Match persons by propensity score

Use the provided propensity scores to match target to comparator persons.

Usage

matchOnPs(
  population,
  caliper = 0.2,
  caliperScale = "standardized logit",
  maxRatio = 1,
  allowReverseMatch = FALSE,
  stratificationColumns = c()
)

Arguments

population

A data frame with the three columns described below.

caliper

The caliper for matching. A caliper is the distance which is acceptable for any match. Observations which are outside of the caliper are dropped. A caliper of 0 means no caliper is used.

caliperScale

The scale on which the caliper is defined. Three scales are supported: caliperScale = 'propensity score', caliperScale = 'standardized', or caliperScale = 'standardized logit'. On the standardized scale, the caliper is interpreted in standard deviations of the propensity score distribution. 'standardized logit' is similar, except that the propensity score is transformed to the logit scale because the PS is more likely to be normally distributed on that scale (Austin, 2011).

maxRatio

The maximum number of persons in the comparator arm to be matched to each person in the treatment arm. A maxRatio of 0 means no maximum: all comparators will be assigned to a target person.

allowReverseMatch

Allows n-to-1 matching if target arm is larger

stratificationColumns

Names or numbers of one or more columns in the data data.frame on which subjects should be stratified prior to matching. No persons will be matched with persons outside of the strata identified by the values in these columns.

Value

Returns a date frame with the same columns as the input data plus one extra column: stratumId. Any rows that could not be matched are removed

Details

The data frame should have the following three columns:

• rowId (numeric): A unique identifier for each row (e.g. the person ID).

• treatment (integer): Column indicating whether the person is in the target (1) or comparator (0) group.

• propensityScore (numeric): Propensity score.

The default caliper (0.2 on the standardized logit scale) is the one recommended by Austin (2011).

References

Rassen JA, Shelat AA, Myers J, Glynn RJ, Rothman KJ, Schneeweiss S. (2012) One-to-many propensity score matching in cohort studies, Pharmacoepidemiology and Drug Safety, May, 21 Suppl 2:69-80.

Austin, PC. (2011) Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies, Pharmaceutical statistics, March, 10(2):150-161.

Examples

rowId <- 1:5
treatment <- c(1, 0, 1, 0, 1)
propensityScore <- c(0, 0.1, 0.3, 0.4, 1)
age_group <- c(1, 1, 1, 1, 1)
data <- data.frame(
  rowId = rowId,
  treatment = treatment,
  propensityScore = propensityScore,
  age_group = age_group
)
result <- matchOnPs(data, caliper = 0, maxRatio = 1, stratificationColumns = "age_group")
#> Population size after matching is 4