kmatch: Create matching vectors based on data

In davidkane9/kmatching: Creates random vectors fulfilling constraints

Description

Given a data frame and the specified matching vectors and weight vector, kmatch extracts the matching vectors and weight vector from the data set, and form them to a system of linear equations, whose matrix expression is Aw=b, where A is the matching vectors and b is the weight vector. The w is the weights we want. kmatch then randomly samples w subject to the non-negativity constraint w_i >= 0 and sum(w)=0. These two make the sampling space bounded so uniform sampling over the whole sampling space is achievable. Then the w is the weights we want.

Usage

kmatch(x, weight.var, match.var, n = 1, thin = 5, replace = FALSE,
  chains = 1, burn = 0.5)

Arguments

x	is a data frame containing needed input data
weight.var	character name of the column of the input weights
match.var	character vector of names of columns of 'data' we wish to match on
n	numeric number of weight vectors desired. Default is 1
thin	only 1 out of every thin vectors will be recorded. Default is 5
replace	if FALSE then any rows that had non-zero weight in weight.var will be set to 0 in the output. Default is FALSE
chains	number of chains. Default is 1.
burn	is the number of points to burn in hitandrun (not very costly, because it just adds to total number of iterations, whereas thin is a multiple)

Value

Returns a list of "chains" matrices of 'n' sets of weights that match the given set of weights in terms of weighted averages to the 'match.var' factors. The columns of each matrix are the sets of weights.

Author(s)

Mike Flynn [e-mail]

Examples

set.seed(40)
x <- data.frame(size = rnorm(10), weight = rep(.02, 10))
weights <- kmatch(x, weight.var = "weight", match.var = "size", n = 100, replace = TRUE)

x2 <- data.frame(var1 = c(-2,-1, 0, 1, 2),
                 var2 = c(1, -1, 2, -1, 0),
                 w = rep(.2, 5))
weights2 <- kmatch(x = x2, weight.var = "w",
                   match.var = c("var1", "var2"),
                   n = 10, thin = 1, replace= TRUE)[[1]]

davidkane9/kmatching documentation built on May 15, 2019, 1:14 a.m.