Type: Package
Title: Propensity Score Matching of Non-Binary Treatments
Version: 1.0.0
Date: 2025-04-03
Description: Propensity score matching for non-binary treatments.
License: GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]
URL: https://jbryer.github.io/TriMatch/, https://github.com/jbryer/TriMatch/
BugReports: https://github.com/jbryer/TriMatch/issues/
Depends: ez, ggplot2, R (≥ 3.0), reshape2, scales
Imports: compiler, grid, gridExtra, PSAgraphics, psych, randomForest, stats
Suggests: bookdown, knitr, MASS, rmarkdown, xtable
VignetteBuilder: knitr
Encoding: UTF-8
RoxygenNote: 7.3.2
NeedsCompilation: no
Packaged: 2025-04-03 20:06:56 UTC; jbryer
Author: Jason Bryer ORCID iD [aut, cre]
Maintainer: Jason Bryer <jason@bryer.org>
Repository: CRAN
Date/Publication: 2025-04-03 20:40:06 UTC

Propensity Score Analysis for Non-Binary Treatments

Description

This packages provides functions to estimate and visualize propensity score analyses including matching for non-binary treatments.

Author(s)

Jason Bryer jason@bryer.org

See Also

PSAgraphics multilevelPSA

This method will use a M1-to-M2-to-1 matching.

Description

In this method, M2 corresponds to the number of times a treat1 unit can be matched with a treat2 unit. The M1 parameter corresponds to the number of times a treat1 unit can be used in total.

Usage

OneToN(tmatch, M1 = 2, M2 = 1, ...)

Arguments

tmatch

initial results from [trimatch()] that contains all possible matches within the specified caliper.

M1

a scaler indicating the number of unique subjects in group one to retain. This applies only to the first group in the matching order.

M2

a scaler indicating the number of unique matches to retain. This applies to the first two groups in the matching order.

...

currently unused.

Convert a list of vectors to a data frame.

Description

This function will convert a list of vectors to a data frame. This function will handle three different types of lists of vectors. First, if all the elements in the list are named vectors, the resulting data frame will have have a number of columns equal to the number of unique names across all vectors. In cases where some vectors do not have names in other vectors, those values will be filled with NA.

Usage

## S3 method for class 'list'
as.data.frame(x, row.names = NULL, optional = FALSE, ...)

Arguments

x

a list to convert to a data frame.

row.names

a vector equal to length(x) corresponding to the row names. If NULL, the row names will be set to names(x).

optional

not used.

...

other parameters passed to [base::data.frame()].

Details

The second case is when all the vectors are of the same length. In this case, the resulting data frame is equivalent to applying rbind across all elements.

The third case handled is when there are varying vector lengths and not all the vectors are named. This condition should be avoided. However, the function will attempt to convert this list to a data frame. The resulting data frame will have a number of columns equal to the length of the longest vector. For vectors with length less than this will fill the row with NAs. Note that this function will print a warning if this condition occurs.

Value

a data frame.

Author(s)

Jason Bryer jason@bryer.org

References

http://stackoverflow.com/questions/4227223/r-list-to-data-frame

Examples

    test1 <- list( c(a='a',b='b',c='c'), c(a='d',b='e',c='f'))
    as.data.frame(test1)
    
    test2 <- list( c('a','b','c'), c(a='d',b='e',c='f'))
    as.data.frame(test2)
    
    test3 <- list('Row1'=c(a='a',b='b',c='c'), 'Row2'=c(var1='d',var2='e',var3='f'))
    as.data.frame(test3)
    
    ## Not run: 
    #This will print a warning.
    test4 <- list('Row1'=letters[1:5], 'Row2'=letters[1:7], 'Row3'=letters[8:14])
    as.data.frame(test4)
    
## End(Not run)
    
    test5 <- list(letters[1:10], letters[11:20])
    as.data.frame(test5)
    
    ## Not run: 
    #This will throw an error.
    test6 <- list(list(letters), letters)
    as.data.frame(test6)
    
## End(Not run)

Balance plot for the given covariate.

