These functions are designed to compute weighted statistics (mean, variance, standard deviation, covariance, correlation, median and quantiles) to perform weighted transformation (scaling, centering, and standardization) using bootstrap weights. These automatically extract bootstrap weights when called from within fwb() to facilitate computation of bootstrap statistics without needing to think to hard about how to correctly incorporate weights.
w_mean(x, w = NULL, na.rm = FALSE)w_var(x, w = NULL, na.rm = FALSE)
w_sd(x, w = NULL, na.rm = FALSE)
w_cov(x, w = NULL, na.rm = FALSE)
w_cor(x, w = NULL)
w_quantile(
x,
w = NULL,
probs = seq(0, 1, by = 0.25),
na.rm = FALSE,
names = TRUE,
type = 7L,
digits = 7L
)
w_median(x, w = NULL, na.rm = FALSE)
w_std(x, w = NULL, na.rm = TRUE, scale = TRUE, center = TRUE)
w_scale(x, w = NULL, na.rm = TRUE)
w_center(x, w = NULL, na.rm = TRUE)
w_mean(), w_var(), w_sd(), and w_median() return a numeric scalar. w_cov() and w_cor() return numeric matrices. w_quantile() returns a numeric vector of length equal to probs. w_std(), w_scale(), and w_center() return numeric vectors of length equal to the length of x.
a numeric variable; for w_cov() and w_cor(), a numeric matrix.
optional; a numeric vector of weights with length equal to the length of x (or number of rows if x is a matrix). If unspecified, will use bootstrapped weights when called from within fwb() or vcovFWB() and unit weights (i.e., for unweighted estimates) otherwise.
logical; whether to exclude NA values in the weights and x when computing statistics. Default is FALSE for the weighted statistics (like with their unweighted counterparts) and TRUE for weighted transformations.
see quantile(). Only type = 7 is allowed.
logical; whether to scale or center the variable.
These function automatically incorporate bootstrap weights when used inside fwb() or vcovFWB(). This works because fwb() and vcovFWB() temporarily set an option with the environment of the function that calls the estimation function with the sampled weights, and the w_*() functions access the bootstrap weights from that environment, if any. So, using, e.g., w_mean() inside the function supplied to the statistic argument of fwb(), computes the weighted mean using the bootstrap weights. Using these functions outside fwb() works just like other functions that compute weighted statistics: when w is supplied, the statistics are weighted, and otherwise they are unweighted.
See below for how each statistic is computed.
For all weighted statistics, the weights are first rescaled to sum to 1. \(w\) in the formulas below refers to these weights.
w_mean()Calculates the weighted mean as
$$\bar{x}_w = \sum_i{w_i x_i}$$
This is the same as weighted.mean().
w_var()Calculates the weighted variance as $$s^2_{x,w} = \frac{\sum_i{w_i(x_i - \bar{x}_w)^2}}{1 - \sum_i{w_i^2}}$$
w_sd()Calculates the weighted standard deviation as $$s_{x,w} = \sqrt{s^2_{x,w}}$$
w_cov()Calculates the weighted covariances as
$$s_{x,y,w} = \frac{\sum_i{w_i(x_i - \bar{x}_w)(y_i - \bar{y}_w)}}{1 - \sum_i{w_i^2}}$$
This is the same as cov.wt().
w_cor()Calculates the weighted correlation as
$$r_{x,y,w} = \frac{s_{x,y,w}}{s_{x,w}s_{y,w}}$$
This is the same as cov.wt().
w_quantile()Calculates the weighted quantiles using linear interpolation of the weighted cumulative density function.
w_median()Calculates the weighted median as w_quantile(., probs = .5).
Weighted transformations use the weighted mean and/or standard deviation to re-center or re-scale the given variable. In the formulas below, \(x\) is the original variable and \(w\) are the weights.
w_center()Centers the variable at its (weighted) mean. $$x_{c,w} = x - \bar{x}_w$$
w_scale()Scales the variable by its (weighted) standard deviation. $$x_{s,w} = x / s_{x,w}$$
w_std()Centers and scales the variable by its (weighted) mean and standard deviation. $$x_{z,w} = (x - \bar{x}_w) / s_{x,w}$$
w_scale() and w_center() are efficient wrappers to w_std() with center = FALSE and scale = FALSE, respectively.
mean() and weighted.mean() for computing the unweighted and weighted mean
var() and sd() for computing the unweighted variance and standard deviation
median() and quantile() for computing the unweighted median and quantiles
cov(), cor(), and cov.wt() for unweighted and weighted covariance and correlation matrices
scale() for standardizing variables using arbitrary (by default, unweighted) centering and scaling factors