VSdecomp (version 0.1.1)

vs_decomp: Skewness and Variance Decomposition

Description

decompose the skewness or the variance of an outcome vector into independent components, either using one or many variables.

Usage

vs_decomp(
  y = NULL,
  X,
  wgt = rep(1, nrow(X)),
  moment = "skewness",
  year = rep(1, nrow(X))
)

Arguments

y

an outcome vector (necessary only when decomposing by one variable).

X

when decomposing by one variable, a matrix or data frame containing the variable for the decomposition. when decomposing by more than one variable, a matrix or data frame containing the linear components of y, calculated by linear_projection.

wgt

an optional vector of weights.

moment

the moment on which the decomposition method is applied. either "variance" (second moment) or "skewness" (third moment).

year

an optional vector of years. if provided, the decomposition is calculated by year. otherwise, the decomposition is calculated for the whole sample.

Value

A "vs_decomp" object. This object is a list containing the estimated components and their standard errors.

Examples

# NOT RUN {
#generate data
n <- 100
men <- rbinom(n, 1, 0.5)
black <- rbinom(n, 1, 0.5)
year <- c(rep(2019, n/2), rep(2020, n/2))
rwage <- function(x){
  m <- x[1]; b <- x[2]; y <- x[3];
  exp(rnorm(1, mean = 1*m + 1*b, sd = 1 + 2020 - y))
}
dat <- data.frame(men, black, year)
dat$wage <- apply(dat, 1, rwage)
#skewness decomposition by one variable:
variable <- as.matrix(dat$men)
decomp_by_gender <- vs_decomp(y = dat$wage, X = variable,
                              moment = "skew", year = dat$year)
summary(decomp_by_gender)
plot(decomp_by_gender)

#skewness decomposition by more than one variable (=linear skewness decomposition):
#first we need to decompose yearly wages into their linear components
#(check ?linear_projection for more details):
wage_linear_comp <- linear_projection(y = "wage",
                                      X.list = list("men", "black"),
                                      data = dat, year = dat$year)
#using the linear components, we can calculate skewness decomposition:
decomp_by_mult_var <- vs_decomp(X = wage_linear_comp,
                                year = dat$year)
#up to 3 components can be summarized or plotted:
colnames(decomp_by_mult_var$components)
#let's take "3cov(epsilon^2,men)" and "3cov(epsilon^2,black)"
summary(decomp_by_mult_var, sum.comp = c(7, 8))
plot(decomp_by_mult_var, plot.comp = c(7, 8))
# }