Description

If the covariate is numeric, boxplots will be drawn with red points for the mean and green error bars for the standard error. For non-numeric covariates a barplot will be drawn.

Usage

balance.plot(
  x,
  covar,
  model,
  nstrata = attr(attr(tmatch, "triangle.psa"), "nstrata"),
  label = "Covariate",
  ylab = "",
  xlab = NULL,
  se.ratio = 2,
  print = TRUE,
  legend.position = "top",
  x.axis.labels,
  x.axis.angle = -45,
  ...
)

Arguments

x

results from [trimatch()].

covar

vector of the covariate to check balance of.

model

an integer between 1 and 3 indicating from which model the propensity scores will be used.

nstrata

number of strata to use.

label

label for the legend.

ylab

label of the y-axis.

xlab

label of the x-axis.

se.ratio

a multiplier for how large standard error bars will be.

print

print the output if the Freidman Rank Sum Test and repeated measures ANOVA (for continuous variables).

legend.position

the position of the legend.

x.axis.labels

labels for the x-axis.

x.axis.angle

angle for x-axis labels.

...

parameters passed to [plot.balance.plots()].

Details

A Friedman rank sum test will be performed for all covariate types, printed, and stored as an attribute to the returned object named friedman. If a continuous covariate a repeated measures ANOVA will also be performed, printed, and returned as an attribute named rmanova.

Value

a ggplot2 figure or a list of ggplot2 figures if covar is a data frame.

Returns a ggplot2 box plot of the differences.

Description

A boxplot of differences between each pair of treatments.

Usage

boxdiff.plot(
  tmatch,
  out,
  plot.mean = TRUE,
  ordering = attr(tmatch, "match.order"),
  ci.width = 0.5
)

Arguments

tmatch

the results from [trimatch()].

out

a vector of the outcome measure of interest.

plot.mean

logical indicating whether the means should be plotted.

ordering

specify the order for doing the paired analysis, that is analysis will be conducted as: ordering[1] - ordering[2], ordering[1] - ordering[3], and ordering[2] - ordering[3].

ci.width

the width for the confidence intervals.

Value

a ggplot2 boxplot of the differences.

Calculate covariate effect size differences before and after stratification.

Description

This function is modified from the [PSAgraphics::cv.bal.psa()] function in the 'PSAgrpahics' package.

Usage

covariateBalance(
  covariates,
  treatment,
  propensity,
  strata = NULL,
  int = NULL,
  tree = FALSE,
  minsize = 2,
  universal.psd = TRUE,
  trM = 0,
  absolute.es = TRUE,
  trt.value = NULL,
  use.trt.var = FALSE,
  verbose = FALSE,
  xlim = NULL,
  plot.strata = TRUE,
  ...
)

Arguments

covariates

dataframe of interest

treatment

binary vector of 0s and 1s (necessarily? what if character, or 1, 2?)

propensity

PS scores from some method or other.

strata

either a vector of strata number for each row of covariate, or one number n in which case it is attempted to group rows by ps scores into n strata of size approximately 1/n. This does not seem to work well in the case of few specific propensity values, as from a tree.

int

either a number m used to divide [0,1] into m equal length subintervals, or a vector of cut points between 0 an 1 defining the subintervals (perhaps as suggested by loess.psa). In either case these subintervals define strata, so strata can be of any size.

tree

logical, if unique ps scores are few, as from a recursively partitioned tree, then TRUE will force each ps value to define a stratum.

minsize

smallest allowable stratum-treatment size. If violated, strata is removed.

universal.psd

If 'TRUE', forces standard deviations used to be unadjusted for stratification.

trM

trimming proportion for mean calculations.

absolute.es

logical, if 'TRUE' routine uses absolute values of all effect sizes.

trt.value

allows user to specify which value is active treatment, if desired.

use.trt.var

logical, if true then Rubin-Stuart method using only treatment variance with be used in effect size calculations.

verbose

logical, controls output that is visibly returned.

xlim

limits for the x-axis.

plot.strata

logical indicating whether to print strata.

...

currently unused.

Details

Note: effect sizes are calculated as treatment 1 - treatment 0, or treatment B - treatment A.

