PanelEstimate {PanelMatch}

PanelEstimate

Description

PanelEstimate estimates a causal quantity of interest, including the average treatment effect for treated or control units (att and atc, respectively), the average effect of treatment reversal on reversed units (art), or average treatment effect (ate), as specified in PanelMatch(). This is done by estimating the counterfactual outcomes for each treated unit using matched sets. Users will provide matched sets that were obtained by the PanelMatch function and obtain point estimates via a weighted average computation with weighted bootstrap standard errors. Point estimates and standard errors will be produced for each period in the lead window specified by the lead argument from PanelMatch(). Users may run multiple estimations by providing lists of each argument to the function. However, in this format, every argument must be explicitly specified in each configuration and must adhere to the same data types/structures outlined below.

Usage

PanelEstimate(
  sets,
  data,
  number.iterations = 1000,
  df.adjustment = FALSE,
  confidence.level = 0.95,
  moderator = NULL,
  se.method = "bootstrap",
  pooled = FALSE,
  include.placebo.test = FALSE,
  parallel = FALSE,
  num.cores = 1
)

Arguments

sets

A PanelMatch object attained via the PanelMatch() function.

data

The same time series cross sectional data set provided to the PanelMatch() function used to produce the matched sets.

number.iterations

If using bootstrapping for calculating standard errors, this is the number of bootstrap iterations. Provide as integer. If se.method is not equal to "bootstrap", this argument has no effect.

df.adjustment

A logical value indicating whether or not a degree-of-freedom adjustment should be performed for the standard error calculation. The default is FALSE. This parameter is only available for the bootstrap method of standard error calculation.

confidence.level

A numerical value specifying the confidence level and range of interval estimates for statistical inference. The default is .95.

moderator

The name of a moderating variable, provided as a character string. If a moderating variable is provided,the returned object will be a list of PanelEstimate objects. The names of the list will reflect the different values of the moderating variable. More specifically, the moderating variable values will be converted to syntactically proper names using make.names().

se.method

Method used for calculating standard errors, provided as a character string. Users must choose between "bootstrap", "conditional", and "unconditional" methods. Default is "bootstrap". "bootstrap" uses a block bootstrapping procedure to calculate standard errors. The conditional method calculates the variance of the estimator, assuming independence across units but not across time. The unconditional method also calculates the variance of the estimator analytically, but makes no such assumptions about independence across units. When the quantity of interest is "att", "atc", or "art", all methods are available. Only "bootstrap" is available for the ate. If pooled argument is TRUE, then only bootstrap is available.

pooled

Logical. If TRUE, estimates and standard errors are returned for treatment effects pooled across the entire lead window. Only available for se.method = ``bootstrap''

include.placebo.test

Logical. If TRUE, a placebo test is run and returned in the results. The placebo test uses the same specifications for calculating standard errors as the main results. That is, standard errors are calculated according to the user provided se.method and confidence.level arguments (and, if applicable, parallelization specifications). If these are invalid for some reason, an error will be thrown.

parallel

Logical. If TRUE and se.method = ``bootstrap'', bootstrap procedure will be parallelized. Default is FALSE. If se.method is not set to bootstrap, this option does nothing.

num.cores

Integer. Specifies the number of cores to use for parallelization. If se.method = ``bootstrap'' and parallel = TRUE, then this option will take effect. Otherwise, it will do nothing.

Value

PanelEstimate returns a list of class ‘PanelEstimate’ containing the following components:

estimates

the point estimates of the quantity of interest for the lead periods specified

se.method

The method used to calculate standard errors. This is the same as the argument provided to the function.

bootstrapped.estimates

the bootstrapped point estimate values, when applicable

bootstrap.iterations

the number of iterations used in bootstrapping, when applicable

method

refinement method used to create the matched sets from which the estimates were calculated

lag

See PanelMatch() argument lag for more information.

lead

The lead window sequence for which PanelEstimate() is producing point estimates and standard errors.

confidence.level

the confidence level

qoi

the quantity of interest

matched.sets

the refined matched sets used to produce the estimations

standard.error

the standard error(s) of the point estimates

pooled

Logical indicating whether or not estimates were calculated for individual lead periods or pooled.

placebo.test

if include.placebo.test = TRUE, a placebo test is conducted using placebo_test() and returned as a list. See documentation for placebo_test() for more about each individual item.

Author(s)

In Song Kim <[e-mail]>, Erik Wang <[e-mail]>, Adam Rauh <[e-mail]>, and Kosuke Imai <[e-mail]>

References

Imai, Kosuke, In Song Kim, and Erik Wang (2023)

Examples

dem.sub <- dem[dem[, "wbcode2"] <= 100, ]
# create subset of data for simplicity
PM.results <- PanelMatch(lag = 4, time.id = "year", unit.id = "wbcode2", 
                        treatment = "dem", refinement.method = "ps.match", 
                         data = dem.sub, match.missing = TRUE, covs.formula = ~ tradewb, 
                         size.match = 5, qoi = "att", outcome.var = "y", 
                         lead = 0:4, forbid.treatment.reversal = TRUE)
PE.results <- PanelEstimate(sets = PM.results, data = dem.sub, se.method = "unconditional")