Nominal data and Kruschke's "conditional logistic" approach

Interfaces brms
5

Hyea šŸ‘‹

Using the sequential ordinal regression (ref. https://doi.org/10.1177/2515245918823199):

library(brms)
library(dplyr)
library(purrr)

dat <- data.frame(Y = ordered(1:4),
                  w = c(80, 114, 138, 143))

fit <- brm(Y | resp_weights(w) ~ 1, 
           family = sratio(), 
           data = dat,
           backend = "cmdstanr")

# True parameters (in Odds Ratios):
map_dbl(1:3, ~ dat$w[.x] / sum(dat$w[-(1:.x)]))
#> [1] 0.2025316 0.4056940 0.9650350

# Recoverd parameters
parameters::model_parameters(fit, test = NULL) |> 
  mutate(across(c(2,3:4), exp))
#> Parameter    | Median |        95% CI |  Rhat |     ESS
#> -------------------------------------------------------
#> Intercept[1] |   0.20 | [0.16, -1.36] | 1.000 | 3875.00
#> Intercept[2] |   0.41 | [0.32, -0.68] | 1.000 | 3687.00
#> Intercept[3] |   0.97 | [0.77,  0.21] | 0.999 | 4128.00
4 Likes