weightit.fit: Generate Balancing Weights with Minimal Input Processing

In WeightIt: Weighting for Covariate Balance in Observational Studies

Generate Balancing Weights with Minimal Input Processing

Description

weightit.fit() dispatches one of the weight estimation methods determined by method. It is an internal function called by \funweightit and should probably not be used except in special cases. Unlike weightit(), weightit.fit() does not accept a formula and data frame interface and instead requires the covariates and treatment to be supplied as a numeric matrix and atomic vector, respectively. In this way, weightit.fit() is to weightit() what \funlm.fit is to \funlm - a thinner, slightly faster interface that performs minimal argument checking.

Usage

weightit.fit(covs,
             treat,
             method = "glm",
             s.weights = NULL,
             by.factor = NULL,
             estimand = "ATE",
             focal = NULL,
             stabilize = FALSE,
             ps = NULL,
             moments = NULL,
             int = FALSE,
             subclass = NULL,
             is.MSM.method = FALSE,
             missing = NULL,
             verbose = FALSE,
             include.obj = FALSE,
             ...)

Arguments

covs	a numeric matrix of covariates.
treat	a vector of treatment statuses.
method	a string of length 1 containing the name of the method that will be used to estimate weights. See \funweightit for allowable options. The default is "glm" for propensity score weighting using a generalized linear model to estimate the propensity score.
s.weights	a numeric vector of sampling weights. See the individual pages for each method for information on whether sampling weights can be supplied.
by.factor	a factor variable for which weighting is to be done within levels. Corresponds to the by argument in \funweightit.
estimand	the desired estimand. For binary and multi-category treatments, can be "ATE", "ATT", "ATC", and, for some methods, "ATO", "ATM", or "ATOS". The default for both is "ATE". This argument is ignored for continuous treatments. See the individual pages for each method for more information on which estimands are allowed with each method and what literature to read to interpret these estimands.
stabilize	logical; whether or not to stabilize the weights. For the methods that involve estimating propensity scores, this involves multiplying each unit's weight by the proportion of units in their treatment group. Default is FALSE.
focal	when multi-category treatments are used and ATT weights are requested, which group to consider the "treated" or focal group. This group will not be weighted, and the other groups will be weighted to be more like the focal group. Must be non-NULL if estimand = "ATT" or "ATC".
ps	a vector of propensity scores. If specified, method will be ignored and set to "glm".
moments, int, subclass	arguments to customize the weight estimation. See \funweightit for details.
is.MSM.method	see \funweightitMSM. Typically can be ignored.
missing	character; how missing data should be handled. The options depend on the method used. If NULL, covs covs will be checked for NA values, and if present, missing will be set to "ind". If "", covs covs will not be checked for NA values; this can be faster when it is known there are none.
verbose	whether to print additional information output by the fitting function.
include.obj	whether to include in the output any fit objects created in the process of estimating the weights. For example, with method = "glm", the glm objects containing the propensity score model will be included. See the individual pages for each method for information on what object will be included if TRUE.
...	other arguments for functions called by weightit.fit() that control aspects of fitting that are not covered by the above arguments.

Details

weightit.fit() is called by weightit() after the arguments to weightit() have been checked and processed. weightit.fit() dispatches the function used to actually estimate the weights, passing on the supplied arguments directly. weightit.fit() is not meant to be used by anyone other than experienced users who have a specific use case in mind. The returned object doesn't contain any information about the supplied arguments or details of the estimation method; all that is processed by weightit.

Less argument checking or processing occurs in weightit.fit() than does in weightit(), which means supplying incorrect arguments can result in errors, crashes, and invalid weights, and error and warning messages may not be helpful in diagnosing the problem. weightit.fit() does check to make sure weights were actually estimated, though.

weightit.fit() may be most useful in speeding up simulation simulation studies that use weightit() because the covariates can be supplied as a numeric matrix, which is often how they are generated in simulations, without having to go through the potentially slow process of extracting the covariates and treatment from a formula and data frame. If the user is certain the arguments are valid (e.g., by ensuring the estimated weights are consistent with those estimated from weightit() with the same arguments), less time needs to be spent on processing the arguments. Also, the returned object is much smaller than a weightit object because the covariates are not returned alongside the weights.

Value

A weightit.fit object with the following elements:

weights	The estimated weights, one for each unit.
ps	The estimated or provided propensity scores. Estimated propensity scores are returned for binary treatments and only when method is "glm", "gbm", "cbps", "super", or "bart".
fit.obj	When include.obj = TRUE, the fit object.
info	Additional information about the fitting. See the individual methods pages for what is included.

The weightit.fit object does not have specialized print(), summary(), or plot() methods. It is simply a list containing the above components.

Author(s)

Noah Greifer

See Also

\fun

weightit, which you should use for estimating weights unless you know better.

Examples

library("cobalt")
data("lalonde", package = "cobalt")

#Balancing covariates between treatment groups (binary)
covs_mat <- as.matrix(lalonde[c("age", "educ", "married",
                                "nodegree", "re74", "re75")])
wf1 <- weightit.fit(covs_mat, treat = lalonde$treat,
                    method = "glm", estimand = "ATT")
str(wf1)

WeightIt documentation built on May 31, 2023, 9:25 p.m.