Author(s)

Robert M. Pruzek RMPruzek@yahoo.com

James E. Helmreich James.Helmreich@Marist.edu

KuangNan Xiong harryxkn@yahoo.com

Convert a list of vectors to a data frame.

Description

This function will convert a list of vectors to a data frame. This function will handle three different types of lists of vectors. First, if all the elements in the list are named vectors, the resulting data frame will have have a number of columns equal to the number of unique names across all vectors. In cases where some vectors do not have names in other vectors, those values will be filled with NA.

Usage

data.frame.to.list(...)

Arguments

...

other parameters passed to [base::data.frame()].

Details

The second case is when all the vectors are of the same length. In this case, the resulting data frame is equivalent to applying rbind across all elements.

The third case handled is when there are varying vector lengths and not all the vectors are named. This condition should be avoided. However, the function will attempt to convert this list to a data frame. The resulting data frame will have a number of columns equal to the length of the longest vector. For vectors with length less than this will fill the row with NAs. Note that this function will print a warning if this condition occurs.

Value

a data frame.

Author(s)

Jason Bryer jason@bryer.org

References

http://stackoverflow.com/questions/4227223/r-list-to-data-frame

Examples

    test1 <- list( c(a='a',b='b',c='c'), c(a='d',b='e',c='f'))
    as.data.frame(test1)
    
    test2 <- list( c('a','b','c'), c(a='d',b='e',c='f'))
    as.data.frame(test2)
    
    test3 <- list('Row1'=c(a='a',b='b',c='c'), 'Row2'=c(var1='d',var2='e',var3='f'))
    as.data.frame(test3)
    
    ## Not run: 
    #This will print a warning.
    test4 <- list('Row1'=letters[1:5], 'Row2'=letters[1:7], 'Row3'=letters[8:14])
    as.data.frame(test4)
    
## End(Not run)
    
    test5 <- list(letters[1:10], letters[11:20])
    as.data.frame(test5)
    
    ## Not run: 
    #This will throw an error.
    test6 <- list(list(letters), letters)
    as.data.frame(test6)
    
## End(Not run)

Euclidean distance calculation.

Description

This method uses a simple Euclidean distance calculation for determining the distances between two matches. That is, |ps1 - ps2|.

Usage

distance.euclid(x, grouping, id, groups, caliper, nmatch = Inf)

Arguments

x

vector of propensity scores.

grouping

vector or factor identifying group membership.

id

vector corresponding to unique identifer for each element in x and grouping.

groups

vector of length two indicating the unique groups to calculate the distance between. The first element will be the rows, the second columns.

caliper

a scaler indicating the caliper to use for matching within each step.

nmatch

number of smallest distances to retain.

Value

a list of length equal to x. Each element of the list is a named numeric vector where the values correspond to the distance and the name to the id.

Barplot for the sum of distances.

Description

Barplot for the sum of distances.

Usage

distances.plot(tmatch, caliper = 0.25, label = FALSE)

Arguments

tmatch

the results of [trimatch()].

caliper

a vector indicating where vertical lines should be drawn as a factor of the standard deviation. Rosenbaum and Rubin (1985) suggested one quarter of one standard deviation.

label

label the bars that exceed the minimum caliper.

See Also

triangle.match

Loess plot for matched triplets.

Description

This function will create a ggplot2 figure with propensity scores on the x-axis and the outcome on the y-axis. Three Loess regression lines will be plotted based upon the propensity scores from model. Since each model produces propensity scores for two of the three groups, the propensity score for the third group in each matched triplet will be the mean of the other two. If model is not specified, the default will be to use the model that estimates the propensity scores for the first two groups in the matching order.

