Sample size for a future study, sized twice

Solves the two-sample design for the number of units per arm needed to reach power at sig_level, once for the observed effect and once for the confidence bound nearer the null.

Usage

cr_power(
  effect_size = NULL,
  ci_low = NULL,
  ci_high = NULL,
  power = 0.8,
  sig_level = 0.05,
  type = c("two.sample", "one.sample", "paired"),
  alternative = c("two.sided", "one.sided")
)

Arguments

effect_size

Numeric vector of observed standardized effects (Cohen's d). May be NULL when only interval bounds are given.

ci_low, ci_high

Numeric vectors of interval bounds for the same effects. Optional.

power

Target power (default 0.8).

sig_level

Significance level (default 0.05).

type

Design: "two.sample" (default), "one.sample" or "paired".

alternative

"two.sided" (default) or "one.sided".

Value

A tibble with one row per effect: d_observed, n_observed, d_conservative, n_conservative, basis, power and sig_level. Sample sizes are units per arm, rounded up.

Details

The conservative figure is the reportable one. Powering a screen's leading compound on its own point estimate is circular: that compound is the largest of the set only by virtue of having been selected for being largest, so a sample size derived from it can hardly fail to be met. n_conservative is NA whenever the interval spans the null, because an interval compatible with no effect cannot be sized.

The solver is stats::power.t.test(), so the effect size is read as a standardized mean difference and n is per group.

See also

cr_power_grid(), cr_conservative_effect(), cr_power_analysis().

Other power: cr_conservative_effect(), cr_power_analysis(), cr_power_grid()

Examples

cr_power(effect_size = c(-1.31, -0.28),
         ci_low = c(-2.35, -1.20),
         ci_high = c(-0.27, 0.64))
#> # A tibble: 2 × 7
#>   d_observed n_observed d_conservative n_conservative basis      power sig_level
#>        <dbl>      <dbl>          <dbl>          <dbl> <chr>      <dbl>     <dbl>
#> 1      -1.31         11          -0.27            217 confidenc…   0.8      0.05
#> 2      -0.28        202          NA                NA interval …   0.8      0.05

# without an interval only the (circular) observed sizing is possible
cr_power(effect_size = 0.8)
#> # A tibble: 1 × 7
#>   d_observed n_observed d_conservative n_conservative basis      power sig_level
#>        <dbl>      <dbl>          <dbl>          <dbl> <chr>      <dbl>     <dbl>
#> 1        0.8         26             NA             NA observed …   0.8      0.05