circ.psa {PSAgraphics}R Documentation

Generates a Propensity Score Assessment Plot

Description

Displays a graphic that summarizes outcomes in a propensity score analysis, based on strata that have been defined in the first Phase of a propensity score analysis (PSA). The graphic displays contributions of individual strata to the overall effect, weighing contributions of individual strata according to the relative sizes of the respective strata. The overall effect is plotted as a heavy diagonal line that runs parallel to the identity diagonal.

Usage

circ.psa(response, treatment = NULL, strata = NULL, summary = FALSE, 
statistic = "mean", trim = 0, revc = FALSE, confint = TRUE, sw = 0.4, ne = 0.5, inc = 0.35, 
pw = 0.4, lab = TRUE, labcex = 1, xlab = NULL, ylab = NULL, main = NULL)

Arguments

response Either a numeric vector containing the response of interest in a propensity score analysis, or a three column array containing response, treatment and strata.
treatment Binary variable of same length as response; generally 0 for 'control,' 1 for 'treatment'. A character vector with two labels or factor with two levels are also accepted.
strata Integer variable; a vector of same length as response indicating the derived strata from estimated propensity scores. Generally 5 or 6 strata used, but function is effective for more strata.
summary Logical (default FALSE). If TRUE then response must have rows corresponding to number of strata; the first two columns should contain treatment and control group sizes for each stratum, and the pair of columns should contain the appropriate summary statistics for each statum. For example, the four summary columns might have been generated by the strata.summary output of loess.psa.
statistic A scalar summary of the center of the response distribution. Seen next item below. Default = "mean". Note that to generate this statistic the full vector of responses must have been input, not summaries.
trim Allows for a trimmed mean as outcome measure, where trim is from 0 to .5 (.5 implying median).
revc Logical; if TRUE then X and Y axes are interchanged in plot.
confint Logical; if TRUE adds an approximate 95% confidence interval for the mean. The interval may not be realistic it the trim argument exceeds zero.
sw Numerical argument (default = 0.4); extends axes on lower ends, effectively moving circles to lower left.
ne Numerical argument (default = 0.5); extends axes on upper ends, effectively moving circles to upper right.
inc Numerical argument (default = 0.35); controls circle sizes, but relative circle sizes are controlled via pw. In general one wants circle areas to appear subjectively to be sized in accordance with strata sizes.
pw numerical argument (default = 0.4); controls relative circle sizes. pw denotes power or exponent for radius of circle.
lab Logical (default TRUE); labels circles with stratum labels.
labcex numerical argument (default = 1); controls the size of the circle labels.
xlab Label for horizontal axis, by default taken from treatment.
ylab Label for vertical axis, by default taken from treatment.
main Main label for graph.

Details

A circle is plotted for each stratum, centered on the means for the treatment and control groups (for the X and Y axes) respectively. The size of the circle corresponds to the size of the stratum. A diagonal line (lower left to upper right) shows the identity, X=Y, so that circles on, say, the lower side of this line show that the corresponding X mean is larger than the Y mean for that stratum, and vice-versa. Parallel projections are made from the centers of the strata-cum-circles to difference scores that are plotted on a line segment on the lower-left corner of the graphic; the average difference, which corresponds to the direct adjustment estimator (DAE) for the overall treatment effect, is plotted as a heavy dashed line parallel to the identity diagonal. Rug plots are shown on the upper and right margins of the graphic, for the X and Y marginal distributions. A 95% confidence interval for the overall effect is plotted to the left of the distribution of the stratum difference scores, centered on the DAE. Trimmed means can replace the conventional mean for both the DAE and the marginal distributions (however, the confidence interval calculations are likely to become less trustworthy as larger values of the trim argument are used).

Value

Generate a Propensity Assessment Plot, as well as numerical data for

summary.strata An array with rows corresponding to strata and four columns; these show counts for control and treatment groups, as well as (possibly trimmed) mean response values for control and treatment.
(Name1).wtd.Mn Weighted mean of response for (Name1) group. Name taken from treatment.
(Name2).wtd.Mn Weighted mean of response for (Name2) group. Name taken from treatment.
dae Direct adjustment estimator.
se.wtd Weighted standard error for dae
approx.t Ratio of the direct adjusted estimaor and a standard error based on weighting of straum variances.
df Estimate of degree of freedom; response vector length minus twice number of strata.
CI.95 Approximate 95% confidence interval for overall effect size.

Author(s)

James E. Helmreich James.Helmreich@Marist.edu

Robert M. Pruzek RMPruzek@yahoo.com

See Also

loess.psa

Examples

##Random data with effect size 0
response <- rnorm(1000)
treatment <- sample(c(0,1), 1000, replace = TRUE)
strata <- sample(1:6, 1000, replace = TRUE)
circ.psa(response, treatment, strata)

##Random data with effect size -.2
response <- c(rnorm(500, 0, 12), rnorm(500, 6, 12))
treatment <- c(rep(0, 500), rep(1,500))
strata <- sample(1:5, 1000, replace = TRUE)
aaa <- cbind(response, treatment, strata)
circ.psa(aaa)

##Random data with effect size -2
response <- c(rt(100,3) * 2 + 20, rt(100,12) * 2 + 18)
treatment <- rep(c("A","B"), each = 100)
strata <- sample(c("X","Y","Z","U","V"), 200, replace = TRUE)
circ.psa(response, treatment, strata)

##Tree derived strata
library(rpart)
data(lindner)
attach(lindner)
lindner.rpart <- rpart(abcix ~ stent + height + female + diabetic +
     acutemi + ejecfrac + ves1proc, data = lindner, method = "class")
lindner.tree<-factor(lindner.rpart$where, labels = 1:6)
circ.psa(log(cardbill), abcix, lindner.tree)

##Loess derived strata
lindner.ps <- glm(abcix ~ stent + height + female + 
      diabetic + acutemi + ejecfrac + ves1proc, 
      data = lindner, family = binomial)
ps<-lindner.ps$fitted
lindner.loess<-loess.psa(log(cardbill), abcix, ps)
circ.psa(lindner.loess$summary.strata[, 1:4], summary = TRUE, 
      inc = .1, labcex = .7)

[Package PSAgraphics version 1.3 Index]