Encodes calipers, or maximum allowable distances within which to
match. The result of a call to
caliper is itself a distance specification between
treated and control units that can be used with
fullmatch(). Calipers can also be combined with
other distance specifications for richer matching problems.
caliper(x, width, exclude = c(), compare = `<=`, values = FALSE) ## S4 method for signature 'InfinitySparseMatrix' caliper(x, width, exclude = c(), compare = `<=`, values = FALSE) ## S4 method for signature 'matrix' caliper(x, width, exclude = c(), compare = `<=`, values = FALSE) ## S4 method for signature 'optmatch.dlist' caliper(x, width, exclude = c(), compare = `<=`, values = FALSE)
A distance specification created with
The width of the caliper: how wide of a margin to
allow in matches. Be careful in setting the width. Vector valued
arguments will be recycled for each of the finite entries in
(Optional) A character vector of observations (corresponding to row and column names) to exclude from the caliper.
A function that decides that whether two
observations are with the caliper. The default is
Should the returned object be made of all zeros
caliper is a generic function with methods for any of the allowed distance
specifications: user created matrices, the results of
match_on, the results
exactMatch, or combinations (using
`+`) of these objects.
width provides the size of the caliper, the allowable distance for
matching. If the distance between a treated and control pair is less than or
equal to this distance, it is allowed kept; otherwise, the pair is discarded
from future matching. The default comparison of "equal or less than can" be
changed to any other comparison function using the
It is important to understand that
width argument is defined on the
scale of these distances. For univariate distances such as propensity scores,
it is common to specify calipers in standard deviations. If a caliper of
this nature is desired, you must either find the standard deviation directly
or use the
match_on function with its
match_on has access to the underlying univariate scores, for
example for the GLM method, it can determine the caliper width in standard
If you wish to exclude specific units from the caliper requirements, pass the names of
these units in the
exclude argument. These units will be allowed to match any other
A matrix like object that is suitable to be given
distance argument to
pairmatch. The caliper will be only zeros and
indicating a possible match or no possible match, respectively.
You can combine the results of
caliper with other distances using the
`+` operator. See the examples for usage.
Mark M. Fredrickson and Ben B. Hansen
P.~R. Rosenbaum and D.~B. Rubin (1985), ‘Constructing a control group using multivariate matched sampling methods that incorporate the propensity score’, The American Statistician, 39 33–38.
data(nuclearplants) ### Caliper of 100 MWe on plant capacity caliper(match_on(pr~cap, data=nuclearplants, method="euclidean"), width=100) ### Caliper of 1/2 a pooled SD of plant capacity caliper(match_on(pr~cap, data=nuclearplants), width=.5) ### Caliper of .2 pooled SDs in the propensity score ppty <- glm(pr ~ . - (pr + cost), family = binomial(), data = nuclearplants) ppty.dist <- match_on(ppty) pptycaliper <- caliper(ppty.dist, width = .2) ### caliper on the Mahalanobis distance caliper(match_on(pr ~ t1 + t2, data = nuclearplants), width = 3) ### Combining a Mahalanobis distance matching with a caliper ### of 1 pooled SD in the propensity score: mhd.pptyc <- caliper(ppty.dist, width = 1) + match_on(pr ~ t1 + t2, data = nuclearplants) pairmatch(mhd.pptyc, data = nuclearplants) ### Excluding observations from caliper requirements: caliper(match_on(pr ~ t1 + t2, data = nuclearplants), width = 3, exclude = c("A", "f")) ### Returning values directly (equal up to the the attributes) all(abs((caliper(ppty.dist, 1) + ppty.dist) - caliper(ppty.dist, 1, values = TRUE)) < .Machine$Double.eps)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.