ps.balance: Propensity Score Population Balancing

Description Usage Arguments Details Value Examples

View source: R/psBalance.R

Description

Calculates the weights and/or perform matching of subject to balance the population

Usage

1
2
3
4
ps.balance(data, covariates, estimand = "ATT", match.subjects = TRUE,
  match.exact = NULL, match.ratio = 1, caliper.sigma = 0,
  use.logit = FALSE, truncate.quantile = 0.95, truncate.method = "cap",
  max.matching = 50000)

Arguments

data

Data Frame - containing the dataset with previously calculated PS. The data frame must contain a treatment indicator variable called 'treat' and a propensity score value called 'ps_values'.

covariates

Vector, containing the variable names to be included as potential confounding variables

estimand

String, specifying the desired estimand. Options are "ATT" (default) for the Average Treatment Effect in the Treated, or "ATE" for the Average Treatment Effect.

match.subjects

Boolean, indicating if matching should be used (default TRUE). This is only applicable when the estimand is "ATT" as "ATE" can only be estimated via IPTW. If the estimand is "ATT", SMRW will always be used to generate weights, but if match.subjects is set to TRUE, matches will also be generated.

match.exact

Vector, containing the list of covariate names to perform exact matching on.

match.ratio

Number, indicating the match ratio of control:treat

caliper.sigma

Number, indicating the width of the caliper to use in matching. A value of 0 indicates that calipers will not be used. A non-zero value will turn on calipers

use.logit

Boolean, indicating if the propensity score should be converted to logit before matching. This is generally recommended when using calipers and leads to better balance in most cases.

truncate.quantile

Number, indicating the upper quantile at which to apply weight trimming.

truncate.method

String, indicating the approach to use to trim the dataset. Large weights can adversely affect the ultimate balance of the population. Two approaches appear in the literature, capping weights at a quantile value or dropping subjects from the dataset. The default for this parameter is "cap", which will downwardly adjust any weights larger than the specified quantile to the value at that quantile. Alternatively, set this parameter to "drop" to remove the subjects from the dataset completely. User will be notified of how many subjects are lost in this step.

max.matching

The maximum number of samples that can be used in matching (default 50k)

Details

This function performs propensity score based population balancing. The details of how the population is balanced depend on the parameters specified by the user, including the requested estimand.

Value

psBalanceData - Object containing parameters used in balancing, along with the resulting data frame. The dataframe has additional variable(s) added for the weights and matches. When matching is used, the is_matched is [0,1] indicating if the subject was matched or not.

Examples

1
2
3
4
5
## Not run: 
ps.balance(myData, covariates)
ps.balance(myData, covariates, match.subjects = TRUE, match.exact = c("GENDER"))

## End(Not run)

OHDSI/Centaur documentation built on May 9, 2017, 3:24 p.m.