Usage

loess3.plot(
  tmatch,
  outcome,
  model,
  ylab = "Outcome",
  plot.connections = FALSE,
  connections.color = "black",
  connections.alpha = 0.2,
  plot.points = geom_point,
  points.alpha = 0.1,
  points.palette = "Dark2",
  ...
)

Arguments

tmatch

the results of [trimatch()].

outcome

a vector representing the outcomes.

model

an integer between 1 and 3 indicating from which model the propensity scores will be used.

ylab

the label for the y-axis.

plot.connections

boolean indicating whether lines will be drawn connecting each matched triplet.

connections.color

the line color of connections.

connections.alpha

number between 0 and 1 representing the alpha levels for connection lines.

plot.points

a ggplot2 function for plotting points. Usually [ggplot2::geom_point()] or [ggplot2::geom_jitter()]. If 'NULL' no points will be drawn.

points.alpha

number between 0 and 1 representing the alpha level for the points.

points.palette

the color palette to use. See [ggplot2::scale_colour_brewer()] and https://colorbrewer2.org for more information.

...

other parameters passed to [ggplot2::geom_smooth()] and [ggplot2::stat_smooth()].

Value

a 'ggplot2' figure.

This method will return at least one treatment from groups one and two within the caliper.

Description

This method will attempt to return enough rows to use each treatment (the first two groups in the matching order) at least once. Assuming treat1 is the first group in the match order and treat2 the second, all duplicate treat1 rows are removed. Next, all treat2 units not in present in after removing duplicate treat1 units are identified. For each of those treat2 units, the matched triplet with the smallest overall distances where treat2 is one of the mathched units is retained.

Usage

maximumTreat(tmatch, ...)

Arguments

tmatch

initial results from [trimatch()] that contains all possible matches within the specified caliper.

...

currently unused.

Merges outcomes with the matched set.

Description

The y parameter should be a subset of the original data used.

Usage

## S3 method for class 'triangle.matches'
merge(x, y, ...)

Arguments

x

the result of [trimatch()]

y

another data frame or vector to merge with.

...

unused

Value

x with the additional column(s) added.

Merges covariate(s) with the results of [trips()].

Description

The y parameter should be a subset of the original data used.

Usage

## S3 method for class 'triangle.psa'
merge(x, y, ...)

Arguments

x

the result of [trips()]

y

another data frame or vector to merge with.

...

unused

Value

x with the additional column(s) added.

Multiple covariate balance assessment plot.

Description

A graphic based upon [cv.bal.psa()] function in the 'PSAgraphics' package. This graphic plots the effect sizes for multiple covariates before and after propensity score adjustment.

Usage

multibalance.plot(tpsa, tmatch, grid = TRUE, cols)

Arguments

tpsa

results of [trips()].

tmatch

results of [trimatch()].

grid

if TRUE, then a grid of three plots for each model will be displayed.

cols

character vector of covariates (i.e. column names) from the original data to include in the plot. By default all covariates used in the logistic regression model are used.

Value

a ggplot2 figure.

Results from the 1987 National Medical Expenditure Study

Description

This file was originally prepared by Anders Corr (corr@fas.harvard.edu) who reports on December 8, 2007 that the resulting numbers closely match with those reported in the published article. It was later modified by Jason Bryer (jason@bryer.org) to an R data object to be included in this package. See http://imai.princeton.edu/research/pscore.html for more information

Format

a data frame with 9,708 observations of 12 variables.

Author(s)

United States Department of Health and Human Services. Agency for Health Care Policy and Research

Source

http://imai.princeton.edu/research/pscore.html

References

National Center For Health Services Research, 1987. National Medical Expenditure Survey. Methods II. Questionnaires and data collection methods for the household survey and the Survey of American Indians and Alaska Natives. National Center for Health Services Research and Health Technology Assessment.

Imai, K., & van Dyk, D.A. (2004). Causal Inference With General Treatment Regimes: Generalizing the Propensity Score, Journal of the American Statistical Association, 99(467), pp. 854-866.

Elizabeth Johnson, E., Dominici, F., Griswold, M., & Zeger, S.L. (2003). Disease cases and their medical costs attributable to smoking: An analysis of the national medical expenditure survey. Journal of Econometrics, 112.

Parallel coordinate plot for the three groups and dependent variable.

Description

Creates a ggplot2 figure of a parallel coordinate plot.

