Statistical Disclosure Control Methods Research Articles

Database Reconstruction Is Not So Easy and Is Different from Reidentification

Abstract In recent years, it has been claimed that releasing accurate statistical information on a database is likely to allow its complete reconstruction. Differential privacy has been suggested as the appropriate methodology to prevent these attacks. These claims have recently been taken very seriously by the U.S. Census Bureau and led them to adopt differential privacy for releasing U.S. Census data. This in turn has caused consternation among users of the Census data due to the lack of accuracy of the protected outputs. It has also brought legal action against the U.S. Department of Commerce. In this article, we trace the origins of the claim that releasing information on a database automatically makes it vulnerable to being exposed by reconstruction attacks and we show that this claim is, in fact, incorrect. We also show that reconstruction can be averted by properly using traditional statistical disclosure control (SDC) techniques. We further show that the geographic level at which exact counts are released is even more relevant to protection than the actual SDC method employed. Finally, we caution against confusing reconstruction and reidentification: using the quality of reconstruction as a metric of reidentification results in exaggerated reidentification risk figures.

Using Saturated Count Models for User-Friendly Synthesis of Large Confidential Administrative Databases

AbstractOver the past three decades, synthetic data methods for statistical disclosure control have continually evolved, but mainly within the domain of survey data sets. There are certain characteristics of administrative databases, such as their size, which present challenges from a synthesis perspective and require special attention. This paper, through the fitting of saturated count models, presents a synthesis method that is suitable for administrative databases. It is tuned by two parameters, σ and α. The method allows large categorical data sets to be synthesized quickly and allows risk and utility metrics to be satisfied a priori, that is, prior to synthetic data generation. The paper explores how the flexibility afforded by two-parameter count models (the negative binomial and Poisson-inverse Gaussian) can be utilised to protect respondents'—especially uniques'—privacy in synthetic data. Finally, an empirical example is carried out through the synthesis of a database which can be viewed as a good substitute to the English School Census.

Vine copula statistical disclosure control for mixed-type data

In this paper, we develop a new statistical disclosure control (SDC) method for mixed-type data based on vine copulas. The use of Gaussian and skew-t copulas has been demonstrated to be capable of incorporating information from the marginal distributions of mixed-type variables, whether they are discrete or continuous. In particular, our proposed SDC method using vine copulas generalizes a data perturbation method using an extended skew-t copula. Our vine-SDC method improves the SDC method using the extended skew-t copula by allowing the bivariate copulas in the vine decomposition to take various forms, thus offering a better fit for the joint distribution of the data and more flexibility in data perturbation. An additional advantage of our vine-SDC method is the significant improvement in computational efficiency compared with that using the extended skew-t copula. We discuss some statistical properties of vine copulas and the methodology of vine-SDC. A simulation and a study of real healthcare survey data are provided to explore the performance and strength of vine-SDC and compare it with a common copula-based SDC method.

Statistical disclosure control for continuous variables using an extended skew‐t copula

AbstractIn this article, we extend the skew‐t data perturbation (STDP) to develop a new statistical disclosure control (SDC) method for data with continuous variables. In this new SDC method, we construct an extended skew‐t (EST) copula to release confidential data for third‐party usage. Using the EST copula for producing perturbed data, we can incorporate rich statistical information in the perturbed data while preserving the marginal distributions of the data. An advancement of this EST‐SDC method is to use a copula distribution, which allows generation of perturbed data from bivariate conditional EST copulas sequentially. We discuss the methodology of EST‐SDC and outline some statistical properties derived from copula theories. Simulations and a real data study are included to demonstrate how the EST‐SDC method can be applied and to compare with the STDP method.

