Skip to contents

Binned calibration plot for categorical DVs

Source: R/categorical.R
catdv_vs_ipred.Rd

A binned alternative to catdv_vs_dvprobs(). The probability column associated with cutpoint is split into bins equally-sized groups, from lowest to highest predicted probability, and for each bin the observed proportion of the categorical DV meeting the cutpoint condition is calculated (i.e. the m/M observations in that bin with the target value).

For a well-specified model, the mean predicted probability of a bin should be close to the bin's observed proportion, so plotted points are expected to fall around the unity (y = x) line.

Usage

catdv_vs_ipred(
  xpdb,
  mapping = NULL,
  cutpoint = 1,
  bins = 10,
  type = "pl",
  guide = TRUE,
  title = "Observed frequency vs. predicted probability | @run",
  subtitle = "Ofv: @ofv, Number of individuals: @nind",
  caption = "@dir",
  tag = NULL,
  xlab = c("probability", "basic"),
  facets,
  .problem,
  quiet,
  ...
)

Arguments

xpdb

<xp_xtras> or <xpose_data> object

mapping

ggplot2 style mapping

cutpoint

<numeric> Of defined probabilities, which one to use in plots.

bins

<numeric> Number of (roughly) equally-sized bins used to group the probability column, from lowest to highest.

type

String setting the type of plot to be used: line l, point p, smooth s and text t, or any combination thereof. See xpose::xplot_scatter().

guide

Include the unity (y = x) guide line?

title

Plot title

subtitle

Plot subtitle

caption

Plot caption

tag

Plot tag

xlab

Either use the typical basic x-axis label (the cutpoint-defined column name) or label it based on the probability/likelihood it is estimating.

facets

Additional facets

.problem

Problem number

quiet

Silence extra debugging output

...

Any additional aesthetics.

Value

The desired plot

See also

catdv_vs_dvprobs()

Examples

# Test M3 model
pkpd_m3 %>%
  # Need to ensure var types are set
  set_var_types(catdv=BLQ,dvprobs=LIKE) %>%
  # Set probs
  set_dv_probs(1, 1~LIKE, .dv_var = BLQ) %>%
  # Optional, but useful to set levels
  set_var_levels(1, BLQ = lvl_bin()) %>%
  # Plot with 5 bins
  catdv_vs_ipred(bins = 5)


# Test categorical model
vismo_xpdb <- vismo_pomod  %>%
  set_var_types(.problem=1, catdv=DV, dvprobs=matches("^P\\d+$")) %>%
  set_dv_probs(.problem=1, 0~P0,1~P1,ge(2)~P23)

# Various cutpoints and bin counts
vismo_xpdb %>%
  catdv_vs_ipred(bins = 8, xlab = "basic")

vismo_xpdb %>%
  catdv_vs_ipred(cutpoint = 2, bins = 8, xlab = "basic")

vismo_xpdb %>%
  catdv_vs_ipred(cutpoint = 3, bins = 8, xlab = "basic")