roblox {RobLox}

Optimally robust estimator for location and/or scale

Description

The function roblox computes the optimally robust estimator and corresponding IC for normal location und/or scale and (convex) contamination neighborhoods. The definition of these estimators can be found in Rieder (1994) or Kohl (2005), respectively.

Usage

roblox(x, mean, sd, eps, eps.lower, eps.upper, initial.est, k = 1L, 
       fsCor = TRUE, returnIC = FALSE, mad0 = 1e-4, na.rm = TRUE)

Arguments

x

vector x of data values, may also be a matrix or data.frame with one row, respectively one column/(numeric) variable.

mean

specified mean.

sd

specified standard deviation which has to be positive.

eps

positive real (0 < eps <= 0.5): amount of gross errors. See details below.

eps.lower

positive real (0 <= eps.lower <= eps.upper): lower bound for the amount of gross errors. See details below.

eps.upper

positive real (eps.lower <= eps.upper <= 0.5): upper bound for the amount of gross errors. See details below.

initial.est

initial estimate for mean and/or sd. If missing median and/or MAD are used.

k

positive integer. k-step is used to compute the optimally robust estimator.

fsCor

logical: perform finite-sample correction. See function finiteSampleCorrection.

returnIC

logical: should IC be returned. See details below.

mad0

scale estimate used if computed MAD is equal to zero

na.rm

logical: if TRUE, the estimator is evaluated at complete.cases(x).

Details

Computes the optimally robust estimator for location with scale specified, scale with location specified, or both if neither is specified. The computation uses a k-step construction with an appropriate initial estimate for location or scale or location and scale, respectively. Valid candidates are e.g. median and/or MAD (default) as well as Kolmogorov(-Smirnov) or von Mises minimum distance estimators; cf. Rieder (1994) and Kohl (2005).

If the amount of gross errors (contamination) is known, it can be specified by eps. The radius of the corresponding infinitesimal contamination neighborhood is obtained by multiplying eps by the square root of the sample size.

If the amount of gross errors (contamination) is unknown, try to find a rough estimate for the amount of gross errors, such that it lies between eps.lower and eps.upper.

In case eps.lower is specified and eps.upper is missing, eps.upper is set to 0.5. In case eps.upper is specified and eps.lower is missing, eps.lower is set to 0.

If neither eps nor eps.lower and/or eps.upper is specified, eps.lower and eps.upper are set to 0 and 0.5, respectively.

If eps is missing, the radius-minimax estimator in sense of Rieder et al. (2008), respectively Section 2.2 of Kohl (2005) is returned.

In case of location, respectively scale one additionally has to specify sd, respectively mean where sd and mean have to be a single number.

For sample size <= 2, median and/or MAD are used for estimation.

If eps = 0, mean and/or sd are computed. In this situation it's better to use function MLEstimator.

Value

Object of class "kStepEstimate".

Author(s)

Matthias Kohl [e-mail]

References

M. Kohl (2005). Numerical Contributions to the Asymptotic Theory of Robustness. Dissertation. University of Bayreuth. https://epub.uni-bayreuth.de/id/eprint/839/2/DissMKohl.pdf.

H. Rieder (1994): Robust Asymptotic Statistics. Springer. doi:10.1007/978-1-4684-0624-5

H. Rieder, M. Kohl, and P. Ruckdeschel (2008). The Costs of Not Knowing the Radius. Statistical Methods and Applications 17(1): 13-40. doi:10.1007/s10260-007-0047-7

M. Kohl, P. Ruckdeschel, and H. Rieder (2010). Infinitesimally Robust Estimation in General Smoothly Parametrized Models. Statistical Methods and Applications 19(3): 333-354. doi:10.1007/s10260-010-0133-0.

M. Kohl and H.P. Deigner (2010). Preprocessing of gene expression data by optimally robust estimators. BMC Bioinformatics 11, 583. doi:10.1186/1471-2105-11-583.

See Also

ContIC-class, rlOptIC, rsOptIC, rlsOptIC.AL, kStepEstimate-class, roptest

Examples

ind <- rbinom(100, size=1, prob=0.05) 
x <- rnorm(100, mean=ind*3, sd=(1-ind) + ind*9)

## amount of gross errors known
res1 <- roblox(x, eps = 0.05, returnIC = TRUE)
estimate(res1)
## don't run to reduce check time on CRAN
## Not run: 
confint(res1)
confint(res1, method = symmetricBias())
pIC(res1)
checkIC(pIC(res1))
Risks(pIC(res1))
Infos(pIC(res1))
plot(pIC(res1))
infoPlot(pIC(res1))

## End(Not run)

## amount of gross errors unknown
res2 <- roblox(x, eps.lower = 0.01, eps.upper = 0.1, returnIC = TRUE)
estimate(res2)
## don't run to reduce check time on CRAN
## Not run: 
confint(res2)
confint(res2, method = symmetricBias())
pIC(res2)
checkIC(pIC(res2))
Risks(pIC(res2))
Infos(pIC(res2))
plot(pIC(res2))
infoPlot(pIC(res2))

## End(Not run)

## estimator comparison
# classical optimal (non-robust)
c(mean(x), sd(x))

# most robust
c(median(x), mad(x))

# optimally robust (amount of gross errors known)
estimate(res1)

# optimally robust (amount of gross errors unknown)
estimate(res2)

# Kolmogorov(-Smirnov) minimum distance estimator (robust)
(ks.est <- MDEstimator(x, ParamFamily = NormLocationScaleFamily()))

# optimally robust (amount of gross errors known)
roblox(x, eps = 0.05, initial.est = estimate(ks.est))

# Cramer von Mises minimum distance estimator (robust)
(CvM.est <- MDEstimator(x, ParamFamily = NormLocationScaleFamily(), distance = CvMDist))

# optimally robust (amount of gross errors known)
roblox(x, eps = 0.05, initial.est = estimate(CvM.est))