Statistical Disclosure Control Methods for Harmonised Protection of Census Data: a Grid Case

The 2011 Population and Housing Census in the Czech Republic was accompanied by a significant change in the technology used to prepare course of the fieldwork, along with changes in how the data are processed and how the outputs are disseminated. Grids are regular polygon networks that divide the territory of country in a grid-like way/pattern into equally large territorial units, to which aggregate statistical data are assigned. The disadvantage of grids is that these are territorially small units that are often minimally populated. This mainly has implications for the protection of individual data, which is associated with statistical disclosure control (SDC). The research question addressed in this paper is whether data protection (perturbation methods) leads to a change in the characteristics of the file either in terms of statistics of the whole file (i.e. for all grids) or in terms of spatial statistics, which indicate the spatial distribution of the analysed phenomenon. Two possible solutions to the issue of grid data protection are discussed. One comes from the Statistical Office of the European Communities (Eurostat) and the other from Cantabular, which is a product of the Sensible Code Company (SCC) based in Belfast. According to the Cantabular methodology, one variant was processed, while according to the Eurostat methodology, two variants were calculated, which differ by the parameter settings for maximum noise D and the variance of noise V. The results of the descriptive statistics show a difference in absolute differences when Cantabular and Europstat solutions are compared. In the case of other statistics, the results are fully comparable. This paper is devoted to one specific type of census output. The question is to what extent these results are relevant for other types of census outputs. They differ fundamentally in the number of dimensions (grids have only two dimensions). It would therefore be appropriate to use SDC procedures that allow greater flexibility in defining SDC parameters.

Reliability of Supervised Machine Learning Using Synthetic Data in Health Care: Model to Preserve Privacy for Data Sharing.

BackgroundThe exploitation of synthetic data in health care is at an early stage. Synthetic data could unlock the potential within health care datasets that are too sensitive for release. Several synthetic data generators have been developed to date; however, studies evaluating their efficacy and generalizability are scarce.ObjectiveThis work sets out to understand the difference in performance of supervised machine learning models trained on synthetic data compared with those trained on real data.MethodsA total of 19 open health datasets were selected for experimental work. Synthetic data were generated using three synthetic data generators that apply classification and regression trees, parametric, and Bayesian network approaches. Real and synthetic data were used (separately) to train five supervised machine learning models: stochastic gradient descent, decision tree, k-nearest neighbors, random forest, and support vector machine. Models were tested only on real data to determine whether a model developed by training on synthetic data can used to accurately classify new, real examples. The impact of statistical disclosure control on model performance was also assessed.ResultsA total of 92% of models trained on synthetic data have lower accuracy than those trained on real data. Tree-based models trained on synthetic data have deviations in accuracy from models trained on real data of 0.177 (18%) to 0.193 (19%), while other models have lower deviations of 0.058 (6%) to 0.072 (7%). The winning classifier when trained and tested on real data versus models trained on synthetic data and tested on real data is the same in 26% (5/19) of cases for classification and regression tree and parametric synthetic data and in 21% (4/19) of cases for Bayesian network-generated synthetic data. Tree-based models perform best with real data and are the winning classifier in 95% (18/19) of cases. This is not the case for models trained on synthetic data. When tree-based models are not considered, the winning classifier for real and synthetic data is matched in 74% (14/19), 53% (10/19), and 68% (13/19) of cases for classification and regression tree, parametric, and Bayesian network synthetic data, respectively. Statistical disclosure control methods did not have a notable impact on data utility.ConclusionsThe results of this study are promising with small decreases in accuracy observed in models trained with synthetic data compared with models trained with real data, where both are tested on real data. Such deviations are expected and manageable. Tree-based classifiers have some sensitivity to synthetic data, and the underlying cause requires further investigation. This study highlights the potential of synthetic data and the need for further evaluation of their robustness. Synthetic data must ensure individual privacy and data utility are preserved in order to instill confidence in health care departments when using such data to inform policy decision-making.

Statistical Disclosure Control Methods for Microdata from the Labour Force Survey

The aim of this article is to analyse the possibility of applying selected perturbative masking methods of Statistical Disclosure Control to microdata, i.e. unit‑level data from the Labour Force Survey. In the first step, the author assessed to what extent the confidentiality of information was protected in the original dataset. In the second step, after applying selected methods implemented in the sdcMicro package in the R programme, the impact of those methods on the disclosure risk, the loss of information and the quality of estimation of population quantities was assessed. The conclusion highlights some problematic aspects of the use of Statistical Disclosure Control methods which were observed during the conducted analysis.

