R: Plot Logit or Probit Fit Summary

plot.logit.spike.fit.summary {BoomSpikeSlab}

R Documentation

Plot Logit or Probit Fit Summary

Description

Two plots can be accessed by this function. The first is a time series plot of the "deviance R-square" statistic, by MCMC iteration. The second is a Hosmer-Lemeshow plot in which the data is divided into 10 groups based on predicted probabilities, and the empirical success probabilities for that group are plotted against the expected probabilities from the model.

Usage

PlotLogitSpikeFitSummary(
    model,
    burn = 0,
    which.summary = c("both", "r2", "bucket"),
    scale = c("logit", "probability"),
    cutpoint.basis = c("sample.size", "equal.range"),
    number.of.buckets = 10,
    ...)

PlotProbitSpikeFitSummary(
    model,
    burn = 0,
    which.summary = c("both", "r2", "bucket"),
    scale = c("probit", "probability"),
    cutpoint.basis = c("sample.size", "equal.range"),
    number.of.buckets = 10,
    ...)

Arguments

`model`	A model object inheriting from `logit.spike` or `probit.spike`.
`burn`	The number of MCMC iterations in the object to be discarded as burn-in. Note that this only affects the deviance R-square plot. The fit summaries in the Hosmer-Lemeshow plot are constructed by `logit.spike` or `probit.spike` in order to keep permanent object sizes small.
`which.summary`	Which plot is desired?
`scale`	The scale to use for the predicted probabilities in the Hosmer-Lemeshow plot.
`cutpoint.basis`	How should cutpoints be determined for the Hosmer-Lemeshow plot? If `"sample.size"` then each bucket will have equal sample size. If `"equal.range"` then each bucket will occupy the same size on the chosen (logit/probit or probability) scale.
`number.of.buckets`	The number of buckets to use in the Hosmer-Lemeshow plot.
`...`	Additional arguments to be passed to `barplot.`

Author(s)

Steven L. Scott

Examples

simulate.logit.spike <- function(n = 100, p = 10, ngood = 3,
                              niter=1000){
  x <- cbind(1, matrix(rnorm(n * (p-1)), nrow=n))
  beta <- c(rnorm(ngood), rep(0, p - ngood))
  prob <- plogis(x %*% beta)
  y <- runif(n) < prob
  x <- x[,-1]
  draws <- logit.spike(y ~ x, niter=niter)
  plot.ts(draws$beta)
  return(invisible(draws))
}
model <- simulate.logit.spike()
plot(model, "fit")
plot(model, "fit", scale = "probability", number.of.buckets = 15)