Skip to contents

qrme_nmix_select

Source: R/model-select.R
qrme_nmix_select.Rd

Fit qrme with a range of mixture component counts and rank the specifications by AIC and BIC. Use this to choose the number of ME mixture components for a single equation.

The number of free ME parameters is:

  • n_mix = 0: \(k = 0\) (no ME; LL from standard QR)

  • n_mix = 1: \(k = 2\) (\(\mu\), \(\sigma\))

  • n_mix >= 2: \(k = 3m - 1\) (\(\mu_1,\ldots,\mu_m\), \(\sigma_1,\ldots,\sigma_m\), \(\pi_1,\ldots,\pi_{m-1}\))

Usage

qrme_nmix_select(
  formula,
  data,
  tau,
  n_mix_vals = 0:3,
  return_fits = FALSE,
  ...
)

Arguments

formula

formula for the outcome model (passed to qrme)

data

data.frame

tau

quantile levels for QR

n_mix_vals

integer vector of component counts to evaluate (default 0:3; 0 is the no-ME baseline)

return_fits

logical; if TRUE the fitted qrme objects are included in the output (default FALSE)

...

additional arguments passed to qrme for every fit (e.g. me_distribution, mcmc_draws, mcmc_burn_in). n_mix and se are set internally and must not be passed.

Value

a list with:

table

data.frame sorted by BIC with columns n_mix, k_me, ll, AIC, BIC

fits

(only when return_fits = TRUE) named list of qrme objects; the n_mix = 0 entry is NULL

Examples

if (FALSE) { # \dontrun{
  set.seed(1)
  n <- 300; X <- runif(n)
  Y <- 1 + 2 * X + rnorm(n) + rnorm(n, sd = 0.5)
  sel <- qrme_nmix_select(
    Y ~ X, data.frame(Y, X),
    tau          = c(0.25, 0.5, 0.75),
    n_mix_vals   = 0:3,
    mcmc_draws   = 100L,
    mcmc_burn_in = 50L
  )
  sel$table
} # }