nlr (version 0.1-3)

nlout.JL: Nonlinear outlier detection.

Description

Detecting outlier for nonlinear regression, is based on mixing statsitics measures and robust estimates through their covariance matrices (hat matrix). The covariance matrix in nonlinear is based on the gradient of nonlinear regression model, but it based on linear approximation of the model, instead Jacobian Leverage is used in this function.

Usage

nlout.JL(nlfited)

Arguments

nlfited

Object of type nl.fitt or nl.fitt.gn for classic estimators, nl.fitt.rob or nl.fitt.rgn for robust estimators.

Value

Result is list of nl.robmeas objects for each statistics.

"jl.vmat"

Jacobian-leverage matrix.

"jl.studres"

nl.robmeas object of Jacobian Leverage Studentised Residuals.

"jl.cook"

nl.robmeas object of Jacobian Leverage Elliptic Norm (Cook Dist).

"jl.hadi"

nl.robmeas object of Jacobian Leverage Hadi potential.

"jl.delstud"

nl.robmeas object of Jacobian Leverage Deletion Studentized.

"jl.dffits"

nl.robmeas object of Jacobian Leverage DFFITS.

"jl.atk"

nl.robmeas object ofJacobian Leverage Atkinson Distance.

Details

The outlier detection measutred used in this function are studentized residuals and Cook Distance. They are mixture of estimators and Jacobians. They are successful for detecting outlier only if combine with robust fits, eventhough the function can work with classic fits but it is not recomended. Riazoshams and Midi (2014)

References

Riazoshams H, Habshah M and Adam MB 2009 On the outlier detection in nonlinear regression. 3(12), 243-250.

Riazoshams H and Midi H 2014 Robust Leverage and outlier detection measures in nonlienar regression, 2014 (Unpublished manuscript).

See Also

nl.fitt, nl.fitt.gn, nl.fitt, nl.fitt.gn, nl.fitt.rob, nl.fitt.rgn, nl.robmeas, nlr, nlout

Examples

# NOT RUN {
d<-list(xr=Weights$Date, yr=Weights$Weight)
wmodel <- nlr(nlrobj1[[2]],data=d,control=nlr.control(method = "OLS",trace=TRUE))
a=nlout.JL(wmodel)
plot(a[[2]])
# }