Data Requirements and Model Selection
This guide explains:
- What type of data is required to use models implemented in JEAM.
- Which model is appropriate for different response geometries.
1. What Data Do You Need?
JEAM implements multi-dimensional diffusion models that jointly explain response time (RT) and continuous response location and assume that response times reflect evidence accumulation process. Therefore, to use these models appropriately, your dataset must contain:
- Response time (RT) for each decision
- Continuous response value for each decision (e.g., angle, slider position, 2D coordinate)
Each row of your dataset should correspond to a single trial.
JEAM may not be appropriate if:
- The response scale was not continuous
- Response times were not recorded.
- Response times are artificially constrained (e.g., fixed response windows).
- The task design makes response times unrelated to decision dynamics.
- You are only interested in modeling response error distributions without response times.
2. Model Selection by Response Scale
JEAM provides different diffusion models depending on the geometry of the response scale.
A. Circular Response Scales
Examples:
- Color judgment tasks (using a color wheel)
- Orientation judgment tasks
Suitable Model: CircularDiffusionModel
Use when responses lie on a circle (0–2π or 0–360°).
B. Circular Response Scales with Multimodal Response Error
Examples:
- Random dot motion task
- Color judgment tasks (using a color wheel)
Suitable Models: SphericalDiffusionModel and HyperSphericalDiffusionModel
Use when responses lie on a circle (0–2π or 0–360°) and the response error is multimodal.
C. One-Dimensional Bounded Scales
Examples:
- Numerosity estimation tasks
- Pricing tasks
Suitable Model: ProjectedSphericalDiffusionModel
Use when responses are bounded but not circular (i.e., endpoints are distinct). For example, sliders with lower and upper bounds or arc-shaped are suitable.
D. Two-Dimensional Bounded Planes
Examples:
- Centriod estimation tasks
- Spatial working memory tasks
Suitable Model: ProjectedHyperSphericalDiffusionModel
Use when responses are continuous in 2D and bounded within a region.
3. Quick Model Selection Table
| Response Geometry | Example Task | Suitable Model in JEAM |
|---|---|---|
| Circular (1D wrapped) | Color or orientation judgement | CircularDiffusionModel |
| Circular (1D wrapped) | Random dot motion | SphericalDiffusionModel or HyperSphericalDiffusionModel |
| 1D bounded (not wrapped) | Estimation, pricing | ProjectedSphericalDiffusionModel |
| 2D bounded plane | Centriod estimation | ProjectedHyperSphericalDiffusionModel |
4. Summary
JEAM is appropriate when:
- The response scale is continuous.
- You have trial-level RTs.
- You have trial-level continuous responses.
- RTs meaningfully reflect decision latency.
Choosing the correct model depends entirely on the structure of the response scale.