tests/testthat/helper-simulation-fixture.R

# Independent strong-signal generator for persistence-specific simulation tests.
# Convergence is separately asserted for every fit used in a regression.
simulation_fixture <- function(nstudent = 72L, nteacher = 12L, seed = 20261024L) {
  stopifnot(nstudent %% nteacher == 0L, nteacher >= 5L)
  old_seed <- get0(".Random.seed", envir = .GlobalEnv, inherits = FALSE)
  on.exit(if (is.null(old_seed)) {
    if (exists(".Random.seed", envir = .GlobalEnv, inherits = FALSE))
      rm(".Random.seed", envir = .GlobalEnv)
  } else assign(".Random.seed", old_seed, envir = .GlobalEnv))
  set.seed(seed)
  nyear <- 3L
  teachers <- vapply(seq_len(nyear), function(g)
    sample(rep(seq_len(nteacher), each = nstudent/nteacher)), integer(nstudent))
  # Centered orthogonal realization columns give each year's teacher effects
  # an exactly full-rank sample covariance instead of an accidentally weak one.
  effects <- lapply(seq_len(nyear), function(g) {
    k <- nyear-g+1L
    z <- matrix(rnorm(nteacher*k), nteacher, k)
    z <- sweep(z, 2L, colMeans(z), "-")
    Q <- qr.Q(qr(z))[, seq_len(k), drop=FALSE]
    covariance <- 1.4^2 * outer(seq_len(k), seq_len(k), function(a,b) 0.35^abs(a-b))
    sqrt(nteacher-1) * Q %*% chol(covariance)
  })
  d <- expand.grid(year = seq_len(nyear), student = seq_len(nstudent))
  d$teacher <- as.character(teachers[cbind(d$student,d$year)])
  d$x <- rnorm(nrow(d))
  student <- as.numeric(scale(rnorm(nstudent))) * 0.5
  error <- matrix(rnorm(nstudent*nyear, sd=0.45), nstudent, nyear)
  d$y <- 1.5 + 0.7*d$year + 0.4*d$x + student[d$student] +
    error[cbind(d$student,d$year)]
  for (i in seq_len(nrow(d))) for (g in seq_len(d$year[i]))
    d$y[i] <- d$y[i] + effects[[g]][teachers[d$student[i],g],d$year[i]-g+1L]
  attr(d,"generating_description") <- paste(
    "Three years, balanced independently reassigned classrooms; teacher SD1.4",
    "with full-rank lag covariance (rho0.35); student SD0.5, residual SD0.45.",
    "The realization has GP/G-compatible structure and stabilized teacher covariance; it is not used for inferential calibration.")
  d
}

Try the GPvam package in your browser

Any scripts or data that you put into this service are public.

GPvam documentation built on Oct. 23, 2026, 1:06 a.m.