across cultures or time periods. bayesactR and inteRact both use actdata to supply data for simulations.