causalSLSE (version 0.4-1)

plot: Plot of Predicted Outcome

Description

For objects of class slseFit, the method plots the predicted outcome with respect to a given covariate. It is the same for objects of class cslseFit, but the predicted outcome is plotted for each treatment group separately.

Usage

# S3 method for cslseFit
plot(x, y, which = y, interval = c("none", "confidence"), 
                        level = 0.95, fixedCov = list(),
                        vcov. = vcovHC,  add = FALSE, addToLegend = NULL,
                        addPoints = FALSE, FUN = mean,
                        plot=TRUE, graphPar=list(), ...)
# S3 method for slseFit
plot(x, y, which = y, interval = c("none", "confidence"), 
                       level = 0.95, fixedCov = NULL,
                       vcov. = vcovHC,  add = FALSE, 
                       addPoints = FALSE, FUN = mean,
                       plot=TRUE, graphPar=list(), ...)

Value

It returns NULL invisibly if the argument plot is TRUE. Otherwise, it returns a data.frame with the necessary variables to manually create the plot. For slseModel

objects, it is a list with the following elements:

Outcome

The outcome variable from the model. The name of this element is the actual name of the outcome variable in the model.

which

The covariate selected by the argument which. The name of this element is the name of the selected covariate.

fit

The predicted outcome from the model fit.

lower, upper

The lower and upper bounds of the confidence interval. It is only available when the argument interval is set to "confidence".

Note that all returned variables are ordered with respect to the selected covariates. See the example below.

For cslseModel, the above list of variables is returned separately for each treatment group.

Arguments

x

Object of class cslseFit or slseFit created by estSLSE.

y

alias for which for compatibility with plot.

which

Which covariate to plot against the outcome variable. It could be an integer or a character.

interval

The type of confidence interval. The default is none.

level

The confidence interval level if included.

fixedCov

List of covariates to fix to specific values for the nontreated and treated groups. By default, covariates not selected by which are set to their group specific sample means.

vcov.

An alternative function to compute the covariance matrix of the least squares estimates. The default is the vcovHC method for lm objects.

add

Should the plot be added to an existing one?

addToLegend

A character string to add to the legend next to treated and control.

addPoints

Should we add the scatterplot of the outcome and covariate on the graph?

FUN

The function to determine the fixed value for the covariates not fixed by fixedCov. The default is mean.

plot

If set to FALSE, a data.frame is returned for each group with the covariate selected by which and the prediction.

graphPar

A list of graphical parameters. See Details.

...

Additional argument to pass to the vcov. function.

Details

The default set of parameters can be obtained by running the command causalSLSE:::.initPar(). It returns a list of four elements: treated for the parameter of the lines or points of the treated, nontreated for the parameters of the nontreated, common for the common parameters not specific to a group like the main title or the axis labels, and legend for the legend parameters. The elements treated and nontreated are lists of two elements: points and lines, which are lists of graphical parameters for the scatterplot (when addPoints is TRUE) and the lines. The argument graphPar can be use to modify existing parameters or to add new ones. It must respect the structure of the initial list. See the example below.

To fix covariates to the same values for both groups, fixedCov is just a named list with values for the covariates associated with the names. To fix the covariates to different values for the treated and nontreated, fixedCov is a named list of at least 1 element with names being either treated, nontreated or both. If only one group is specified, the covariates for the other groups are determined by FUN.

Examples

data(simDat3)

##  For cslse objects

mod <- cslseModel(Y ~ Z | ~ X1 + X2, data = simDat3)
fit <- causalSLSE(mod) 
plot(fit, "X1")

## Let's change the type of points for the treated and lines for the
## nontreated

gpar <- list(treated = list(points = list(pch = 24, col = 5)),
             nontreated = list(lines = list(lty = 5, col =
             "darkgreen")), common = list(xlab = "New X",  main =
             "Plot with changed parameters"))  

plot(fit, "X1", addPoints = TRUE, graphPar = gpar)

## For slseModel objects:

mod2 <- slseModel(Y ~ X1 + X2, data = simDat3)
fit2 <- estSLSE(mod2)
plot(fit2, "X1", interval = "confidence", addPoints = TRUE)

##  The same graph produced manually

p2 <- plot(fit2, "X1", interval = "confidence", plot = FALSE)
plot(p2$X1, p2$Y, pch = 21, main = "Y against X1", xlab = "X1", ylab =
     "Y") 
lines(p2$X1, p2$fit, lwd = 2)
lines(p2$X1, p2$lower, lty = 3, lwd = 2)
lines(p2$X1, p2$upper, lty = 3, lwd = 2)