plot_ndfa function

Plot Log-OR vs. X for Normal Discriminant Function Approach

When p_ndfa is fit with constant_or = FALSE, the log-OR for X depends on the value of X (and covariates, if any). This function plots the log-OR vs. X for one or several sets of covariate values.

plot_ndfa(estimates, varcov = NULL, xrange, xname = "X",
  cvals = NULL, set_labels = NULL, set_panels = TRUE)

Arguments

• estimates: Numeric vector of point estimates for (gamma_0, gamma_y, gamma_c^T, sigsq).
• varcov: Numeric matrix with variance-covariance matrix for estimates. If NULL, 95% confidence bands are omitted.
• xrange: Numeric vector specifying range of X values to plot.
• xname: Character vector specifying name of X variable, for plot title and x-axis label.
• cvals: Numeric vector or list of numeric vectors specifying covariate values to use in log-odds ratio calculations.
• set_labels: Character vector of labels for the sets of covariate values. Only used if cvals is a list.
• set_panels: Logical value for whether to use separate panels for each set of covariate values, as opposed to using different colors on a single plot.

Returns

Plot of log-OR vs. X generated by ggplot.

Examples

# Fit discriminant function model for poolwise X vs. (Y, C), without assuming
# a constant log-OR. Note that data were generated with a constant log-OR of
# 0.5.
data(dat_p_ndfa)
dat <- dat_p_ndfa$dat
fit <- p_ndfa(
  g = dat$g,
  y = dat$numcases,
  xtilde = dat$x,
  c = dat$c,
  errors = "neither",
  constant_or = FALSE
)

# Plot estimated log-OR vs. X, holding C fixed at the sample mean.
p <- plot_ndfa(
  estimates = fit$estimates,
  varcov = fit$theta.var,
  xrange = range(dat$x[dat$g == 1]),
  cvals = mean(dat$c / dat$g)
)
p