stm (version 1.0.1)

plot.estimateEffect: Plot effect of covariates on topics

Description

Plots the effect of a covariate on a set of topics selected by the user. Different effect types available depending on type of covariate. Before running this, the user should run a function to simulate the necessary confidence intervals. See estimateEffect.

Usage

## S3 method for class 'estimateEffect':
plot(x, covariate, model=NULL, topics=x$topics,
                          method="pointestimate",
                          cov.value1=NULL,
                          cov.value2=NULL,int.value=NULL,
                          npoints=100, nsims=100, ci.level=.95,
                          xlim=NULL, ylim=NULL, ylab=NULL,
                          main="",printlegend=TRUE,
                          labeltype="numbers",n=7,frexw=.5,
                          xlab="", add=FALSE, linecol=NULL,
                          width=25, verbose.labels=TRUE,
                          family=NULL, custom.labels=NULL,...)

Arguments

x

Output of estimateEffect, which calculates simulated betas for plotting

covariate

String of the name of the main covariate of interest. Must be enclosed in quotes. All other covariates within the formula specified in estimateEffect will be kept at their median.

model

Model output, only necessary if labeltype is "prob", "frex", "score", or "lift".

topics

Topics to plot.

method

Method used for plotting. "pointestimate" estimates mean topic proportions for each value of the covariate. "difference" estimates the mean difference in topic proportions for two different values of the covariate (cov.value1 and cov.v

cov.value1

For method "difference", the value or set of values of interest at which to set the covariate. In the case of calculating a treatment/control contrast, set the treatment to cov.value1.

cov.value2

For method "difference", the value or set of values which will be set as the comparison group. cov.value1 and cov.value2 must be vectors of the same length.

int.value

For methods "pointestimate" and "difference", when two terms are interacted and one variable in the interaction is the covariate of interest, the user can specify the value of the interaction term.

npoints

Number of unique points to use for simulation along the support of a continuous covariate. For method "continuous" only.

nsims

Number of simulations for estimation.

n

Number of words to print if "prob", "score", "lift", or "frex" is chosen.

ci.level

Confidence level for confidence intervals.

frexw

If "frex" labeltype is used, this will be the frex weight.

add

Logical parameter for whether the line should be added to the plot, or a new plot should be drawn.

linecol

For continuous covariates only. A vector that specifies the colors of the lines within the plot. If NULL, then colors will be randomly generated.

verbose.labels

For method "difference" -- verboselabels will specify the comparison covariate values of the covariate on the plot.

xlim

Vector of x axis minimum and maximum values.

ylim

Vector of y axis minimum and maximum values.

main

Character string that is plot title.

printlegend

Whether to plot a topic legend in the case of a continuous covariate.

labeltype

Determines the labeltype for the topics. The default is "number" which prints the topic number. Other options are "prob", which prints the highest probability words, "score", "lift", and "frex", from labeltopics (see labeltopics() for mor

xlab

Character string that is x axis title.

ylab

Character string that is y axis title.

width

Character string that is y axis title.

custom.labels

A vector of custom.labels if labeltype is equal to "custom".

family

Font family.

...

Other plotting parameters

Examples

prep <- estimateEffect(1:3 ~ treatment, gadarianFit, gadarian)
plot.estimateEffect(prep, "treatment", model=gadarianFit,
method="pointestimate")
plot.estimateEffect(prep, "treatment", model=gadarianFit,
method="pointestimate")
plot.estimateEffect(prep, "treatment", model=gadarianFit,
method="difference",cov.value1=1,cov.value2=0)

#If the covariate were a binary factor, 
#the factor labels can be used to  
#specify the values of cov.value1 (e.g., cov.value1="treat"). 

# String variables must be turned to factors prior to plotting. 
#If you see this error, Error in rep.int(c(1, numeric(n)), n - 1L) : 
# invalid 'times' value, then you likely have not done this.

#Example of binary times binary interaction
gadarian$binaryvar <- sample(c(0,1), nrow(gadarian), replace=T)
temp <- textProcessor(gadarian$open.ended.response,metadata=gadarian)
out <- prepDocuments(temp$documents, temp$vocab, temp$meta)
stm1 <- stm(out$documents, out$vocab, 3, prevalence=~treatment*binaryvar,
 data=gadarian)
prep <- estimateEffect(c(2) ~ treatment*binaryvar, stmobj=stm1,
metadata=gadarian)

par(mfrow=c(1,2))
plot.estimateEffect(prep, "binaryvar", method="pointestimate",
cov.value1=1, cov.value2=0, xlim=c(-1,1), int.value=1)
plot.estimateEffect(prep, "binaryvar", method="pointestimate",
cov.value1=1, cov.value2=0, xlim=c(-1,1), int.value=0)