Home

ACTR_DDM is a project to analyzing ACT-R simulations with HDDM, or, viceversa, to interpret HDDM simulations in terms of ACT-R modules, processes, and parameters.

Why?

Because DDM is simple and elegant, but ACT-R provides a direct mapping to internal processes. Furthermore, ACT-R parameters can be interpreted in terms of quantities that are potentially task-independent. Suppose you estimate model parameters from two tasks; in DDM, you get values like v and a that are solely task-dependent. In ACT-R, you could use behavioral results to estimate parameters (such as W, S, f, and alpha) that are participant-dependent.

Which ACT-R model?

ACT-R is very flexible, and many models are possible. We have tried to capture various possible phases of a decision process within a single, general-purpose model.

How?

Here is the general strategy, as envisioned.

1. A single ACT-R model is instantiated, with a number of parameters set to predefined values. When running many simulations, these parameters will form a hyperparameter space that will be examined extensively. Within each run, there is only one condition, and only one type of trials.

2. The model is run a predefined number of times, so that parameters can be estimated properly. Technically, each run corresponds to a different "subject", even if each subject is, in effects, a "clone" of the same individual.

3. HDDM is then run on the simulations. The expected values of v, a, and T are recorded.

The net result of this process is that, for every hyperparameter in ACT-R space, we have a corresponding hyperparameter (a 3D point in v, a, and T axes) in DDM space.