geoWeight: Split weights across geographies

In donboyd5/geoWeight: Distribute Microdata Record Weights Across Different Geographies

Description

geoWeight calculates state weights for each household in a microdata file that add up to the household total weight, such that weighted state totals for selected characteristics hit or come close to desired targets

Usage

geoWeight(
  wh,
  xmat,
  targets,
  dweights = get_dweights(targets),
  method = "Broyden",
  betavec = rep(0, length(targets)),
  maxiter = NULL,
  opts = NULL
)

Arguments

wh	vector of household total weights, length h (see h, s, k definitions below)
xmat	h x k matrix of data for households
targets	s x k matrix of desired target values
dweights	optional vector of weighting factors for targets, length s * k
method	optional parameter for approach to use; must be one of c('Broyden', 'Newton', 'LM'); default is 'Broyden'
betavec	optional vector of initial guess at parameters, length s * k; default is zero for all

Details

geoWeight uses the solver nleqslv or the solver nls.lm depending on user choice.

Value

A list with the following elements:

h

number of households (or individuals, records, tax returns, etc.)

s

number of states (or other geographies or subgroups)

k

number of characteristics each household has

solver_message

message from the solver that was used

etime

elapsed time

beta_opt_mat

s x k matrix of optimal parameters

whs

h x s matrix of state weights for each household, computed using the optimal parameters

wh

the input vector of household total weights, length h

output

list of output from the solver that was used

Examples

# Example 1: Determine state weights for a simple problem with random data
p <- make_problem(h=10, s=3, k=2)
dw <- get_dweights(p$targets)

res1 <- geoWeight(wh = p$wh, xmat = p$xmat, targets = p$targets,
  dweights = dw)

res2 <- geoWeight(wh = p$wh, xmat = p$xmat, targets = p$targets,
  dweights = dw, method = 'Newton')

res3 <- geoWeight(wh = p$wh, xmat = p$xmat, targets = p$targets,
  dweights = dw, method = 'LM')

res1
res2
res3
c(res1$sse_unweighted, res2$sse_unweighted, res3$sse_unweighted)

# verify that the state weights produce the desired targets
whs <- get_weights(beta, delta, xmat)
t(res2$whs) %*% p$xmat
p$targets

donboyd5/geoWeight documentation built on July 5, 2020, 8:55 p.m.