Usage

parallel.plot(tmatch, outcome)

Arguments

tmatch

results from [trimatch()].

outcome

vector of the outcome

Internal method for plotting. Finds a point d distance from x, y

Description

Internal method for plotting. Finds a point d distance from x, y

Usage

perpPt(x, y, d = 0.05)

Arguments

x

x coordinate

y

y coordinate

d

the distance

Prints a grid of balance plots.

Description

Prints a grid of balance plots.

Usage

## S3 method for class 'balance.plots'
plot(x, rows, cols, byrow = TRUE, plot.sequence = seq_along(bplots), ...)

Arguments

x

the results of [balance.plot()] when a data frame is specified.

rows

if covar is a data frame of covariates, the number of rows in the grid of figures.

cols

if covar is a data frame of covariates, the number of columns in the grid of figures.

byrow

if TRUE (default), plots will be drawn by rows, otherwise by columns.

plot.sequence

the sequence (or subset) of plots to draw.

...

currently unused.

Triangle plot drawing matched triplets.

Description

This plot function adds a layer to [plot.triangle.psa()] drawing matched triplets. If 'p' is supplied, this function will simply draw on top of the pre-existing plot, otherwise [plot.triangle.psa()] will be called first.

Usage

## S3 method for class 'triangle.matches'
plot(
  x,
  sample = 0.05,
  rows = sample(nrow(tmatch), nrow(tmatch) * sample),
  line.color = "black",
  line.alpha = 0.5,
  point.color = "black",
  point.size = 3,
  p,
  ...
)

Arguments

x

matched triplets from [triangle.match()].

sample

an number between 0 and 1 representing the percentage of matched triplets to draw.

rows

an integer vector corresponding to the rows in 'tmatch' to draw.

line.color

the line color.

line.alpha

the alpha for the lines.

point.color

color of matched triplet points.

point.size

point size for matched triplets.

p

a ggplot to add the match lines. If NULL, then [plot.triangle.psa()].

...

other parameters passed to [plot.triangle.psa()].

Details

If this function calls [plot.triangle.psa()], it will only draw line segments and points for those data rows that were used in the matching procedure. That is, data elements not matched will be excluded from the figure. To plot all segments and points regardless if used in matching, set 'p = plot(tpsa)'.

Value

a 'ggplot2' graphic.

See Also

plot.triangle.psa

triangle.match

Triangle plot.

Description

Triangle plot showing the fitted values (propensity scores) for three different models.

Usage

## S3 method for class 'triangle.psa'
plot(
  x,
  point.alpha = 0.3,
  point.size = 1.5,
  legend.title = "Treatment",
  text.size = 4,
  draw.edges = FALSE,
  draw.segments = TRUE,
  edge.alpha = 0.2,
  edge.color = "grey",
  edge.labels = c("Model 1", "Model 2", "Model 3"),
  sample = c(1),
  ...
)

Arguments

x

the results from [trips()].

point.alpha

alpha level for points.

point.size

point size.

legend.title

title for the legend.

text.size

text size.

draw.edges

draw edges of the triangle.

draw.segments

draw segments connecting points across two models.

edge.alpha

alpha level for edges if drawn.

edge.color

the color for edges if drawn.

edge.labels

the labels to use for each edge of the triangle.

sample

a vector of length 1 or 3 representing the sample of points to plot. The position of each element corresponds to the groups as returned by attr(tpsa,'groups'). If equal to one, all points will be plotted. Values less than one will plot a percentage of points. Values greater than one exactly that number of points will be plotted.

...

currently unused.

Value

ggplot2 figure

See Also

triangle.psa

Print the results of [balance.plot()] for a data frame of covariates.

Description

Print the results of [balance.plot()] for a data frame of covariates.

Usage

## S3 method for class 'balance.plots'
print(x, ...)

Arguments

x

the results of [balance.plot()] when a data frame is specified.

...

parameters passed to [plot.balance.plots()] and [summary.balance.plots()].

Print method for [plot.triangle.psa()]. The primary purpose is to suppress the "Removed n rows containing missing values" warning printed by 'ggplot2'.

Description

Print method for [plot.triangle.psa()]. The primary purpose is to suppress the "Removed n rows containing missing values" warning printed by 'ggplot2'.

