bmm model explorer

What does each parameter do? Move it and see.

The pages below belong to the measurement models in the bmm R package. Each one lets you move a model’s parameters and watch the predicted responses change, shows the task the model is written for, and, where the model has something to say about it, what happens when the task changes while the parameters do not. Nothing is installed and nothing is sent anywhere: the curves are computed in your browser. New here? Read how to read a page first.

Start from your response

What does one response in your task look like? Find it below. Under each kind of response, every model after the first names a model above it and says what it adds to that model or does differently.

• A position on a circle, such as a colour reproduced on a colour wheel
  • mixture2p: is a response a memory of the target or a guess? The share of memory responses and how precise they are.
  • mixture3p: mixture2p plus responses to the other items on the display. Do some responses go to a non-target?
  • imm: an alternative to mixture3p, with activations instead of proportions. Every item adds activation, more the closer it is to the cued location: does the target’s share fall as items are added?
  • sdm: an alternative to the mixture models, with no separate guessing parameter: guessing is the limit of weak memory. How strong is memory and how precise, in one distribution?
• A choice among candidates, such as recalling a cued item from a list when the response set is shown
  • m3: recall as a competition of activations. How strongly do the correct item, the other list items, distractors and lures each draw a response, and how does recall change with the number of candidates?
  • mpt: an alternative to m3 for simple-span recall, with discrete states instead of activations. Is the cued item in memory, is its position retrieved, or is the response a guess?
• A yes/no answer, a choice among m alternatives, or a confidence rating, such as old/new recognition
  • sdt_yn: how far apart are the evidence distributions of noise and signal trials, and where does the observer put the criterion, the point above which the answer is yes?
  • sdt_mafc: a choice among m alternatives, with a d on the scale of the d of sdt_yn and no criterion. How does accuracy fall as m grows?
  • sdt_rating: sdt_yn with several criteria at once: ratings on K levels. Where do K − 1 thresholds cut the evidence axis, and which points of the ROC (receiver operating characteristic) do they give?
• A response time and one of two responses, such as a lexical decision
  • ezdm: what probability of each response and what mean and variance of the response times follow from the drift, the boundary, the non-decision time and, in 4par, the starting point?
  • ddm: the diffusion model that ezdm summarises, adding the whole response-time distribution of both responses. How do speed and accuracy trade as the boundary moves?
  • cswald: a simpler alternative to ddm for tasks with few errors: an error is a censored trial (simple) or the response of a second accumulator that wins the race (crisk). How far does it part from ddm as errors become frequent?

Models by family

The same models by family, with what each page shows.

Continuous reproduction tasks

• Interference Measurement Model (imm): responses on a colour wheel drawn to the target, to the other items by their closeness to the cued location, and to noise. Set size and spacing are part of the model.
• Two-parameter mixture model (mixture2p): responses centred on the target, or guesses. No task setting enters the likelihood.
• Three-parameter mixture model (mixture3p): adds responses to the non-targets. Their colours and their number enter the density; from set size 2 on, more non-targets split the same weight.
• Signal discrimination model (sdm): one distribution with a memory strength and a precision, and no separate guessing parameter. No task setting enters the likelihood.

Categorical decision tasks

• Memory Measurement Model (m3): a correct item, other list items, distractors and new lures compete for a response, in simple and complex span, under three choice rules. Candidate counts are part of the model.
• Multinomial processing tree, span task (mpt): recall as discrete states, with guessing rates set by the response set, and how it relates to the M3.

Signal detection

• Yes/no (sdt_yn): noise and signal evidence against one criterion, given from the noise location or as a bias from the midpoint, under four evidence distributions, with the ROC and why one condition cannot identify sdratio.
• m-alternative forced choice (sdt_mafc): the signal among m alternatives. The number of alternatives is part of the model, and accuracy across m is what tells the evidence distributions apart.
• Confidence ratings (sdt_rating): K − 1 thresholds on one evidence axis, the category proportions, and the cumulative rates as points on the ROC.

Response times

• EZ Diffusion Model (ezdm): the diffusion model’s probability of an upper response and the mean and variance of the response times in closed form, and how far the observed ones scatter around them for a given number of trials per cell.
• Censored Shifted Wald Model (cswald): evidence for the correct response accumulates to one boundary, errors censored or racing; compared with the diffusion model.
• Diffusion Decision Model (ddm): response-time densities of both responses, and how speed and accuracy trade as the boundary separation changes.