README

Installation

You can install the {brms.exgaussian} from GitHub:

remotes::install_github("mattansb/brms.exgaussian")

(You will need to specify the full path to the files. E.g., C:\\User\\Cool R Code\\brms.exgaussian_0.0.2.zip.)

Demo

This README demos {brms.exgaussian} with parameter recovery.

library(brms)
library(brms.exgaussian)

Generate Data

Using retimes::rexgauss() with standard Ex-Gauss parameterization. (Also available as brms.exgaussian::rexgaussian2().)

truth <- data.frame(
  Group = 1:2,
  mu = c(150, 180),
  sigma = c(50, 70),
  tau = c(200, 100)
)

g <- sample(1:2, size = 1000, replace = TRUE)

d <- data.frame(
  Group = factor(g),
  rt = retimes::rexgauss(1000,
    mu = truth$mu[g],
    sigma = truth$sigma[g],
    tau = truth$tau[g]
  )
)

Fit brms Model

Functions from {brms.exgaussian} for standard Ex-Gaussian distribution.

exg_stanvars <- exgaussian2_stancode()
exg_family <- exgaussian2()

Fit the model:

fit <- brm(
  bf(
    rt ~ Group,
    tau ~ Group,
    sigma ~ Group
  ),
  data = d,
  
  # brms.exgaussian additions
  family = exg_family,
  stanvars = exg_stanvars
)

Support Via Internal Functions

These all work…

loo(fit)
waic(fit)
bridgesampling::bridge_sampler(fit)

posterior_epred(fit)

posterior_predict(fit)

Parameter Recovery

PP-Check

pp_check(fit, group = "Group", type = "dens_overlay_grouped")

Actual Parameters

Recovered with {brms.exgaussian}:

nd <- data.frame(Group = factor(1:2))

mu <- posterior_linpred(fit,
  newdata = nd,
  dpar = "mu", transform = TRUE
)

sigma <- posterior_linpred(fit,
  newdata = nd,
  dpar = "sigma", transform = TRUE
)

tau <- posterior_linpred(fit,
  newdata = nd,
  dpar = "tau", transform = TRUE
)




nd$mu <- apply(mu, 2, mean)
nd$sigma <- apply(sigma, 2, mean)
nd$tau <- apply(tau, 2, mean)

nd
#>   Group       mu    sigma       tau
#> 1     1 144.0746 43.65374 207.68073
#> 2     2 193.6895 73.64891  92.83953

Recovered with {retimes}:

cbind.data.frame(
  Group = 1:2, 
  by(d$rt, d$Group, retimes::timefit) |> 
    lapply(function(p) as.data.frame(as.list(p@par))) |> 
    do.call(what = "rbind")
)
#>   Group       mu    sigma       tau
#> 1     1 144.0544 43.29852 207.28435
#> 2     2 192.9664 72.99485  93.22048

Ground truth:

truth
#>   Group  mu sigma tau
#> 1     1 150    50 200
#> 2     2 180    70 100