General Confidentiality and Utility Metrics for Privacy-Preserving Data Publishing Based on the Permutation Model

Anonymization for privacy-preserving data publishing, also known as statistical disclosure control (SDC), can be viewed under the lens of the permutation model. According to this model, any SDC method for individual data records is functionally equivalent to a permutation step plus a noise addition step, where the noise added is marginal, in the sense that it does not alter ranks. Here, we propose metrics to quantify the data confidentiality and utility achieved by SDC methods based on the permutation model. We distinguish two privacy notions: in our work, anonymity refers to subjects and hence mainly to protection against record re-identification, whereas confidentiality refers to the protection afforded to attribute values against attribute disclosure. Thus, our confidentiality metrics are useful even if using a privacy model ensuring an anonymity level ex ante . The utility metric is a general-purpose metric that can be conveniently traded off against the confidentiality metrics, because all of them are bounded between 0 and 1. As an application, we compare the utility-confidentiality trade-offs achieved by several anonymization approaches, including privacy models ( $k$ k -anonymity and $\epsilon$ e -differential privacy) as well as SDC methods (additive noise, multiplicative noise and synthetic data) used without privacy models.

An Empirical Study of Applying Statistical Disclosure Control Methods to Public Health Research

Patient data or information collected from public health and health care surveys are of great research value. Usually, the data contain sensitive personal information. Doctors, nurses, or researchers in the public health and health care sector do not analyze the available datasets or survey data on their own, and may outsource the tasks to third parties. Even though all identifiers such as names and ID card numbers are removed, there may still be some occasions in which an individual can be re-identified via the demographic or particular information provided in the datasets. Such data privacy issues can become an obstacle in health-related research. Statistical disclosure control (SDC) is a useful technique used to resolve this problem by masking and designing released data based on the original data. Whilst ensuring the released data can satisfy the needs of researchers for data analysis, there is high protection of the original data from disclosure. In this research, we discuss the statistical properties of two SDC methods: the General Additive Data Perturbation (GADP) method and the Gaussian Copula General Additive Data Perturbation (CGADP) method. An empirical study is provided to demonstrate how we can apply these two SDC methods in public health research.

Feedback-Based Integration of the Whole Process of Data Anonymization in a Graphical Interface

The interactive, web-based point-and-click application presented in this article, allows anonymizing data without any knowledge in a programming language. Anonymization in data mining, but creating safe, anonymized data is by no means a trivial task. Both the methodological issues as well as know-how from subject matter specialists should be taken into account when anonymizing data. Even though specialized software such as sdcMicro exists, it is often difficult for nonexperts in a particular software and without programming skills to actually anonymize datasets without an appropriate app. The presented app is not restricted to apply disclosure limitation techniques but rather facilitates the entire anonymization process. This interface allows uploading data to the system, modifying them and to create an object defining the disclosure scenario. Once such a statistical disclosure control (SDC) problem has been defined, users can apply anonymization techniques to this object and get instant feedback on the impact on risk and data utility after SDC methods have been applied. Additional features, such as an Undo Button, the possibility to export the anonymized dataset or the required code for reproducibility reasons, as well its interactive features, make it convenient both for experts and nonexperts in R—the free software environment for statistical computing and graphics—to protect a dataset using this app.

Using the Complex Measure in an Assessment of the Information Loss Due to the Microdata Disclosure Control

The paper contains a proposal of original method of assessment of information loss resulted from an application of the Statistical Disclosure Control (SDC) conducted during preparation of the resulting data to the publication and disclosure to interested users. The SDC tools enable protection of sensitive data from their disclosure – both direct and indirect. The article focuses on pseudonimised microdata, i.e. individual data without fundamental identifiers, used for scientific purposes. This control is usually to suppress, swapping or disturbing of original data. However, such intervention is connected with the loss of some information. Optimization of choice of relevant SDC method requires then a minimization of such loss (and risk of disclosure of protected data). Traditionally used methods of measurement of such loss are not rarely sensitive to dissimilarities resulting from scale and scope of values of variables and cannot be used for ordinal data. Many of them weakly take also connections between variables into account, what can be important in various analyses. Hence, this paper is aimed at presentation of a proposal (having the source in papers by Zdzisław Hellwig) concerning use of a method of normalized and easy interpretable complex measure (called also the synthetic indicator) for connected features based on benchmark and anti–benchmark of development to the assessment of information loss resulted from an application of some SDC techniques and at studying its practical utility. The measure is here constructed on the basis of distances between original data and data after application of the SDC taking measurement scales into account.

