autoplot.ref_fit

Usage

autoplot.ref_fit(
  object,
  type = c("centiles", "trajectories", "support", "adaptation"),
  outcome = NULL,
  x = NULL,
  newdata = NULL,
  by = NULL,
  data = NULL,
  id = NULL,
  time = NULL,
  centiles = c(0.05, 0.25, 0.5, 0.75, 0.95),
  ...
)

autoplot.ref_assessment(
  object,
  type = c("calibration", "worm", "qq", "conditional"),
  level = 0.95,
  ...
)

autoplot.ref_scores(object, type = c("profile", "heatmap"), id = NULL, ...)

autoplot.ref_forecast(
  object,
  type = c("fan"),
  centiles = c(0.05, 0.25, 0.5, 0.75, 0.95),
  ...
)

autoplot.ref_derivative(object, type = c("velocity"), level = 0.95, ...)

autoplot.ref_transition(object, type = c("velocity", "innovation", "change"), ...)

autoplot.ref_dynamics(
  object,
  type = c("kernel", "calibration", "thrive"),
  data = NULL,
  id = NULL,
  time = NULL,
  level = NULL,
  outcome = NULL,
  x = NULL,
  centiles = c(0.05, 0.25, 0.5, 0.75, 0.95),
  anchors = c(0.05, 0.25, 0.5, 0.75, 0.95),
  from = NULL,
  horizon = 1,
  thrive = 0.025,
  ...
)

Every autoplot() method returns a ggplot object styled with theme_referent(). Individual observations are overlays, not part of the fitted geometry.

Arguments

Argument	Description
object	A fit, assessment, score table, forecast, transition, or dynamics object.
type	Plot kind.
outcome	Outcome name for multi-outcome fits.
x	Covariate mapped to the x-axis (default age if present).
newdata	Observations to overlay or to classify.
by	Optional grouping factor for faceted charts.
data	Visit-level data for trajectory and dynamics-calibration plots.
id	Subject identifier column.
time	Time column.
centiles	Probability levels for centile and fan charts.
...	Unused.
level	Coverage of the envelope on the Q-Q and worm plots, of the pointwise interval on a velocity plot, or of the correlation interval around a thrive line.
anchors	Starting centiles for type = "thrive".
from	Anchor times for type = "thrive" (default: a grid over the reference range).
horizon	Forward step(s) for type = "thrive"; a vector traces a continuous path rather than a single segment.
thrive	Conditional centile of the step for type = "thrive"; 0.5 gives the pure regression-to-the-mean path.

Details

For a ref_fit:

• "centiles": the fitted conditional distribution along x (median, dashed outer centiles, and shaded bands), optionally faceted by a factor covariate and with newdata overlaid as points.
• "trajectories": the centile chart with subject paths from data drawn over it.
• "support": where newdata falls relative to the reference support (histogram of x coloured by ref_support() status).
• "adaptation": location offsets (with standard errors) estimated by ref_adapt().

For a ref_assessment, "calibration" plots observed against nominal coverage, "conditional" the largest fitted drift of z against each numeric covariate (factor levels are in the table only), and "qq" / "worm" the ordered Z scores against their expected normal order statistics, raw or detrended (observed minus expected), inside the envelope a calibrated model implies. Under a correctly specified model the held-out centiles are iid uniform, so the i-th ordered centile is Beta(i, n - i + 1): the pointwise band is its normal transform in closed form, and the simultaneous band uses equal local levels (Aldor-Noiman et al. 2013, Am. Stat. 67:249) with the common local level calibrated by Monte Carlo so the whole ordered sample stays inside with probability level. The corner tally names both bands, since about 1 - level of the points are expected outside the pointwise band even when the model is right.

For a ref_dynamics, "kernel" draws the fitted process correlation against lag, "calibration" draws the same Q-Q chart of held-out innovation Z from ref_transition() on data (which needs data, id, and time), and "thrive" overlays conditional forecast quantile paths on the centile chart. A thrive line starts on the anchors centile at each time in from and follows the thrive quantile of the forecast distribution horizon time units later; it is ref_forecast() on a grid of one-visit histories, so the shaded band is the level interval implied by the standard error of the estimated process correlation. Paths whose lag falls outside the range the dynamics were estimated on are drawn dotted.

For a ref_derivative, "velocity" draws the chart velocity of each centile against time with its pointwise delta-method interval. That interval is the sampling error of the fitted coefficients; it does not cover the bias of the spline itself, which dominates when the basis dimension k of the time smooth is too small. The subtitle says so.

Value

A ggplot.

Examples

ref <- ref_simulate(150, seed = 1)
fit <- ref_fit(ref_spec(ref_gaussian(), ~ s(age, k = 5) + sex), ref, "y")
autoplot(fit, type = "centiles", by = sex, newdata = ref[1:20, ])

sc <- predict(fit, newdata = ref[1:6, ], uncertainty = "conditional")
autoplot(sc, type = "heatmap")