Package 'pgsc'

Wrapper function for GSC estimation

Description

Wrapper function for GSC estimation

Usage

pgsc(dta, dep.var, indep.var, b.init, method, sol.it = NULL,
  wt.init = NULL, print.level = 0, g.i = NULL, g.i.grad = NULL,
  ...)

Arguments

dta	A data frame
dep.var	A string defining the dependent variable
indep.var	A vector of strings defining the independent (treatment) variables
b.init	An initial value for the treatment variable coefficients. Must have same length as 'indep.var'
method	The GSC iteration method to be used. Must be one of: onestep: "Plain" GSC solution, without weights twostep.aggte: Observations weighted by unit MSEs from the one-step solution. twostep.indiv: Observations weighted by unit MSEs from individual, unit-by-unit unweighted solutions.
sol.it	The first step solution used in the two-step methods. If omitted, a new one-step solution is computed.
wt.init	An initial value for the weighting matrix
print.level	The level of detail provided in the printed output
g.i	A function defining a restriction on the parameters. Used in hypothesis testing.
g.i.grad	The gradient of g.i.
...	Other arguments to be passed to the optimization

Details

See the vignette "Using pgsc" for an extended example.

Value

Returns the point estimate of the model as a gsc object, a list with entries:

b

The point estimate of the coefficients on the dependent variables

diff

The difference between successive iterations

err

The maximum error on the within-iteration optimization problems

it

Number of iterations require to solve

sig.i

The unit-specific MSEs from the solution

W

The "full" weighting matrix for counterfactuals, containing own-unit weights (all zero) and unit-N weights

wt

The "minimal" weighting matrix, omitting own-unit weights and weights on unit N (which can be computed as one-minus-rowsum)

Examples

data("pgsc.dta")
sol <- pgsc(pgsc.dta, dep.var = 'y', indep.var = c('D1','D2'), 
b.init = c(0,0), method='onestep' )
summary(sol)
g.i <- function(b) b[1] ; g.i.grad <- function(b) c(1,0)
sol.r <- pgsc(pgsc.dta, dep.var = 'y', indep.var = c('D1','D2'), 
b.init = sol$b, method='onestep', g.i=g.i, g.i.grad=g.i.grad )
summary(sol.r)

---

Synthetic data for PGSC testing.

Description

A dataset with an outcome given by a treatment and a set of factors.

Usage

pgsc.dta

Format

A data frame with 750 rows and 8 variables:

n

The unit, here labeled as a US state

t

The time period

y

The outcome variable

D1

The first treatment variable

D2

The second treatment variable

X1

The first observed confounding factor

X2

The second observed confounding factor

X3

The third observed confounding factor

Source

Generated by code in the package vignette "Using pgsc".

---

A wrapper for the wald test of a restricted solution

Description

A wrapper for the wald test of a restricted solution

Usage

pgsc.wald.test(dta, dep.var, indep.var, sol.rest, n.boot = 10000,
  seed = 42)

Arguments

dta	A dataframe
dep.var	A vector of strings of names of dependent variables.
indep.var	A vector of strings of names of independent (treatment) variables.
sol.rest	A restricted solution which is being tested
n.boot	The number of bootstrapped samples for the variance calculation. Default is 10000.
seed	Randomization seed. Default is 42.

Details

See the vignette "Using pgsc" for an extended example.

Value

Returns the wald test as gsc.wald object, a list with entries:

b

The point estimate of the coefficients on the dependent variables

S

The Wald statistic

s.boot

The bootstrapped Wald statistic

p.value

The p-value for the Wald statistic.

Examples

data("pgsc.dta")
g.i <- function(b) b[1] ; g.i.grad <- function(b) c(1,0)
sol.r <- pgsc(pgsc.dta, dep.var = 'y', indep.var = c('D1','D2'), 
b.init = c(0,1), method='onestep', g.i=g.i, g.i.grad=g.i.grad )
wald <- pgsc.wald.test( pgsc.dta, 'y', indep.var = c('D1','D2'), sol.r )
summary(wald)
plot(wald)

---