Subsample bootstrapped communities to assess power for detecting change across community coverage
subsample_boots.RdDownsample bootstrapped communities to lower sample sizes representing a gradient of coverage values, and assess the proportion of correct-direction detections within each coverage interval.
Usage
subsample_boots(
boots,
pilot,
method = "single",
power = c(80),
target_eff_size = NULL,
cost_per_sample = NULL,
seed = NULL,
analysis_type = "sign",
effect_minimum = NULL
)Arguments
- boots
tibble of bootstrapped communities across multiple effect sizes. Output from `bootstrap_pilot()`.
- pilot
original pilot study. Identical to the pilot input in `bootstrap_pilot()`
- method
"single" or "two" for single-treatment power analysis (one community over time), or two-treatment power analysis (comparing differences between two treatment communities)
- power
statistical power with which the user wants to detect richness change (defaults to 80).
- target_eff_size
if supplied, the level of richness difference the user wants to detect (in log2 ratio units). Adding a target value will change outputs of the function, specifying sampling needs to reach the richness difference target.
- cost_per_sample
cost per unit sample (linear only), which will add an axis to output plots for total cost per unit power.
- seed
random seed. Defaults to 1 so repeat runs will be identical, but since simulations rely on random draws, changing the seed will result in different answers.
- analysis_type
"sign" or "minimum". Describes the type of power analysis to conduct. "sign" identifies the number of samples needed to detect an effect in the same direction as the true effect, "minimum" identifies the number of samples needed to detect an effect both in the same direction and above user-supplied biologically-relevant effect.
- effect_minimum
if `analysis_type` is "minimum", a user-defined minimum effect size to form a minimum-effect null distribtuion.
Value
List of outputs: (1) Plot of convergence of detections across community coverage in multiple effect size bins. (2) Plot of mean detection across community coverage within multiple effect size bins. (3) Plot of proportion of correct detections across community coverage within multiple effect size bins. (4) Plot of functional relationship between sample size and power to detect richness differences. (5) Tibble of input and output values.