exactMatch function

Generate an exact matching set of subproblems.

An exact match is one based on a factor. Within a level, all observations are allowed to be matched. An exact match can be combined with another distance matrix to create a set of matching subproblems. methods

exactMatch(x, ...)

## S4 method for signature 'vector'
exactMatch(x, treatment)

## S4 method for signature 'formula'
exactMatch(x, data = NULL, subset = NULL, na.action = NULL, ...)

Arguments

• x: A factor vector or a formula, used to select method.

• ...: Additional arguments for methods.

• treatment: A logical or binary vector the same length as x

  indicating treatment and control for each unit in the study. TRUE or 1 represents a treatment unit, FALSE or 0 represents a control unit. NA units are excluded.

• data: A data.frame or matrix that contains the variables used in the formula x.

• subset: an optional vector specifying a subset of observations to be used

• na.action: A function which indicates what should happen when the data contain NAs

Returns

A matrix like object, which is suitable to be given as distance argument to fullmatch or pairmatch. The exact match will be only zeros and Inf values, indicating a possible match or no possible match, respectively. It can be added to a another distance matrix to create a subclassed matching problem.

Details

exactMatch creates a block diagonal matrix of 0s and Infs. The pairs with 0 entries are within the same level of the factor and legitimate matches. Inf indicates units in different levels. exactMatch replaces the structure.fmla argument to several functions in previous versions of optmatch. For the factor method, the two vectors x and treatment must be the same length. The vector x is interpreted as indicating the grouping factors for the data, and the vector treatment indicates whether a unit is in the treatment or control groups. At least one of these two vectors must have names. For the formula method, the data argument may be omitted, in which case the method attempts to find the variables in the environment from which the function was called. This behavior, and the arguments subset

and na.action, mimics the behavior of lm.

Examples

data(nuclearplants)

### First generate a standard propensity score
ppty <- glm(pr~.-(pr+cost), family = binomial(), data = nuclearplants)
ppty.distances <- match_on(ppty)

### Only allow matches within the partial turn key plants
pt.em <- exactMatch(pr ~ pt, data = nuclearplants)
as.matrix(pt.em)

### Blunt matches:
match.pt.em <- fullmatch(pt.em)
print(match.pt.em, grouped = TRUE)

### Combine the propensity scores with the subclasses:
match.ppty.em <- fullmatch(ppty.distances + pt.em)
print(match.ppty.em, grouped = TRUE)

See Also

caliper, antiExactMatch, match_on, fullmatch, pairmatch

Author(s)

Mark M. Fredrickson