Towards the adaptation of SDC methods to stream mining

Most of the existing statistical disclosure control (SDC) standards, such as k-anonymity or l-diversity, were initially designed for static data. Therefore, they cannot be directly applied to stream data which is continuous, transient, and usually unbounded. Moreover, in streaming applications, there is a need to offer strong guarantees on the maximum allowed delay between incoming data and its corresponding anonymous output. In order to full-fill with these requirements, in this paper, we present a set of modifications to the most standard SDC methods, efficiently implemented within the Massive Online Analysis (MOA) stream mining framework. Besides, we have also developed a set of performance metrics to evaluate Information Loss and Disclosure Risk values continuously. Finally, we also show the efficiency of our new methods with a large set of experiments.

Disclosure risk reduction for generalized linear model output in a remote analysis system

Remote analysis systems allow analysts to obtain statistical results without providing direct access to confidential data stored in a secure server system. An attacking analyst could send queries to a remote server to obtain outputs of statistical analyses and use those outputs for a disclosure attack. Statistical disclosure control (SDC) methods are used to modify remote analysis system (RAS) outputs in the protection of confidential information. Confidentiality protection through perturbation is one of the most commonly adopted SDC methods. In the case of generalized linear modelling, random noise is added to the estimated coefficients or to the associated estimating equation prior to getting estimates. This inflates the variances of estimators, and some efficiency and utility of estimators are lost. Thus the application of any perturbation based SDC method could result in an inefficient estimator, with the danger of producing worthless inferences. To date, little attention has been given to systematically controlling the disclosure risk and utility in SDC methods for RAS. In this paper, we develop a framework for the perturbation of estimating equations that enables an RAS to release modified generalized linear model output in such a way that the disclosure risk is not only reduced but also a good utility is maintained. Finally, we present some empirical results demonstrating the application of our framework for obtaining estimates from perturbed estimating equations of binary and count response models.

Measuring Disclosure Risk and Data Utility for Flexible Table Generators

Abstract Statistical agencies are making increased use of the internet to disseminate census tabular outputs through web-based flexible table-generating servers that allow users to define and generate their own tables. The key questions in the development of these servers are: (1) what data should be used to generate the tables, and (2) what statistical disclosure control (SDC) method should be applied. To generate flexible tables, the server has to be able to measure the disclosure risk in the final output table, apply the SDC method and then iteratively reassess the disclosure risk. SDC methods may be applied either to the underlying data used to generate the tables and/or to the final output table that is generated from original data. Besides assessing disclosure risk, the server should provide a measure of data utility by comparing the perturbed table to the original table. In this article, we examine aspects of the design and development of a flexible table-generating server for census tables and demonstrate a disclosure risk-data utility analysis for comparing SDC methods. We propose measures for disclosure risk and data utility that are based on information theory.

Statistical Disclosure Control for Micro-Data Using theRPackagesdcMicro

The demand for data from surveys, censuses or registers containing sensible information on people or enterprises has increased significantly over the last years. However, before data can be provided to the public or to researchers, confidentiality has to be respected for any data set possibly containing sensible information about individual units. Confidentiality can be achieved by applying statistical disclosure control (SDC) methods to the data in order to decrease the disclosure risk of data. The R package sdcMicro serves as an easy-to-handle, object-oriented S4 class implementation of SDC methods to evaluate and anonymize confidential micro-data sets. It includes all popular disclosure risk and perturbation methods. The package performs automated recalculation of frequency counts, individual and global risk measures, information loss and data utility statistics after each anonymization step. All methods are highly optimized in terms of computational costs to be able to work with large data sets. Reporting facilities that summarize the anonymization process can also be easily used by practitioners. We describe the package and demonstrate its functionality with a complex household survey test data set that has been distributed by the International Household Survey Network.

