Posterior credible interval tests

Modeled after t.test, returns the credible interval of the parameter or test and what proportion of the posterior distribution (or the difference in posterior distributions in case of a two sample test) overlaps with mu. For a one sample test provide x and for two sample also provide y. Note that for comparisons within one model, we recommend using hypothesis() if the priors were well chosen.

Usage

# S3 method for class 'emc'
credible(
  x,
  x_name = NULL,
  x_fun = NULL,
  x_fun_name = "fun",
  selection = "mu",
  y = NULL,
  y_name = NULL,
  y_fun = NULL,
  y_fun_name = "fun",
  x_subject = NULL,
  y_subject = NULL,
  mu = 0,
  alternative = c("less", "greater")[1],
  probs = c(0.025, 0.5, 0.975),
  digits = 2,
  p_digits = 3,
  print_table = TRUE,
  ...
)

credible(x, ...)

Arguments

x

An emc object

x_name

A character string. Name of the parameter to be tested for x

x_fun

Function applied to the MCMC chains to create variable to be tested.

x_fun_name

Name to give to quantity calculated by x_fun

selection

A character string designating parameter type (e.g. alpha or covariance)

y

A second emc object

y_name

A character string. Name of the parameter to be tested for y

y_fun

Function applied to the MCMC chains to create variable to be tested.

y_fun_name

Name to give to quantity calculated by y_fun

x_subject

Integer or name selecting a subject

y_subject

Integer or name selecting a subject

mu

Numeric. NULL value for single sample test if y is not supplied (default 0)

alternative

less or greater determining direction of test probability

probs

Vector defining quantiles to return.

digits

Integer, significant digits for estimates in printed results

p_digits

Integer, significant digits for probability in printed results

print_table

Boolean (defaults to TRUE) for printing results table

...

Additional optional arguments that can be passed to get_pars

Value

Invisible results table with no rounding.

Examples

if (FALSE) { # \dontrun{
# Run a credible interval test (Bayesian ''t-test'')
# Here the full model is an emc object with the hypothesized effect
design_full <- design(data = forstmann,model=DDM,
                           formula =list(v~0+S,a~E, t0~1, s~1, Z~1, sv~1, SZ~1),
                           constants=c(s=log(1)))

full_model <- make_emc(forstmann, design_full)
full_model <- fit(full_model)
credible(full_model, x_name = "v")
# We can also compare between two sets of emc objects
# Now without a ~ E
design_null <- design(data = forstmann,model=DDM,
                           formula =list(v~0+S,a~1, t0~1, s~1, Z~1, sv~1, SZ~1),
                           constants=c(s=log(1)))

null_model <- make_emc(forstmann, design_null)
null_model <- fit(null_model)
credible(x = null_model, x_name = "a", y = full_model, y_name = "a")

# Or provide custom functions
credible(x = full_model, x_fun = function(d) d["a_Eaccuracy"] - d["a_Eneutral"])
} # }