Skip to contents

Evaluation of demographic forecast accuracy

Source: R/fdm.R
compare.demogdata.Rd

Computes mean forecast errors and mean square forecast errors for each age level. Computes integrated squared forecast errors and integrated absolute percentage forecast errors for each year.

Usage

compare.demogdata(
  data,
  forecast,
  series = names(forecast$rate)[1],
  ages = data$age,
  max.age = min(max(data$age), max(forecast$age)),
  years = data$year,
  interpolate = FALSE
)

Arguments

data

Demogdata object such as created using read.demogdata containing actual demographic rates.

forecast

Demogdata object such as created using forecast.fdm or forecast.lca.

series

Name of series to use. Default: the first matrix within forecast$rate.

ages

Ages to use for comparison. Default: all available ages.

max.age

Upper age to use for comparison.

years

Years to use in comparison. Default is to use all available years that are common between data and forecast.

interpolate

If TRUE, all zeros in data are replaced by interpolated estimates when computing the error measures on the log scale. Error measures on the original (rate) scale are unchanged.

Value

Object of class "errorfdm" with the following components:

label

Name of region from which data taken.

age

Ages from data object.

year

Years from data object.

<error>

Matrix of forecast errors on rates.

<logerror>

Matrix of forecast errors on log rates.

mean.error

Various measures of forecast accuracy averaged across years. Specifically ME=mean error, MSE=mean squared error, MPE=mean percentage error and MAPE=mean absolute percentage error.

int.error

Various measures of forecast accuracy integrated across ages. Specifically IE=integrated error, ISE=integrated squared error, IPE=integrated percentage error and IAPE=integrated absolute percentage error.

life.expectancy

If data$type="mortality", function returns this component which is a matrix containing actual, forecast and actual-forecast for life expectancies.

Note that the error matrices have different names indicating if the series forecast was male, female or total.

See also

forecast.fdm,plot.errorfdm

Author

Rob J Hyndman

Examples

fr.test <- extract.years(fr.sm,years=1921:1980)
fr.fit <- fdm(fr.test,order=2)
fr.error <- compare.demogdata(fr.mort, forecast(fr.fit,20))
#> Warning: No population data available for combining upper ages
plot(fr.error)

par(mfrow=c(2,1))
plot(fr.error$age,fr.error$mean.error[,"ME"],
     type="l",xlab="Age",ylab="Mean Forecast Error")
plot(fr.error$int.error[,"ISE"],
     xlab="Year",ylab="Integrated Square Error")