A minimum spanning tree equipartition algorithm for microaggregation

In this paper, we propose a solution on microaggregation problem based on the hierarchical tree equi-partition (HTEP) algorithm. Microaggregation is a family of methods for statistical disclosure control of microdata, that is, for masking microdata, so that they can be released without disclose private information on the underlying individuals. Knowing that the microaggregation problem is non-deterministic polynomial-time-hard, the goal is to partition N given data into groups of at least K items, so that the sum of the within-partition squared error is minimized. The proposed method is general and it can be applied to any tree partition problem aiming at the minimization of a total score. The method is divisive, so that the tree with the highest ‘score’ is split into two trees, resulting in a hierarchical forest of trees with almost equal ‘score’ (equipartition). We propose a version of HTEP for microaggregation (HTEPM), that is applied on the minimum spanning tree (MST) of the graph defined by the data. The merit of the HTEPM algorithm is that it solves optimally some instances of the multivariate microaggregation problem on MST search space in . Experimental results and comparisons with existing methods from literature prove the high performance and robustness of HTEPM.

A derivative‐free algorithm for refining numerical microaggregation solutions

AbstractThe biggest challenge in disclosing private data is to share information contained in databases, while protecting people from being individually identified. Microaggregation is a family of methods for statistical disclosure control. The principle of microaggregation states that confidentiality rules permit the publication of individual records if they are divided into groups of g or more data, where none of the records is more representative than the others in the same group. The application of such rules leads to replacing individual values by those computed from small groups, which are denoted as microaggregates, before data publication. The microaggregation procedure should be developed to reduce the information loss caused by this replacement process to an extent, so that valuable and accurate information could still be extracted from the microaggregated data. This work proposes a derivative‐free optimization algorithm for the numerical microaggregation problem based on the evaluation of a simplified surrogate function. Computational experiments show that the proposed method can considerably improve the results obtained by other heuristics found in the literature, as it improves some of its best‐known results.

Secure and efficient anonymization of distributed confidential databases

Let us consider the following situation: $$t$$ t entities (e.g., hospitals) hold different databases containing different records for the same type of confidential (e.g., medical) data. They want to deliver a protected version of this data to third parties (e.g., pharmaceutical researchers), preserving in some way both the utility and the privacy of the original data. This can be done by applying a statistical disclosure control (SDC) method. One possibility is that each entity protects its own database individually, but this strategy provides less utility and privacy than a collective strategy where the entities cooperate, by means of a distributed protocol, to produce a global protected dataset. In this paper, we investigate the problem of distributed protocols for SDC protection methods. We propose a simple, efficient and secure distributed protocol for the specific SDC method of rank shuffling. We run some experiments to evaluate the quality of this protocol and to compare the individual and collective strategies for solving the problem of protecting a distributed database. With respect to other distributed versions of SDC methods, the new protocol provides either more security or more efficiency, as we discuss through the paper.

Providing Data With High Utility And No Disclosure Risk For The Public and Researchers: An Evaluation By Advanced Statistical Disclosure Risk Methods

The demand of data from surveys, registers or other data sets containing sensibleinformation on people or enterprises have been increased significantly over the last years.However, before providing data to the public or to researchers, confidentiality has to berespected for any data set containing sensible individual information. Confidentiality canbe achieved by applying statistical disclosure control (SDC) methods to the data. Theresearch on SDC methods becomes more and more important in the last years because ofan increase of the awareness on data privacy and because of the fact that more and moredata are provided to the public or to researchers. However, for legal reasons this is onlyvisible when the released data has (very) low disclosure risk.In this contribution existing disclosure risk methods are review and summarized. Thesemethods are finally applied on a popular real-world data set - the Structural EarningsSurvey (SES) of Austria. It is shown that the application of few selected anonymisationmethods leads to well-protected anonymised data with high data utility and low informationloss.

Column generation bounds for numerical microaggregation

The biggest challenge when disclosing private data is to share information contained in databases while protecting people from being individually identified. Microaggregation is a family of methods for statistical disclosure control. The principle of microaggregation is that confidentiality rules permit the publication of individual records if they are partitioned into groups of size larger or equal to a fixed threshold value, where none is more representative than the others in the same group. The application of such rules leads to replacing individual values by those computed from small groups (microaggregates), before data publication. This work proposes a column generation algorithm for numerical microaggregation in which its pricing problem is solved by a specialized branch-and-bound. The algorithm is able to find, for the first time, lower bounds for instances of three real-world datasets commonly used in the literature. Furthermore, new best known solutions are obtained for these instances by means of a simple heuristic method with the columns generated.