Computes a Gaussian response forecast at future occasions of already
observed series. The forecast conditions on the complete observed response
vector at fitted parameter values. se.fit is the corresponding
conditional predictive standard deviation; it excludes uncertainty in the
fitted parameters and is not a calibrated prediction interval.
Arguments
- object
A fitted native temporal
gllvmTMB()model.- newdata
A complete trait panel for one or more future series–occasion pairs. Each series must occur in the fitted data and each requested time must be strictly after that series' final fitted time.
- se.fit
Return the fitted-parameter conditional predictive standard deviation.
Value
newdata, in its input row order, with est on the Gaussian
identity-link scale and, when requested, se.fit.
Details
This first route supports an unreplicated, temporal_indep(), Gaussian
identity-link model. It deliberately refuses new series, replicated panels,
other temporal covariance modes, non-temporal random-effect tiers, source
combinations, past or observed occasions, and non-Gaussian families. Those
cases require their own conditioning and validation contracts.
Examples
d <- expand.grid(series = c("a", "b"), occasion = 1:3,
trait = c("t1", "t2", "t3"), KEEP.OUT.ATTRS = FALSE)
d$value <- with(d, occasion + as.numeric(factor(trait)) / 10)
fit <- gllvmTMB(value ~ 0 + trait +
temporal_indep(0 + trait | series, time = occasion),
data = d, unit = "series", family = gaussian(), silent = TRUE,
control = gllvmTMBcontrol(se = FALSE))
future <- expand.grid(series = c("a", "b"), occasion = 4,
trait = c("t1", "t2", "t3"), KEEP.OUT.ATTRS = FALSE)
forecast_temporal(fit, future, se.fit = TRUE)
#> series occasion trait est se.fit
#> 1 a 4 t1 2.1 0.8164966
#> 2 b 4 t1 2.1 0.8164966
#> 3 a 4 t2 2.2 0.8164966
#> 4 b 4 t2 2.2 0.8164966
#> 5 a 4 t3 2.3 0.8164966
#> 6 b 4 t3 2.3 0.8164966