Skip to contents

Calibrate a Bayesian t-test design

calibrate_bf_ttest.Rd

Searches an integer sample-size range and returns the smallest design that satisfies the requested frequentist and Bayesian operating-characteristic constraints for a one-sample or paired Bayesian t-test.

Usage

calibrate_bf_ttest(
  design = c("one.sample", "paired", "two.sample.pooled"),
  hypothesis = c("two.sided.pointnull", "greater.pointnull", "greater.composite"),
  n_min,
  n_max,
  k1 = 1/10,
  k0 = 10,
  analysis_prior_sd = 0.5,
  analysis_prior = NULL,
  freq_t1e = NULL,
  freq_power = NULL,
  bayes_power = NULL,
  bayes_t1e = NULL,
  pce_h0 = NULL,
  design_prior_h1 = list(type = "point", effect = 0.4),
  design_prior_h0 = list(type = "point", effect = 0),
  analysis_prior_h0 = NULL,
  analysis_prior_h1 = NULL,
  control = NULL
)

Arguments

design

Design type. One of "one.sample", "paired", or "two.sample.pooled". The current implementation supports "one.sample" and "paired".

hypothesis

Hypothesis configuration. One of "two.sided.pointnull", "greater.pointnull", or "greater.composite".

n_min

Minimum candidate sample size.

n_max

Maximum candidate sample size.

k1

Evidence threshold for evidence in favour of the alternative: evidence for H1 is declared when BF01 <= k1.

k0

Evidence threshold for evidence in favour of the null: evidence for H0 is declared when BF01 >= k0.

analysis_prior_sd

Standard deviation of the default analysis prior. Used when a relevant explicit analysis prior is not supplied.

analysis_prior

Analysis prior for point-null hypotheses. If NULL, a default prior determined by hypothesis and analysis_prior_sd is used.

freq_t1e

Optional upper bound on the frequentist type-I error, calculated under the null effect.

freq_power

Optional list of frequentist power specifications. Each element must contain effect and may contain a target target.

bayes_power

Optional lower bound on Bayesian power, averaged under design_prior_h1.

bayes_t1e

Optional upper bound on the Bayesian type-I error, averaged under design_prior_h0.

pce_h0

Optional lower bound on the probability of compelling evidence for H0, averaged under design_prior_h0.

design_prior_h1

Design-prior specification under H1.

design_prior_h0

Design-prior specification under H0.

analysis_prior_h0

Analysis prior under H0 for "greater.composite" hypotheses.

analysis_prior_h1

Analysis prior under H1 for "greater.composite" hypotheses.

control

Optional named list controlling numerical integration, root-finding, search strategy, verification, and progress reporting.

Value

An object of class "bf_ttest_design". On success, it contains the selected sample size (n), decision regions, operating characteristics, a constraint table, search results, and plot data. If no feasible design is found, it returns a "bf_ttest_design" object with success = FALSE.

Examples

if (FALSE) { # \dontrun{
design <- calibrate_bf_ttest(
  design = "one.sample",
  hypothesis = "greater.composite",
  n_min = 20,
  n_max = 120,
  k1 = 1 / 20,
  k0 = 3,
  freq_t1e = 0.05,
  freq_power = list(main = list(effect = 0.40, target = 0.80)),
  design_prior_h0 = list(
    type = "truncnormal",
    mean = -0.10,
    sd = 0.25,
    lower = -Inf,
    upper = 0
  ),
  design_prior_h1 = list(
    type = "truncnormal",
    mean = 0.40,
    sd = 0.20,
    lower = 0,
    upper = Inf
  ),
  control = list(progress = FALSE)
)
} # }