Usage

## S3 method for class 'triangle.plot'
print(x, ...)

Arguments

x

a plot from [plot.triangle.psa()].

...

other parameters passed to ggplot2.

Prints the results of [summary.triangle.matches()].

Description

This is an S3 generic function to print the results of [summary.triangle.matches()].

Usage

## S3 method for class 'trimatch.summary'
print(x, ...)

Arguments

x

results of [summary.triangle.matches()].

...

multiple results of [summary.triangle.matches()]. These must be named. For example, "Method 1" = summary(tmath, outcome).

Internal method for plotting. Position along the left side segment

Description

Internal method for plotting. Position along the left side segment

Usage

segment1(d)

Arguments

d

the distance

Internal method for plotting. Position along the right side segment

Description

Internal method for plotting. Position along the right side segment

Usage

segment2(d)

Arguments

d

the distance

Returns significance level.

Description

Returns the significance level as stars, or NA if a non-numeric value is passed in.

Usage

star(x)

Arguments

x

p-value.

Prints a summary table of the test statistics of each balance plot.

Description

The [balance.plot()] function will create a grid of balance plots if a data frame is provided. The returned object is a list of ggplot2 figures with the statistical tests (i.e. Friedmen Rank Sum tests and if a continuous variable, repeated measures ANOVA as well) saved as attributes. This function will return a data frame combining all of those results.

Usage

## S3 method for class 'balance.plots'
summary(object, ...)

Arguments

object

the results of [balance.plot()] when a data frame is specified.

...

currently unused.

Value

a data frame

Provides a summary of the matched triplets including analysis of outcome measure if provided.

Description

If an outcome measure is provided this function will perform a Freidman Rank Sum Test and repeated measures ANOVA. If either test has a statistically significant difference (as determined by the value of the p parameter), a Pairwise Wilcoxon Rank Sum Test will also be provided.

Usage

## S3 method for class 'triangle.matches'
summary(object, outcome, p = 0.05, ordering = attr(object, "match.order"), ...)

Arguments

object

result of [trimatch()].

outcome

vector representing the outcome measure.

p

threshold of the p value to perform a

ordering

specify the order for doing the paired analysis, that is analysis will be conducted as: ordering[1] - ordering[2], ordering[1] - ordering[3], and ordering[2] - ordering[3].

...

parameters passed to other statistical tests.

Value

a trimatch.summary object.

See Also

[stats::friedman.test()], [ez::ezANOVA()], [stats::pairwise.wilcox.test()]

Prints the summary results of the logistic regression models.

Description

The [trips()] function estimates three separate logistic regression models for each pair of groups. This function will print a combined table of the three summaries.

Usage

## S3 method for class 'triangle.psa'
summary(object, ...)

Arguments

object

the results of [trips()].

...

currently unused.

Provides a summary of unmatched subjects.

Description

Will return as a list and print the percentage of total unmatched rows and percent by treatment.

Usage

## S3 method for class 'unmatched'
summary(object, digits = 3, ...)

Arguments

object

results of [unmatched()].

digits

number of digits to print.

...

currently unused.

Value

a list of summary results.

Creates matched triplets.

Description

Create matched triplets by minimizing the total distance between matched triplets within a specified caliper.

Usage

trimatch(
  tpsa,
  caliper = 0.25,
  nmatch = c(15),
  match.order,
  exact,
  method = maximumTreat,
  ...
)

Arguments

tpsa

the results from [trips()]

caliper

a vector of length one or three indicating the caliper to use for matching within each step. This is expressed in standardized units such that .25 means that matches must be within .25 of one standard deviation to be kept, otherwise the match is dropped.

nmatch

number of closest matches to retain before moving to next edge. This can be Inf in which case all matches within the caliper will be retained through to the next step. For large datasets, evaluating all possible matches within the caliper could be time consuming.

match.order

character vector of length three indicating the order in which the matching algorithm will processes. The default is to use start with the group the middle number of subjects, followed by the smallest, and then the largest.

exact

a vector or data frame of representing covariates for exact matching. That is, matched triplets will first be matched exactly on these covariates before evaluating distances.

method

This is a function that specifies which matched triplets will be retained. If 'NULL', all matched triplets within the specified caliper will be returned (equivalent to caliper matching in two group matching). The default is [maximumTreat()] that attempts include each treatment at least once. Another option is [OneToN()] which mimicks the one-to-n matching where treatments are matched to multiple control units.

