Skip to contents

Get fitted (component) means from a hhh4 model.


# S3 method for class 'hhh4'
predict(object, newSubset=object$control$subset,
        type="response", ...)



fitted hhh4 model (class "hhh4").


subset of time points for which to return the predictions. Defaults to the subset used for fitting the model, and must be a subset of 1:nrow(object$stsObj).


the type of prediction required. The default ("response" or, equivalently, "mean") is on the scale of the response variable (mean = endemic plus epidemic components). The alternatives are: "endemic", "epidemic", "epi.own" (i.e. the autoregressive part), and "epi.neighbours" (i.e. the spatio-temporal part).


unused (argument of the generic).


matrix of fitted means for each time point (of newSubset) and region.


Predictions for “newdata”, i.e., with modified covariates or fixed weights, can be computed manually by adjusting the control list (in a copy of the original fit), dropping the old terms, and using the internal function meanHHH directly, see the Example.


Michaela Paul and Sebastian Meyer


## simulate simple seasonal noise with reduced baseline for t >= 60
t <- 0:100
y <- rpois(length(t), exp(3 + sin(2*pi*t/52) - 2*(t >= 60)))
obj <- sts(y)

## fit true model
fit <- hhh4(obj, list(end = list(f = addSeason2formula(~lock)),
                      data = list(lock = as.integer(t >= 60)),
                      family = "Poisson"))
coef(fit, amplitudeShift = TRUE, se = TRUE)

## compute predictions for a subset of the time points
stopifnot(identical(predict(fit), fitted(fit)))
lines(40:80, predict(fit, newSubset = 40:80), lwd = 2)

## advanced: compute predictions for "newdata" (here, a modified covariate)
mod <- fit
mod$terms <- NULL  # to be sure
mod$control$data$lock[t >= 60] <- 0.5
pred <- meanHHH(mod$coefficients, terms(mod))$mean
plot(fit, xaxis = NA)
lines(mod$control$subset, pred, lty = 2)