magnitude: Magnitudes of Breakpoints

In appelmar/strucchange: Testing, Monitoring, and Dating Structural Changes: C++ Version

Description

Computes magnitude statistics for breakpoints.

Usage

## S3 method for class 'breakpointsfull'
magnitude(object,
  interval = 0.1, breaks = NULL, component = "trend", ...)

Arguments

object	an object of class "breakpointsfull" as computed by breakpoints.
interval	the (one-side) interval over which to compute the magnitude for the columns RMSD and MAD. If below 1, fraction of the time series length, otherwise number of observations.
breaks	how many breaks to use or which statistic to use for estimating it, see breakpoints for details. If NULL, uses the defaults.
component	which covariate(s) used for fitting the breakpointsfull object to use for magnitude estimation. The components are additive.
...	ignored

Details

The breakpoint magnitude is estimated using several statistics: the difference (diff) in the fitted value immediately before and after the break, and the root mean squared deviation (RMSD), mean absolute deviation (MAD) and mean deviation (MD) between the fitted values of the model preceding the break extrapolated over interval samples after the break, and vice versa. RMSD and MAD should be more robust estimators of magnitude compared to diff.

Value

A list with the following elements, compatible with the magnitude format as output by bfast:

Mag

a matrix containing the magnitude estimates (in columns) for each breakpoint (in rows),

m.x

sample number at which the largest break was detected, twice

y.x

fitted values before and after the largest break

Magnitude

the maximum diff magnitude of the largest break

Time

time of the largest break (same as m.x, but single value)

See Also

breakpoints

Examples

data(Nile)

trend <- 1:length(Nile)
bp.nile <- breakpoints(Nile ~ trend)

# The trend component is default, set "component" to the
# name of your coviariate(s), if it is different.
magnitude(bp.nile)

appelmar/strucchange documentation built on Nov. 27, 2021, 8:55 a.m.