...

other parameters passed to 'method'.

Details

The [trips()] function will estimate the propensity scores for three models. This method will then find the best matched triplets based upon minimizing the summed differences between propensity scores across the three models. That is, the algorithm works as follows:

Examples

## Not run: 
data(turoing)
formu <- ~ Gender + Ethnicity + Military + ESL + EdMother + EdFather + Age +
     Employment + Income + Transfer + GPA
tpsa <- trips(tutoring, tutoring$treat, formu)
tmatch <- trimatch(tpsa, status=FALSE)

## End(Not run)

Recursive function to find possible matched triplets using the apply functions.

Description

Internal method. This version does not use the exact matching. Instead, this function should be called separately for each grouping.

Usage

trimatch.apply2(tpsa, caliper, nmatch, match.order, sd1, sd2, sd3)

Arguments

tpsa

the results from [trips()]

caliper

a vector of length one or three indicating the caliper to use for matching within each step. This is expressed in standardized units such that .25 means that matches must be within .25 of one standard deviation to be kept, otherwise the match is dropped.

nmatch

number of closest matches to retain before moving to next edge. This can be Inf in which case all matches within the caliper will be retained through to the next step. For large datasets, evaluating all possible matches within the caliper could be time consuming.

match.order

character vector of length three indicating the order in which the matching algorithm will processes. The default is to use start with the group the middle number of subjects, followed by the smallest, and then the largest.

sd1

standard deviation for propensity scores from model 1.

sd2

standard deviation for propensity scores from model 2.

sd3

standard deviation for propensity scores from model 3.

Estimates propensity scores for three groups

Description

The propensity score is

e(X)=P({ W }=1|X)

This function will estimate the propensity scores for each pair of groups (e.g. two treatments and one control).

Usage

trips(
  thedata,
  treat,
  formu = ~.,
  groups = unique(treat),
  nstrata = 5,
  method = "logistic",
  ...
)

Arguments

thedata

the data frame.

treat

vector or factor indicating the treatment/control assignment for thedata. Length must be equal to nrow(thedata).

formu

the logistic regression formula. Note that the dependent variable should not be specified and will be modified.

groups

a vector of exactly length three corresponding the values in treat for each control/treatment.

nstrata

the number of strata marks to plot on the edge.

method

the method to use to estimate the propensity scores. Current options are logistic or randomForest.

...

other parameters passed to [stats::glm()].

Details

{ PS }_{ 1 }=e({ X }_{ { T }_{ 1 }C })=Pr(z=1|{ X }_{ { T }_{ 1 }C })

{ PS }_{ 2 }=e({ X }_{ { T }_{ 2 }C })=Pr(z=1|{ X }_{ { T }_{ 2 }C })

{ PS }_{ 3 }=e({ X }_{ { T }_{ 2 }{ T }_{ 1 } })=Pr(z=1|{ X }_{ { T }_{ 2 }{ T }_{ 1 } })

Examples

## Not run: 
data(tutoring)
formu <- ~ Gender + Ethnicity + Military + ESL + EdMother + EdFather + Age +
     Employment + Income + Transfer + GPA
tpsa <- trips(tutoring, tutoring$treat, formu)
head(tpsa)

## End(Not run)

Results from a study examining the effects of tutoring services on course grades.

Description

Format

a data frame with 17 variables.

Returns rows from [trips()] that were not matched by [trimatch()].

Description

This function returns a subset of [trips()] that were not matched by [trimatch()]. All data frame methods work with the returned object but special 'summary' function will provided relevant information.

Usage

unmatched(tmatch)

Arguments

tmatch

the results of [trimatch()].

Value

a data frame of unmatched rows.