optmatch (version 0.10.8)

pairmatch: Optimal 1:1 and 1:k matching

Description

Given a treatment group, a larger control reservoir, and a method for creating discrepancies between each treatment and control unit (or optionally an already created such discrepancy matrix), finds a pairing of treatment units to controls that minimizes the sum of discrepancies.

Usage

pairmatch(x, controls = 1, data = NULL, remove.unmatchables = FALSE, ...)
pair(x, controls = 1, data = NULL, remove.unmatchables = FALSE, ...)

Value

A optmatch object (factor) indicating matched groups.

Arguments

x

Any valid input to match_on. If x is a numeric vector, there must also be passed a vector z indicating grouping. Both vectors must be named.

Alternatively, a precomputed distance may be entered.

controls

The number of controls to be matched to each treatment

data

Optional data set.

remove.unmatchables

Should treatment group members for which there are no eligible controls be removed prior to matching?

...

Additional arguments to pass to match_on (e.g. within)) or to fullmatch (e.g. tol). It is an error to pass min.controls, max.controls, mean.controls or omit.fraction as pairmatch must set these values.

Details

This is a wrapper to fullmatch; see its documentation for more information, especially on additional arguments to pass, additional discussion of valid input for parameter x, and feasibility recovery.

If remove.unmatchables is FALSE, then if there are unmatchable treated units then the matching as a whole will fail and no units will be matched. If TRUE, then this unit will be removed and the function will attempt to match each of the other treatment units. As of version 0.9-8, if there are fewer matchable treated units than matchable controls then pairmatch will attempt to place each into a matched pair each of the matchable controls and a strict subset of the matchable treated units. (Previously matching would have failed for subclasses of this structure.)

Matching can still fail, even with remove.unmatchables set to TRUE, if there is too much competition for certain controls; if you find yourself in that situation you should consider full matching, which necessarily finds a match for everyone with an eligible match somewhere.

The units of the optmatch object returned correspond to members of the treatment and control groups in reference to which the matching problem was posed, and are named accordingly; the names are taken from the row and column names of distance (with possible additions from the optional data argument). Each element of the vector is the concatenation of: (i) a character abbreviation of subclass.indices, if that argument was given, or the string 'm' if it was not; (ii) the string .; and (iii) a non-negative integer. Unmatched units have NA entries. Secondarily, fullmatch returns various data about the matching process and its result, stored as attributes of the named vector which is its primary output. In particular, the exceedances attribute gives upper bounds, not necessarily sharp, for the amount by which the sum of distances between matched units in the result of fullmatch exceeds the least possible sum of distances between matched units in a feasible solution to the matching problem given to fullmatch. (Such a bound is also printed by print.optmatch and by summary.optmatch.)

References

Hansen, B.B. and Klopfer, S.O. (2006), ‘Optimal full matching and related designs via network flows’, Journal of Computational and Graphical Statistics, 15, 609--627.

See Also

matched, caliper, fullmatch

Examples

data(nuclearplants)

### Pair matching on a Mahalanobis distance
( pm1 <- pairmatch(pr ~ t1 + t2, data = nuclearplants) )
summary(pm1)

### Pair matching within a propensity score caliper.
ppty <- glm(pr ~ . - (pr + cost), family = binomial(), data = nuclearplants)
### For more complicated models, create a distance matrix and pass it to fullmatch.
mhd <- match_on(pr ~ t1 + t2, data = nuclearplants) + caliper(match_on(ppty), 2)
( pm2 <- pairmatch(mhd, data = nuclearplants) )
summary(pm2)

### Propensity balance assessment. Requires RItools package.
if(require(RItools)) summary(pm2, ppty)

### 1:2 matched triples
( tm <- pairmatch(pr ~ t1 + t2, controls = 2, data = nuclearplants) )
summary(tm)

### Creating a data frame with the matched sets attached.
### match_on(), caliper() and the like cooperate with pairmatch()
### to make sure observations are in the proper order:
all.equal(names(tm), row.names(nuclearplants))
### So our data frame including the matched sets is just
cbind(nuclearplants, matches=tm)

### In contrast, if your matching distance is an ordinary matrix
### (as earlier versions of optmatch required), you'll
### have to align it by observation name with your data set.
cbind(nuclearplants, matches = tm[row.names(nuclearplants)])


### Match in subgroups only. There are a few ways to specify this.
m1 <- pairmatch(pr ~ t1 + t2, data=nuclearplants,
                within=exactMatch(pr ~ pt, data=nuclearplants))
m2 <- pairmatch(pr ~ t1 + t2 + strata(pt), data=nuclearplants)
### Matching on propensity scores within matching in subgroups only:
m3 <- pairmatch(glm(pr ~ t1 + t2, data=nuclearplants, family=binomial),
                data=nuclearplants,
                within=exactMatch(pr ~ pt, data=nuclearplants))
m4 <- pairmatch(glm(pr ~ t1 + t2 + pt, data=nuclearplants,
                    family=binomial),
                data=nuclearplants,
                within=exactMatch(pr ~ pt, data=nuclearplants))
m5 <- pairmatch(glm(pr ~ t1 + t2 + strata(pt), data=nuclearplants,
                    family=binomial), data=nuclearplants)
# Including `strata(foo)` inside a glm uses `foo` in the model as
# well, so here m4 and m5 are equivalent. m3 differs in that it does
# not include `pt` in the glm.