Learn R Programming

PSAgraphics (version 2.1.3)

cv.bal.psa: Multiple Covariate Balance Assessment Plot

Description

Provides a graphic that depicts covarite effect size differences between treatment groups both before and after stratification. Function will create stata internally if desired, and returns numerical output used to create graphic.

Usage

cv.bal.psa(
  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,
  ...
)

Value

Graphic plots covariate balance before and after stratication on propensity scores. The default version (absolute.es = TRUE) plots the absolute values of effect sizes for each stratum, though the overall estimate is the weighted mean before taking the absolute values. Numerical output consists of seven addressable objects. If verbose is FALSE (default), output is not printed.

original.strata

Matrix of strata-treatment counts as originally input.

strata.used

Matrix of strata-treatment counts used in effectsize calculations after any minsize reductions.

mean.diff.strata.wtd

Matrix of strata by covariate weighted (by strata size) average differences.

mean.diff.unadj

Matrix of covariate effects sizes before stratification.

effect.sizes

Matrix of effect sizes by covariate and statum.

treatment.levels

Names of treatments.

effects.strata.treatment

Matrix of standard deviations and stratum-treatment covariate means used to calculate the effect.sizes.

Arguments

covariates

Dataframe of covariates. Factors should be recoded using cv.trans.psa

treatment

Binary vector or factor defining the two treatments

propensity

Vector of same length as treatment containing estimated propensity scores.

strata

Either a vector of same length as treatment of predefined stratum number, or one integer n used to assign rows to n strata propensity scores, each of approximately the same number of cases. If relatively few unique propensity scores have been defined (as from a classification tree) then the logical tree should be set equal to TRUE.

int

Either a number m used to divide [0,1] into m equal length subintervals, or a vector containing cut points between 0 and 1 that define subintervals (perhaps as suggested by loess.psa). In either case the subintervals define strata, for which sizes can differ.

tree

Logical, default FALSE. If there are few unique propensity scores, say from a recursively partitioned tree, then TRUE forces strata to be defined by the unique propensity scores.

minsize

Smallest allowable stratum-treatment size. If violated, rows in the stratum are removed. User may wish to redefine strata.

universal.psd

Logical, default = TRUE. Forces standard deviations used to be unadjusted for stratification.

trM

Numeric, default = 0; passed to mean for trimming purposes.

absolute.es

Logical, default TRUE. If TRUE, graphic depicts absolute values of all effect sizes. Note that the adjusted effect size plotted is the absolute value of weighted averages of the signed by-stratum effect size values when absolute.es is TRUE.

trt.value

Character string; if desired allows the name of an active treatment to be given. Should be a level (value) of the treatment factor (vector).

use.trt.var

Logical, default FALSE. If TRUE, uses just active treatment standard deviations for effect size, as per a suggestion of Rubin and Stuart (see reference below).

verbose

Logical, default FALSE. Numerical output is returned invisibly.

xlim

Binary vector passed to plot for overriding default choices. Default NULL.

plot.strata

Logical, default TRUE. Adds effect size values for individual strata to graphic.

...

Other graphical parameters passed to plot.

Author

Robert M. Pruzek RMPruzek@yahoo.com

James E. Helmreich James.Helmreich@Marist.edu

KuangNan Xiong harryxkn@yahoo.com

Details

Effect sizes between treatments for each covariate are presented in one graphic, both before and after stratification.

References

``Matching Methods for Causal Inference: A review and a look forward." Forthcoming in Statistical Science.

See Also

cv.bal.psa, loess.psa, cstrata.psa, cv.trans.psa

Examples

Run this code

data(lindner)
attach(lindner)
lindner.ps <- glm(abcix ~ stent + height + female +
      diabetic + acutemi + ejecfrac + ves1proc,
      data = lindner, family = binomial)
ps<-lindner.ps$fitted
lindner.cv <- lindner[,4:10]
cv.bal.psa(lindner.cv, abcix, ps, strata = 5)
cv.bal.psa(lindner.cv, abcix, ps, strata = 10)
cv.bal.psa(lindner.cv, abcix, ps, int = c(.2, .5, .6, .75, .8))

Run the code above in your browser using DataLab