Extracts weights from the outputs of preprocessing functions.
get.w(x, ...) # S3 method for matchit get.w(x, ...) # S3 method for ps get.w(x, stop.method = NULL, estimand, s.weights = FALSE, ...) # S3 method for mnps get.w(x, stop.method = NULL, s.weights = FALSE, ...) # S3 method for iptw get.w(x, stop.method = NULL, s.weights = FALSE, ...) # S3 method for ps.cont get.w(x, s.weights = FALSE, ...) # S3 method for Match get.w(x, ...) # S3 method for CBPS get.w(x, estimand, ...) # S3 method for ebalance get.w(x, treat, ...) # S3 method for optmatch get.w(x, estimand, ...) # S3 method for cem.match get.w(x, estimand, ...) # S3 method for weightit get.w(x, s.weights = FALSE, ...) # S3 method for mimids get.w(x, ...) # S3 method for wimids get.w(x, ...) # S3 method for designmatch get.w(x, treat, estimand, ...) # S3 method for sbwcau get.w(x, ...)
output from the corresponding preprocessing packages.
the name of the stop method used in the original call to
mnps() in twang, e.g.,
"es.mean". If empty, will return weights from all stop method available into a data.frame. Abbreviations allowed.
if weights are computed using the propensity score (i.e., for the
CBPS methods), which estimand to use to compute the weights. If
"ATE", weights will be computed as
1/ps for the treated group and
1/(1-ps) for the control group. If
"ATT", weights will be computed as
1 for the treated group and
ps/(1-ps) for the control group. If not specified,
get.w() will try to figure out which estimand is desired based on the object.
If weights are computed using subclasses/matching strata (i.e., for the
designmatch methods), which estimand to use to compute the weights. First, a subclass propensity score is computed as the proportion of treated units in each subclass, and the one of the formulas above will be used based on the estimand requested. If not specified,
"ATT" is assumed.
a vector of treatment status for each unit. This is required for methods that include
treat as an argument. The treatment variable that was used in the original preprocessing function call should be used.
whether the sampling weights included in the original call to the fitting function should be included in the weights. If
TRUE, the returned weights will be the product of the balancing weights estimated by the fitting function and the sampling weights. If
FALSE, only the balancing weights will be returned.
further arguments passed to or from other methods.
The output of
get.w() can be used in calls to the formula and data frame methods of
bal.tab() (see example below). In this way, the output of multiple preprocessing packages can be viewed simultaneously and compared. The weights can also be used in
weights statements in regression methods to compute weighted effects.
twang has a function called
get.weights() that performs the same function on
ps objects but offers slightly finer control. Note that the weights generated by
ps objects do not include sampling weights by default.
When sampling weights are used with
CBPS() in CBPS, the returned weights will already have the sampling weights incorporated. To retrieve the balancing weights on their own, divide the returned weights by the original sampling weights. For other packages, the balancing weights are returned separately unless
s.weights = TRUE, which means they must be multiplied by the sampling weights for effect estimation.
Match() in Matching is used with
CommonSupport = TRUE, the returned weights will be incorrect. This option is not recommended by the package authors.
A vector or data frame of weights for each unit. These may be matching weights or balancing weights.
twang::get.weights() in twang.
data("lalonde", package = "cobalt") library("MatchIt"); library("WeightIt") m.out <- matchit(treat ~ age + educ + race, data = lalonde) w.out <- weightit(treat ~ age + educ + race, data = lalonde, estimand = "ATT") bal.tab(treat ~ age + educ + race, data = lalonde, weights = data.frame(matched = get.w(m.out), weighted = get.w(w.out)), method = c("matching", "weighting"), estimand = "ATT") #> Balance Measures #> Type Diff.matched Diff.weighted #> age Contin. -0.0280 0.1078 #> educ Contin. 0.0161 -0.0633 #> race_black Binary 0.3730 -0.0020 #> race_hispan Binary -0.2703 0.0009 #> race_white Binary -0.1027 0.0011 #> #> Effective sample sizes #> Control Treated #> All 429. 185 #> matched 185. 185 #> weighted 116.94 185