Transcript Who`s doing what?

Summary of Topic ii (Tabular Data Protection)
 Frequency
Tables
 Magnitude
Tables
 Web
Access
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Summary of Topic ii (Tabular Data Protection)
Frequency Tables
C. Dwork et al. (Microsoft Research) suggest a new concept for
disclosure risk avoidance for frequency data - “differential privacy” – and
techniques to ensure it. These techniques are based on adding noise to
Fourier coefficients corresponding to a given contingency table.

A new method
for assessing disclosure
riskother
(i.e.SDC
risk of
attributeensure
Limitations
(No of variables/Categories)?
Could
methods
disclosure)
for tables
of counts,
the subtraction
- attribution
probability
differential
privacy?
Applicable
to tabulations
from a survey
with hundreds
of
(SAP) method has been proposed by D. Smith and M. Elliot (University of
variables?
Manchester).
 N. Shlomo (University of Southampton) compares the performance of
several techniques to protect population counts tables with respect to
disclosure risk and information loss.
 Statistics New Zealand uses Random Rounding, a mean cell size rule
and a threshold rule for SDC of population count tables. M. Camden et al.
calculate measures for utility and safety assessing the quality of this SDC
concept.


J.J. Salazar (Univerity La Laguna) explains advantages and
disadvantages of the mathematical models for Controlled (Integer)
Rounding vs. (continous) Tabular Adjustment.
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Summary of Topic ii (Tabular Data Protection)
Frequency Tables
C. Dwork et al. (Microsoft Research) suggest a new concept for
disclosure risk avoidance for frequency data - “differential privacy” – and
techniques to ensure it. These techniques are based on adding noise to
Fourier coefficients corresponding to a given contingency table.
 A new method for assessing disclosure risk (i.e. risk of attribute
disclosure) for tables of counts, the subtraction - attribution probability
(SAP) method has been proposed by D. Smith and M. Elliot (University of
Manchester).
How
N. to
Shlomo
of Southampton)
embed(University
the SAP method
into an SDCcompares
strategy? the performance of
several techniques to protect population counts tables with respect to
disclosure risk and information loss.
 Statistics New Zealand uses Random Rounding, a mean cell size rule
and a threshold rule for SDC of population count tables. M. Camden et al.
calculate measures for utility and safety assessing the quality of this SDC
concept.


J.J. Salazar (Univerity La Laguna) explains advantages and
disadvantages of the mathematical models for Controlled (Integer)
Rounding vs. (continous) Tabular Adjustment.
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Summary of Topic ii (Tabular Data Protection)
Frequency Tables
C. Dwork et al. (Microsoft Research) suggest a new concept for
disclosure risk avoidance for frequency data - “differential privacy” – and
techniques to ensure it. These techniques are based on adding noise to
Fourier coefficients corresponding to a given contingency table.
 A new method for assessing disclosure risk (i.e. risk of attribute
disclosure) for tables of counts, the subtraction - attribution probability
(SAP) method has been proposed by D. Smith and M. Elliot (University of
Manchester).
 N. Shlomo (University of Southampton) compares the performance of
several techniques to protect population counts tables with respect to
disclosure risk and information loss.
Cell
Statistics
New Zealand
uses Random
Rounding,
a mean
cell size
rule
suppression/simple
Imputation
least distortion
– Cell
suppression
best
and a threshold rule for SDC of population count tables. M. Camden et al.
method???
calculate measures for utility and safety assessing the quality of this SDC
concept.


J.J. Salazar (Univerity La Laguna) explains advantages and
disadvantages of the mathematical models for Controlled (Integer)
Rounding vs. (continous) Tabular Adjustment.
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Summary of Topic ii (Tabular Data Protection)
Frequency Tables





