Rfit (version 0.27.0)

drop.test: Drop (Reduction) in Dispersion Test

Description

Given two full model fits, this function performs a reduction in dispersion test.

Usage

drop.test(fitF, fitR = NULL)

Value

F

Value of the F test statistic

p.value

The observed significance level of the test (using an F quantile)

RD

Reduced model dispersion minus Full model dispersion

tauhat

Estimate of the scale parameter (using the full model residuals)

df1

numerator degrees of freedom

df2

denominator degrees of freedom

Arguments

fitF

An object of class rfit. The full model fit.

fitR

An object of class rfit. The reduced model fit.

Author

John Kloke, Joseph McKean

Details

Rank-based inference procedure analogous to the traditional (LS) reduced model test.

The full and reduced model dispersions are calculated. The reduction in dispersion test, or drop test for short, has an asymptotic chi-sq distribution. Simulation studies suggest using F critical values. The p-value returned is based on a F-distribution with df1 and df2 degrees of freedom where df1 is the difference in the number of parameters in the fits of fitF and fitR and df2 is the residual degrees of freedom in the fit fitF.

Both fits are based on a minimization routine. It is possible that resulting solutions are such that the fitF$disp > fitRdisp. We recommend starting the full model at the reduced model fit as a way to avoid this situation. See examples.

Checks to see if models appear to be proper subsets. The space spanned by the columns of the reduced model design matrix should be a subset of the space spanned by the columns of the full model design matrix.

References

Hettmansperger, T.P. and McKean J.W. (2011), Robust Nonparametric Statistical Methods, 2nd ed., New York: Chapman-Hall.

See Also

rfit

Examples

y<-rnorm(47)
x1<-rnorm(47)
x2<-rnorm(47)
fitF<-rfit(y~x1+x2)
fitR<-rfit(y~x1)
drop.test(fitF,fitR)

## try starting the full model at the reduced model fit ##
fitF<-rfit(y~x1+x2,yhat0=fitR$fitted)
drop.test(fitF,fitR)