Predict from a longitudinal GAMLSS-copula model

Usage

# S3 method for class 'gamlss.longitudinal'
predict(
  object,
  newdata = NULL,
  type = c("response", "mean", "mu", "median", "parameters", "quantile", "cdf",
    "density", "probability"),
  probs = c(0.025, 0.5, 0.975),
  q = NULL,
  y = NULL,
  direction = c("below", "above"),
  se.fit = FALSE,
  interval = c("none", "confidence"),
  level = 0.95,
  vcov_method = "analytical",
  ...
)

Arguments

object: A fitted gamlss.longitudinal object.
newdata: Optional new data. If omitted, fitted rows are used.
type: Prediction type: "mean" returns the fitted marginal response mean, "median" returns the fitted marginal median, "mu" returns the fitted GAMLSS mu parameter, "response" is retained as a compatibility alias for "mu", "parameters" returns all fitted marginal distribution parameters, "quantile" returns fitted marginal quantiles, "cdf"/"density" evaluate the fitted marginal CDF or density, and "probability" returns probabilities below or above a threshold.
probs: Quantile probabilities when type = "quantile".
q: Threshold value for type = "cdf" or type = "probability". Defaults to observed responses for type = "cdf" when available.
y: Evaluation value for type = "density". Defaults to observed responses when available.
direction: Probability direction for type = "probability".
se.fit: Logical; for type = "response", "mu", or "mean", return approximate delta-method standard errors for the fitted mu linear predictor contribution. For non-mu response means this is a first-order approximation and should be treated as exploratory.
interval: Interval type. "confidence" adds response-scale confidence limits when type = "response", "mu", or "mean".
level: Confidence level for interval = "confidence".
vcov_method: Variance-covariance method passed to vcov.gamlss.longitudinal().
...: Additional arguments reserved for future methods.

Value

A numeric vector for type = "response", "mu", "mean", or "median" unless standard errors or intervals are requested; a data frame otherwise.

Details

predict() returns marginal summaries from the fitted distribution at each requested row. The fitted copula/dependence structure is not used to condition a row's prediction on that subject's other observed responses. Instead, dependence affects prediction indirectly through the coefficients estimated by the joint copula likelihood, and through se.fit/confidence intervals when the covariance matrix is computed from the joint model. Use simulate.gamlss.longitudinal() for fitted-data trajectory simulation that preserves the fitted copula dependence structure. With newdata, simulation is unconditional and uses the model-implied dependence evaluated on the supplied panel.

type = "response" is a soft-deprecated compatibility alias for type = "mu" because GAMLSS mu is not the response mean for every family. New code should use type = "mean" for response-mean estimands or type = "mu" for the distribution parameter.