How to read a page

Every model page opens with one screen of panels: a drawing of the task, the task settings, the parameters, the predicted responses, and the prediction across task settings. Move any control and every panel follows. The dashed lines and outlines show the model at bmm’s prior means, so a change is always a comparison. Each slider covers the range bmm’s default priors consider plausible. The (i) next to a control opens a note on what that parameter or setting does, with an example from the task. An axis that has to change its limit eases to the new one, and it keeps its limit as long as the curves fill more than half of it, so a slider moved back and forth does not make the axis jump.

The simulate switch in the prediction panel draws a number of trials from the predicted distribution. On the continuous-reproduction pages they appear as a histogram behind the curve, with the expected mean error, circular SD and share near the target next to the same values in the sample. On the M3, MPT, yes/no and rating pages a black line on each bar marks the sample’s share, and on the yes/no and rating pages the sample’s rates also appear as points on the ROC. On the m-AFC page a black dot marks the sample’s proportion correct at the current d. The sample keeps its random numbers while the sliders move, and New sample draws other ones; how far a sample of 50 or 200 trials strays from the expected values is a view of the noise in a real data set. For the circular SD of widely spread errors part of the gap is a bias, not noise: a small sample gives a lower value than expected.

Below the panels, every page has the same five parts.

  1. What the model says. The model in a few sentences and its formula, with the parameters named as bmm names them.
  2. The task. A drawing of the experimental trial the model is written for. Where the task has settings that enter the model (how many items, how many response candidates), the drawing follows the controls.
  3. Explore. How to read the panels, and where the slider ranges come from.
  4. Across task settings. Some models predict how responses change when the task changes while the parameters stay the same: more candidates, more items on the screen. Those pages show that prediction. Models whose likelihood has no such setting say so, which is worth knowing before an experiment is designed around one.
  5. In bmm. The call that fits the model you are looking at, and the density call with the current values, ready to paste into R.

The curves are computed in your browser by small JavaScript copies of bmm’s density functions. Each copy is tested against values produced by bmm itself; the test and the fixtures are in the repository.