pscl (version 0.5)

predict.zeroinfl: Generate predictions from zero-inflated regression count models

Description

Generate predicted counts from zero-inflated regression models for count data, fit with the zeroinfl function, producing objects of class zeroinfl.

Usage

## S3 method for class 'zeroinfl':
predict(object, newdata, se.fit = FALSE, conf = 0.95,
              MC = 1000, type = c("response", "prob"),
              na.action = na.pass, ...)

Arguments

object

an object of class zeroinfl

newdata

optionally, a data frame in which to look for variables with which to predict. If omitted, the fitted linear predictors are used.

se.fit

logical switch indicating if standard errors on predicted values are to be computed

conf

a proportion, width of confidence intervals computed if se.fit=TRUE, defaults to .95

MC

number of Monte Carlo iterates for computing standard errors and confidence intervals around predicted values

type

the type of prediction required. The default is "response", generating predictions on the scale of the observed counts. Choosing type="prob" generates a matrix of predicted probabilities over the range of counts obser

na.action

function determining what should be done with missing values in newdata. The default is to predict NA.

...

further arguments passed to or from other methods.

Value

• a list, with components
• yhat
  a vector of predicted counts
• mu
  predictions from the count part of the model, on the scale of the counts
• phi
  predicted probabilities from the zero-inflated component of the model, the probability of zero count
• se
  if se.fit=T, the standard errors of the predicted counts
• lower,uppper
  if se.fit=T, the lower/upper bounds of confidence intervals of content conf
• prob
  if type="prob", a matrix of predicted probabilities, with each row containing predicted probabilities over the range of counts observed in the data, i.e., ncol(prob) = length(min(y):max(y))

Details

Monte Carlo methods are used to generate standard errors and confidence intervals as follows: MC samples are drawn from a multivariate normal distribution centered on the MLEs of the parameters and variance-covariance matrix equal to the MLE's variance-covariance matrix. With sampled set of parameter values, predicted values are generated. The standard deviation of the predicted values is returned as se.fit along with the lower and upper quantiles corresponding to confidence level conf.

See Also

zeroinfl, predict.glm, predprob.zeroinfl

Examples

data(bioChemists)
zip <- zeroinfl(count=art ~ .,
                x = ~ fem + mar + kid5 + phd + ment,
                z = ~ fem + mar + kid5 + phd + ment,
                data=bioChemists,trace=TRUE)

newdata <- expand.grid(list(fem="Men",
                            mar="Married",
                            kid5=1,
                            phd=3.103,
                            ment=0:77))

yhat <- predict(zip,newdata=newdata,
                se.fit=TRUE,MC=2500)

plot(x=newdata$ment,
     y=yhat$yhat,
     xlab="Mentor Articles",
     ylab="Predicted Counts",
     ylim=range(zip$y),
     type="n")
polygon(x=c(newdata$ment,rev(newdata$ment)),
        y=c(yhat$lower,rev(yhat$upper)),
        border=FALSE,
        col=gray(.75))
lines(x=newdata$ment,
      y=yhat$yhat,
      lwd=2)
rug(quantile(bioChemists$ment,c(.05,.50,.95)))
title("Predicted Counts and 95