Spower (version 0.5.1)

p_prop.test: p-value from proportion test simulation

Description

Generates single and multi-sample data for proportion tests and return a p-value. Uses binom.test for one-sample applications and prop.test otherwise.

Usage

p_prop.test(
  n,
  h,
  prop = NULL,
  pi = 0.5,
  n.ratios = rep(1, length(prop)),
  two.tailed = TRUE,
  correct = TRUE,
  exact = FALSE,
  gen_fun = gen_prop.test,
  return_analysis = FALSE,
  ...
)
gen_prop.test(
  n,
  h,
  prop = NULL,
  pi = 0.5,
  n.ratios = rep(1, length(prop)),
  ...
)

Value

a single p-value

Arguments

n

sample size per group

h

Cohen's h effect size; only supported for one-sample analysis.

Note that it's important to specify the null value pi when supplying this effect size as the power changes depending on these specific values (see example below).

prop

sample probability/proportions of success. If a vector with two-values or more elements are supplied then a multi-samples test will be used. Matrices are also supported

pi

probability of success to test against (default is .5). Ignored for two-sample tests

n.ratios

allocation ratios reflecting the sample size ratios. Default of 1 sets the groups to be the same size (n * n.ratio)

two.tailed

logical; should a two-tailed or one-tailed test be used?

correct

logical; use Yates' continuity correction?

exact

logical; use fisher's exact test via fisher.test? Use of this flag requires that prop was specified as a matrix

gen_fun

function used to generate the required discrete data. Object returned must be a matrix with two rows and 1 or more columns. Default uses gen_prop.test. User defined version of this function must include the argument ...

return_analysis

logical; return the analysis object for further extraction and customization?

...

additional arguments to be passed to gen_fun. Not used unless a customized gen_fun is defined

Author

Phil Chalmers

See Also

gen_prop.test

Examples

# one sample, 50 observations, tested against pi = .5 by default
p_prop.test(50, prop=.65)

# return analysis model
p_prop.test(50, prop=.65, return_analysis = TRUE)

# specified using h and pi
h <- pwr::ES.h(.65, .4)
p_prop.test(50, h=h, pi=.4)
p_prop.test(50, h=-h, pi=.65)

# two-sample test
p_prop.test(50, prop=c(.5, .65))

# two-sample test, unequal ns
p_prop.test(50, prop=c(.5, .65), n.ratios = c(1,2))

# three-sample test, group2 twice as large as others
p_prop.test(50, prop=c(.5, .65, .7), n.ratios=c(1,2,1))

# Fisher exact test
p_prop.test(50, prop=matrix(c(.5, .65, .7, .5), 2, 2))

# \donttest{
    # compare simulated results to pwr package

    # one-sample tests
    (h <- pwr::ES.h(0.5, 0.4))
    pwr::pwr.p.test(h=h, n=60)

    # uses binom.test (need to specify null location as this matters!)
    Spower(p_prop.test(n=60, h=h, pi=.4))
    Spower(p_prop.test(n=60, prop=.5, pi=.4))

    # compare with switched null
    Spower(p_prop.test(n=60, h=h, pi=.5))
    Spower(p_prop.test(n=60, prop=.4, pi=.5))

    # two-sample test, one-tailed
    (h <- pwr::ES.h(0.67, 0.5))
    pwr::pwr.2p.test(h=h, n=80, alternative="greater")
    p_prop.test(n=80, prop=c(.67, .5), two.tailed=FALSE,
      correct=FALSE) |> Spower()

    # same as above, but with continuity correction (default)
    p_prop.test(n=80, prop=c(.67, .5), two.tailed=FALSE) |>
      Spower()

    # three-sample joint test, equal n's
    p_prop.test(n=50, prop=c(.6,.4,.7)) |> Spower()

# }