M2 response + RT + gaze reference specification

Creates a thin M2 specialization of the established [multimodal_irt_spec()] / `irt_model_spec()` architecture. The measurement likelihood follows Man, Harring, and Zhan (2022): Rasch response, lognormal response time, and negative-binomial gaze-fixation counts. Priors are implemented in Stan using either a regularized profile or a paper-centered profile; the latter is not claimed to reproduce every published hyperprior exactly.

Usage

multimodal_m2_spec(
  backend = "cmdstanr",
  prior_profile = c("regularized", "paper_centered"),
  missingness = "ignorable"
)

Arguments

backend

Currently `"cmdstanr"` only.

prior_profile

`"regularized"` or `"paper_centered"`.

missingness

Currently `"ignorable"` only. Channel-specific missing observations contribute no level-1 likelihood term.

Value

An `eye_multimodal_m2_spec`, inheriting the established `eye_multimodal_irt_spec` and `eye_irt_model_spec` classes.