autoSmoothAPC {smoothAPC}

Smooths demographic data using automatically estimated parameters and optionally taking into account period and cohort effects

Description

If period and cohort effects are taken into account (effects = TRUE) the method uses all available years and diagonals for estimation of the period and cohort effects.

Usage

autoSmoothAPC(data, effects = TRUE, cornerLength = 7,
  affdDiagonals = NULL, affdYears = NULL, lower = head(c(0.01, 0.01, 0.01,
  2, 0.001, 2, 0.001), 3 + effects * 4), upper = head(c(1.2, 1.8, 1.2, 12,
  0.4, 12, 0.4), 3 + effects * 4), init = head(c(0.1, 0.1, 0.2, 4, 0.001, 4,
  0.001), 3 + effects * 4), reltol = 0.001, parameters = NULL, trace = F,
  control = list(nnzlmax = 1e+06, nsubmax = 2e+06, tmpmax = 2e+05),
  weights = NULL)

Arguments

data

Demographic data (log mortality) presented as a matrix. Row numbers represent ages and column numbers represet time.

effects

Controls if the cohort and period effects are taken into account.

cornerLength

Sets the smallest length of a diagonal to be considered for cohort effects.

affdDiagonals

Diagonals to be used for cohort effects. The first diagonal is at the bottom left corner of the data matrix (maximal age and minimal time in the data matrix).

affdYears

Years to be used for period effects.

lower

Lowest possible values for the optimization procedure.

upper

Highest possible values for the optimization procedure.

init

Initial values for the optimization procedure.

reltol

Relative tolerance parameter to be supplied to optim function.

parameters

Optional model parameters. If not provided, they are estimated.

trace

Controls if tracing is on.

control

The control data passed directly to rq.fit.sfn function.

weights

Define how much every observation effect the resulting smooth surface. The parameter must have same dimentions as data parameter. Weights can be set to reciprocal of estimated standard deviation of the data.

Value

A list of four components: smooth surface, period effects, cohort effects and parameters used for smoothing (passed as a parameter or estimated).

Author(s)

Alexander Dokumentov

References

http://robjhyndman.com/publications/mortality-smoothing/

See Also

smoothAPC and signifAutoSmoothAPC. The latter might give slightly better performance.

Examples

library(demography)
m <- log(fr.mort$rate$female[1:30, 150:160])
plot(m)
sm <- autoSmoothAPC(m)
plot(sm)
plot(sm, "period")
plot(sm, "cohort")