poppr (version 2.9.6)

diversity_ci: Perform bootstrap statistics, calculate, and plot confidence intervals.

Description

This function is for calculating bootstrap statistics and their confidence intervals. It is important to note that the calculation of confidence intervals is not perfect (See Details). Please be cautious when interpreting the results.

Usage

diversity_ci(
  tab,
  n = 1000,
  n.boot = 1L,
  ci = 95,
  total = TRUE,
  rarefy = FALSE,
  n.rare = 10,
  plot = TRUE,
  raw = TRUE,
  center = TRUE,
  ...
)

Value

raw = TRUE

• obs a matrix with observed statistics in columns, populations in rows

• est a matrix with estimated statistics in columns, populations in rows

• CI an array of 3 dimensions giving the lower and upper bound, the index measured, and the population.

• boot a list containing the output of boot::boot() for each population.

raw = FALSE

a data frame with the statistic observations, estimates, and confidence intervals in columns, and populations in rows. Note that the confidence intervals are converted to characters and rounded to three decimal places.

Arguments

tab

a genind(), genclone(), snpclone(), OR a matrix produced from mlg.table().

n

an integer defining the number of bootstrap replicates (defaults to 1000).

n.boot

an integer specifying the number of samples to be drawn in each bootstrap replicate. If n.boot < 2 (default), the number of samples drawn for each bootstrap replicate will be equal to the number of samples in the data set. See Details.

ci

the percent for confidence interval.

total

argument to be passed on to mlg.table() if tab is a genind object.

rarefy

if TRUE, bootstrapping will be performed on the smallest population size or the value of n.rare, whichever is larger. Defaults to FALSE, indicating that bootstrapping will be performed respective to each population size.

n.rare

an integer specifying the smallest size at which to resample data. This is only used if rarefy = TRUE.

plot

If TRUE (default), boxplots will be produced for each population, grouped by statistic. Colored dots will indicate the observed value.This plot can be retrieved by using p <- last_plot() from the ggplot2 package.

raw

if TRUE (default) a list containing three elements will be returned

center

if TRUE (default), the confidence interval will be centered around the observed statistic. Otherwise, if FALSE, the confidence interval will be bias-corrected normal CI as reported from boot::boot.ci()

...

parameters to be passed on to boot::boot() and diversity_stats()

Author

Zhian N. Kamvar

Details

Bootstrapping

For details on the bootstrapping procedures, see diversity_boot(). Default bootstrapping is performed by sampling N samples from a multinomial distribution weighted by the relative multilocus genotype abundance per population where N is equal to the number of samples in the data set. If n.boot > 2, then n.boot samples are taken at each bootstrap replicate. When rarefy = TRUE, then samples are taken at the smallest population size without replacement. This will provide confidence intervals for all but the smallest population.

Confidence intervals

Confidence intervals are derived from the function boot::norm.ci(). This function will attempt to correct for bias between the observed value and the bootstrapped estimate. When center = TRUE (default), the confidence interval is calculated from the bootstrapped distribution and centered around the bias-corrected estimate as prescribed in Marcon (2012). This method can lead to undesirable properties, such as the confidence interval lying outside of the maximum possible value. For rarefaction, the confidence interval is simply determined by calculating the percentiles from the bootstrapped distribution. If you want to calculate your own confidence intervals, you can use the results of the permutations stored in the $boot element of the output.

Rarefaction

Rarefaction in the sense of this function is simply sampling a subset of the data at size n.rare. The estimates derived from this method have straightforward interpretations and allow you to compare diversity across populations since you are controlling for sample size.

Plotting

Results are plotted as boxplots with point estimates. If there is no rarefaction applied, confidence intervals are displayed around the point estimates. The boxplots represent the actual values from the bootstrapping and will often appear below the estimates and confidence intervals.

References

Marcon, E., Herault, B., Baraloto, C. and Lang, G. (2012). The Decomposition of Shannon’s Entropy and a Confidence Interval for Beta Diversity. Oikos 121(4): 516-522.

See Also

diversity_boot() diversity_stats() poppr() boot::boot() boot::norm.ci() boot::boot.ci()

Examples

library(poppr)
data(Pinf)
diversity_ci(Pinf, n = 100L)
if (FALSE) {
# With pretty results
diversity_ci(Pinf, n = 100L, raw = FALSE)

# This can be done in a parallel fasion (OSX uses "multicore", Windows uses "snow")
system.time(diversity_ci(Pinf, 10000L, parallel = "multicore", ncpus = 4L))
system.time(diversity_ci(Pinf, 10000L))

# We often get many requests for a clonal fraction statistic. As this is 
# simply the number of observed MLGs over the number of samples, we 
# recommended that people calculate it themselves. With this function, you
# can add it in:

CF <- function(x){
 x <- drop(as.matrix(x))
 if (length(dim(x)) > 1){
   res <- rowSums(x > 0)/rowSums(x)
 } else {
   res <- sum(x > 0)/sum(x)
 }
 return(res)
}
# Show pretty results

diversity_ci(Pinf, 1000L, CF = CF, center = TRUE, raw = FALSE)
diversity_ci(Pinf, 1000L, CF = CF, rarefy = TRUE, raw = FALSE)
}