Likelihood Tools for the ggdmc Package

Description

Efficient computation of likelihoods in design-based choice response time models, including the Decision Diffusion Model, is supported. The package enables rapid evaluation of likelihood functions for both single- and multi-subject models across trial-level data. It also offers fast initialisation of starting parameters for genetic sampling with many Markov chains, facilitating estimation in complex models typically found in experimental psychology and behavioural science. These optimisations help reduce computational overhead in large-scale model fitting tasks.

Author(s)

Yi-Shin Lin <yishinlin001@gmail.com>

See Also

Useful links:

• https://github.com/yxlin/ggdmcLikelihood

Compute Likelihood for Behavioural Models

Description

Computes the likelihood for multiple subjects by aggregating results from individual subject computations.

Usage

compute_likelihood(dmis, parameter_r, debug = FALSE)

compute_subject_likelihood(dmi, parameter_r, debug = FALSE)

Arguments

Details

These functions compute likelihoods for behavioural models, with compute_subject_likelihood handling a single subject and compute_likelihood handling multiple subjects.

These functions provide access to the internal mechanism of the design-based likelihood computation. They primarily intended to initialise new 'samples' or to verify that the likelihood evaluations, when associated with a particular design, are computed accurately.

Value

with:

• compute_likelihood returns a list. Each element is the likelihood for a subject. The element in the inner list is the likelihood for a condition.

• compute_subject_likelihood returns also a list. Each element is the likelihood for a condition.

Examples

# Example dataset
hdat <- data.frame(
  RT = round(runif(15, min = 0.4, max = 1.2), 7),
  R  = sample(c("r1", "r2", "r3"), size = 15, replace = TRUE),
  s  = rep(1:3, each = 5),
  S  = rep(c("s1", "s2", "s3"), each = 5),
  stringsAsFactors = FALSE
)
dat <- hdat[hdat$s==1, ]

p_vector <- c(A = .75, B = 1.25, mean_v.false = 1.5, mean_v.true = 2.5, t0 = .15)
nsubject <- length(unique(hdat$s))

if(requireNamespace("ggdmcModel", quietly = TRUE)) {
    BuildModel <- getFromNamespace("BuildModel", "ggdmcModel")
    BuildDMI   <- getFromNamespace("BuildDMI", "ggdmcModel")

    model <- BuildModel(
        p_map = list(A = "1", B = "1", t0 = "1", mean_v = "M", sd_v = "1", st0 = "1"),
        match_map = list(M = list(s1 = "r1", s2 = "r2")),
        factors = list(S = c("s1", "s2")),
        constants = c(st0 = 0, sd_v = 1),
        accumulators = c("r1", "r2"),
        type = "lba")
    pop_dmis <- BuildDMI(hdat, model)
    sub_dmis <- BuildDMI(dat, model)

    parameters <- list()
    for (i in seq_len(nsubject)) {
        new_p_vector <- p_vector[model@pnames]
        parameters[[i]] <- new_p_vector
    }

    result1 <- compute_subject_likelihood(sub_dmis[[1]], parameters[[1]], FALSE)
    result2 <- compute_likelihood(pop_dmis, parameters, FALSE)
}

print(result1)
print(result2)