C. Dwork et al. (Microsoft Research) suggest a new concept for
disclosure risk avoidance for frequency data - “differential privacy” – and
techniques to ensure it. These techniques are based on adding noise to
Fourier coefficients corresponding to a given contingency table.
A new method for assessing disclosure risk (i.e. risk of attribute
disclosure) for tables of counts, the subtraction - attribution probability
(SAP) method has been proposed by D. Smith and M. Elliot (University of
Manchester).
N. Shlomo (University of Southampton) compares the performance of
several techniques to protect population counts tables with respect to
disclosure risk and information loss.
Statistics New Zealand uses Random Rounding, a mean cell size rule
and a threshold rule for SDC of population count tables. M. Camden et al.
calculate measures for utility and safety assessing the quality of this SDC
concept.
J.J. Salazar (Univerity La Laguna) explains advantages and
disadvantages of the mathematical models for Controlled (Integer)
Rounding vs. (continous) Tabular Adjustment.
Do integrality problems matter for magnitude tables? Could variable controlled
rounding be modelled (efficiently)?
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Summary of Topic ii (Tabular Data Protection)
Magnitude Tables
The US Census Bureau adds noise to the underlying microdata prior to
tabulation. The paper by L. Zayatz also addresses other SDC research
areas at the USBC like synthetic micro data generation (also used to
protect frequency tabular data) and a remote microdata analysis system.

L. reactions?
Cox (US NCHS) compares properties of two methods for Controlled
User
Tabular Adjustment, one based on LP technology, the other on iterative
proportional fitting.
 Using tabular structures of EIA publications, and artificial microdata, R.
Dandekar compares empirically the performance of various methods for
tabular data protection, i.e. CTA, USBC’s noise method and cell
suppression.
 P.P. de Wolf (CBS Netherlands) discusses a possible way to describe a
simple class of linked tables that is often considered at NSI's.

Web Access

The USDA Economic Research Service has developed web-based data
delivery tools for access to farm survey data (M. Morchart, C. Towe)
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Summary of Topic ii (Tabular Data Protection)
Magnitude Tables




The US Census Bureau adds noise to the underlying microdata prior to
tabulation. The paper by L. Zayatz also addresses other SDC research
areas at the USBC like synthetic micro data generation (also used to
protect frequency tabular data) and a remote microdata analysis system.
L. Cox (US NCHS) compares properties of two methods for Controlled
Tabular Adjustment, one based on LP technology, the other on iterative
proportional fitting.
Using tabular structures of EIA publications, and artificial microdata, R.
Dandekar compares empirically the performance of various methods for
tabular data protection, i.e. CTA, USBC’s noise method and cell
suppression.
P.P. de Wolf (CBS Netherlands) discusses a possible way to describe a
simple class of linked tables that is often considered at NSI's.
Any plans for Linked Tables version of t-ARGUS HiTaS?
Web Access

The USDA Economic Research Service has developed web-based data
delivery tools for access to farm survey data (M. Morchart, C. Towe)
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Summary of Topic ii (Tabular Data Protection)
Magnitude Tables




The US Census Bureau adds noise to the underlying microdata prior to
tabulation. The paper by L. Zayatz also addresses other SDC research
areas at the USBC like synthetic micro data generation (also used to
protect frequency tabular data) and a remote microdata analysis system.
L. Cox (US NCHS) compares properties of two methods for Controlled
Tabular Adjustment, one based on LP technology, the other on iterative
proportional fitting.
Using tabular structures of EIA publications, and artificial microdata, R.
Dandekar compares empirically the performance of various methods for
tabular data protection, i.e. CTA, USBC’s noise method and cell
suppression.
P.P. de Wolf (CBS Netherlands) discusses a possible way to describe a
simple class of linked tables that is often considered at NSI's.
Web Access

The USDA Economic Research Service has developed web-based data
delivery tools for access to farm survey data (M. Morchart, C. Towe)
Details on cell suppression approach within the tool?
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden
Discussion/Questions to the authors







Dwork et al.: Limitations (No of variables/Categories)? Could
other SDC methods ensure differential privacy? Applicable to
tabulations from a survey with hundreds of variables?
Smith/Elliot: How to embed the SAP method into an SDC
strategy?
Shlomo: Cell suppression+simple imputation least distortion –
Cell suppression best method???
Salazar: Do integrality problems matter for magnitude tables?
Is variable controlled rounding a realistic option?
Zayatz: User reactions?
De Wolf: Any plans for Linked Tables version of t-ARGUS
HiTaS?
Morehart/Towe: Details on cell suppression approach within
the tool?
© Statistisches Bundesamt, IIA - Mathematisch Statistische Methoden