sdcMicro (version 4.1.0)

shuffle: Shuffling and EGADP

Description

Data shuffling and General Additive Data Perturbation.

Usage

shuffle(obj, form, method="ds", weights=NULL, covmethod="spearman",
        regmethod="lm", gadp=TRUE)

Arguments

obj

An object of class sdcMicroObj or a data.frame including the data.

form

An object of class formula (or one that can be coerced to that class): a symbolic description of the model to be fitted. The responses have to consists of at least two variables of any class and the response variables have t

method

currently either the original form of data shuffling (ds - default), mvn or mlm, see the details section. The last method is in experimental mode and almost untested.

weights

Survey sampling weights. Automatically chosen when obj is of class sdcMicroObj.

covmethod

Method for covariance estimation. spearman, pearson and dQuote{mcd} are possible. For the latter one, the implementation in package robustbase is used.

regmethod

Method for multivariate regression. lm and MM are possible. For method MM, the function rlm from package MASS is applied.

gadp

TRUE, if the egadp results from a fit on the origianl data is returned.

Value

• If obj is of class sdcMicroObj the corresponding slots are filled, like manipNumVars, risk and utility. If obj is of class data.frame an object of class micro with following entities is returned:
• shConf
  the shuffled numeric key variables
• egadp
  the perturbed (using gadp method) numeric key variables

Details

Perturbed values for the sensitive variables are generated. The sensitive variables have to be stored as responses in the argument form, which is the usual formula interface for regression models in R. For method ds the EGADP method is applied on the norm inverse percentiles. Shuffling then ranks the original values according to the GADP output. For further details, please see the references. Method mvn uses a simplification and draws from the normal Copulas directly before these draws are shuffled. Method mlm is also a simplification. A linear model is applied the expected values are used as the perturbed values before shuffling is applied.

References

K. Muralidhar, R. Parsa, R. Saranthy (1999). A general additive data perturbation method for database security. Management Science, 45, 1399-1415. K. Muralidhar, R. Sarathy (2006). Data shuffling - a new masking approach for numerical data. Management Science, 52(5), 658-670, 2006. M. Templ, B. Meindl. (2008). Robustification of Microdata Masking Methods and the Comparison with Existing Methods, in: Lecture Notes on Computer Science, J. Domingo-Ferrer, Y. Saygin (editors.); Springer, Berlin/Heidelberg, 2008, ISBN: 978-3-540-87470-6, pp. 14-25.

See Also

rankSwap, lm

Examples

data(Prestige,package="car")
form <- formula(income + education ~ women + prestige + type, data=Prestige)
sh <- shuffle(obj=Prestige,form)
plot(Prestige[,c("income", "education")])
plot(sh$sh)
colMeans(Prestige[,c("income", "education")])
colMeans(sh$sh)
cor(Prestige[,c("income", "education")], method="spearman")
cor(sh$sh, method="spearman")

## for objects of class sdcMicro:
data(testdata2)
sdc <- createSdcObj(testdata2,
  keyVars=c('urbrur','roof','walls','water','electcon','relat','sex'), 
  numVars=c('expend','income','savings'), w='sampling_weight')
sdc <- shuffle(sdc, method=c('ds'),regmethod= c('lm'), covmethod=c('spearman'), 
		form=savings+expend ~ urbrur+walls)