predict.Mort2Dsmooth: Predict Method for 2D P-splines Fits

Description

Obtains predictions and optionally estimates standard errors of those predictions from a fitted Mort2Dsmooth object.

Usage

## S3 method for class 'Mort2Dsmooth'
## S3 method for class 'Mort2Dsmooth':
predict(object, newdata = NULL, 
        type = c("link", "response"), 
        se.fit = FALSE, ...)

Arguments

object

An object of class "Mort2Dsmooth", usually, a result of a call to Mort2Dsmooth.

newdata

Optionally, a list in which to look for x and/or y with which to predict. If omitted, the fitted linear predictors are used.

type

The type of prediction required. The default ("link") is on the scale of the linear predictors; the alternative "response" is on the scale of the response variable.

se.fit

Logical switch indicating if standard errors are required. Default: FALSE.

...

Other predict parameters to passed to predict.

Value

If se.fit = FALSE, a matrix of predictions. If se.fit = TRUE, a list with components
fitA matrix of predictions.
se.fitA matrix of estimated standard errors.

Details

If newdata is omitted the predictions are based on the data used for the fit. Note that, in common with other prediction functions, any offset supplied as an argument is always ignored when predicting, unlike offsets specified in modelling. The user can provide also a single predictor (either x or y) within the argument newdata. The name within the list newdata must be named x and y. Forecast is not possible, therefore newdata has to be include within the range of the original x and y.

Examples

Run this code

# selected data
years <- 1980:2006
ages <- 80:100
death <- selectHMDdata("Denmark", "Deaths", "Females",
                       ages = ages, years = years)
exposure <- selectHMDdata("Denmark", "Exposures", "Females",
                          ages = ages, years = years)
# fit
fit <- Mort2Dsmooth(x=ages, y=years, Z=death, offset=log(exposure),
                    method=3, lambdas=c(100,500))

# predict and computing standard errors
pre <- predict(fit, se.fit=TRUE)

# plotting over ages and years 
# !hard to distinguish between upper and lower confidence bounds
grid. <- expand.grid(x = ages, y = years, gr = 1:2)
grid.$lmx <- c(c(pre$fit - 2*pre$se.fit), c(pre$fit + 2*pre$se.fit))
wireframe(lmx ~ x * y, data = grid., groups = gr,
          scales = list(arrows = FALSE),
          drape = TRUE, colorkey = TRUE)

# plotting age 90
plot(years, log(death[11,] / exposure[11,]),
     main="Mortality rates, log-scale. Danish females, age 90, 1980:2006")
lines(years, pre$fit[11,], lwd=2, col=2)
lines(years, pre$fit[11,] + 2*pre$se.fit[11,], lwd=2, col=2, lty=2)
lines(years, pre$fit[11,] - 2*pre$se.fit[11,], lwd=2, col=2, lty=2)

# compute log-death rates for each calendar month and calendar ages
newyears12 <- seq(1990, 2000, length=11*11)
newages12 <- seq(90, 100, length=11*11)
newdata12 <- list(x=newages12, y=newyears12)
pre12 <- predict(fit, newdata=newdata12, se.fit=TRUE)

# death rates in June 1995 at age 95.5
which.age <- which(newages12==95.5)
which.year <- which(newyears12==1995.5)
exp(pre12$fit[which.age, which.year] +
    c(-2*pre12$se.fit[which.age, which.year], 0, 2*pre12$se.fit[which.age, which.year]))