Introduction to 'easySdcTable'"

In easySdcTable: Easy Interface to the Statistical Disclosure Control Package 'sdcTable' Extended with Own Implementation of 'GaussSuppression'

Introduction and background

Below is given an introductory demonstration of the function ProtectTable() which enables an easy interface to the statistical disclosure control package 'sdcTable' (https://CRAN.R-project.org/package=sdcTable). To see the input and output to functions in sdcTable consult the function ProtectTable1() which is an underlying function of ProtectTable(). Note that 'easySdcTable' is not as general as 'sdcTable'.

This package was originally developed as a part of the modernization of the production of the key figures on municipal activities in Norway (https://www.ssb.no/en/offentlig-sektor/kostra). The fictitious example data is generated to be similar to realistic data from Norwegian municipalities and the variable names are (unfortunately) in Norwegian.

library(knitr)
library(easySdcTable)

The demonstration below is based on the data from example 2 in the package and first we will use the unstacked data.

Before demonstrating ProtectTable() a few words about other possibilities.

Note after easySdcTable version 0.8.0

Method "Gauss" has been made default (See NEWS). This is an additional method that is not available in sdcTable.

News in easySdcTable version 0.9.0

Method "Gauss" improved when zeros omitted in input data.

• A potential pitfall when doing secondary suppression by Gaussian elimination is cases with protectZeros = TRUE and where zeros are omitted in input data. The underlying function, GaussSuppression, produce a warning in such cases (introduced in SSBtools version 1.2.2) with text: "Suppressed cells with empty input will not be protected. Extend input data with zeros?". Cases where this warning occur is now avoided within ProtectTable. Internally data are automatically extended when needed.

Another comment about "Gauss" and zeros omitted in input

When hierarchies are supplied as input (parameter dimList) and when there exist input codes in the hierarchies that are totally missing in the data, it is still possible to create a situation with warning: "Suppressed cells with empty input will not be protected. Extend input data with zeros?". This behavior will not be changed. Ignore the warning if such codes represent structural zeros. If not, extend data with zero frequencies (see parameter freqVar) so that these code are represented in data.

Graphical user interface and $\tau$-ARGUS

A graphical user interface based on 'shiny' can be started by:

PTgui()

To start the gui with example data and catch output:

out <- PTgui(data=EasyData("z1w"))

To start the gui with possibility to run tau-argus:

exeArgus <- "C:/Tau/TauArgus4.1.4/TauArgus.exe" # Tau-argus executable 
pathArgus <- "C:/Users/nnn/Documents" # Folder for (temporary) tau-argus files
PTgui(exeArgus=exeArgus, pathArgus=pathArgus) 

The interface to tau-argus make use of functionality in ‘sdcTable’. See the documentation of ProtectTable() for more information.

Unstacked data

The input data

The function EasyData() in ‘easySdcTable’ returns example data.

z2w <- EasyData("z2w") 
print(z2w, row.names=FALSE)

By unstacked data we mean that counts (cell frequencies) are in more than a single column.

Running ProtectTable

In this case we have counts in columns four to seven. Using the dimensional variable in the first column we can run ProtectTable by:

ex2w <- ProtectTable(z2w,1,4:7) 

The data with computed totals

The output element freq contains the data with computed totals.

print(ex2w$freq, row.names=FALSE)

SdcStatus

In the output element sdcStatus the cells are coded as "u" (primary suppressed), "x" (secondary suppression), and "s" (can be published).

print(ex2w$sdcStatus, row.names=FALSE)

Suppressed data

The output element suppressed is the same as freq with the exception that suppressed cells ("u" and "x") are set to missing (NA).

print(ex2w$suppressed, row.names=FALSE)

Using named input and the HITAS method

Now we specify the variables using names instead of numbers. Furthermore we use the "HITAS" method. The default method is "SIMPLEHEURISTIC" and other possibilities are "OPT" and "HYPERCUBE". The latter is not possible in cases with two linked tables.

ex2wHITAS <- ProtectTable(z2w,dimVar = c("region"),freqVar = c("annet", "arbeid", "soshjelp", "trygd"), method="HITAS")  

print(ex2wHITAS$suppressed, row.names=FALSE)

More advanced use of ProtectTable

Here we include the tree first variables as dimensional variables. It will be detected automatically that "fylke" and "kostragr" are hierarchically related to "region" and that they are not hierarchically related to each other. Zeros will not be suppressed and we will only primarily suppress ones and twos.

ex2wAdvanced <- ProtectTable(z2w,dimVar = c("region", "fylke","kostragr"),freqVar = c("annet", "arbeid", "soshjelp", "trygd"), maxN=2, protectZeros=FALSE, method = "SIMPLEHEURISTIC", addName=TRUE)  

Suppressed data with totals and sub-totals

Now the output data will contain sub-totals of the additional variables and the secondary suppression has taken those sub-totals into account. Since addName is TRUE, sub-totals are named using "fylke" and "kostragr".

print(ex2wAdvanced$suppressed, row.names=FALSE)

Info

The output element info contains three parts.

1. Since we have unstacked data an extra variable, named var1, is created. How the categories of this variable are related to the variable names are described. Here these categories are simply the variable names. In more advanced cases it is possible that more than a single variable are created from the variable names.

2. Secondly, it is described how the tables(s) are created from the variables. In this case the problem is solved using two linked tables. The first table involves "fylke" and the second table involves "kostragr".

3. The last part contains summary output for each of the two linked tables.

prmatrix(ex2wAdvanced$info,rowlab=rep("",99),collab="",quote=FALSE)

Stacked data

Now we will use a stacked variant of the same data. A single column ("ant") holds cell counts and the variable "hovedint" contains the four categories "annet", "arbeid", "soshjelp" and "trygd".

z2 <- EasyData("z2") 
print(z2)

We run ProtectTable with stacked data the same way as with unstacked data.

ex2 <- ProtectTable(z2,dimVar = c("region", "hovedint", "kostragr"), freqVar = "ant") 

Instead of three output elements we now have the single element data:

print(ex2$data)

Unlike above addName is FALSE (default) and therefore the sub-totals "300" and "400" are written without "kostragr".

Assuming micro data

Below no columns holds cell counts (no freqVar input) and therefore it is assumed that each cell count is one. For this data set this is not realistic, but in other cases rows are replicated.

ex2micro <- ProtectTable(z2,dimVar = c("region", "hovedint", "kostragr")) 

print(ex2micro$data)

.

easySdcTable documentation built on Dec. 28, 2022, 2:29 a.m.