Search

This article is an introduction to the 13 articles in the JSSAM special issue on Privacy, Confidentiality, and Disclosure Protection. We also provide background information to place the articles into context.

Record linkage aims to bring records together from two or more files that belong to the same statistical entity. Naively treating a linked file as if there are no linkage errors may lead to biased inference. We present two general approaches for compensating for linkage error when calculating and analysing a two-way contingency table for categorical data, and study the following question: under what conditions can a compensation approach improve on the naive approach, where linkage error is not compensated for? To this end, we compare estimation errors, bias, variance and mean square error for the naive approach and two compensation approaches by means of an analytical study as well as a simulation study.

Web surveys have become increasingly popular over the last decade, but they tend to suffer from breakoffs, which take place when respondents start the survey but do not complete it. Many studies have investigated the factors impacting breakoffs, but they often ignored the breakoff timing and gave scant attention to two factors: question topics and filter question formats (grouped vs. interleafed as defined by whether filter questions are presented upfront or not). Using survival analysis, this study first identifies when breakoffs are more likely to happen and what are the time-varying predictors of breakoffs. Then, by using a web survey that experimentally manipulates the filter question format and randomly orders the question topic, this study investigates the effect of question topics and filter question formats on the breakoff event and its timing. We find that most breakoffs tend to happen at the beginning of the questionnaire and at the place where a new question topic is introduced. While item nonresponse is associated with more breakoffs, it is surprising to see that open-ended and long questions are associated with a lower breakoff risk. Additionally, we discover that grouping the filter questions leads to fewer breakoffs at the beginning compared to the interleafed counterpart, but the breakoff risk in the grouped format catches up quickly when respondents realise their previous answers will trigger more questions. This study also shows that questions about insurance have more breakoffs while questions on demographics and income have fewer breakoffs despite their sensitivity level.

We review the influential research carried out by Chris Skinner in the area of statistical disclosure control, and in particular quantifying the risk of re-identification in sample microdata from a random survey drawn from a finite population. We use the sample microdata to infer population parameters when the population is unknown, and estimate the risk of re-identification based on the notion of population uniqueness using probabilistic modelling. We also introduce a new approach to measure the risk of re-identification for a subpopulation in a register that is not representative of the general population, for example a register of cancer patients. In addition, we can use the additional information from the register to measure the risk of re-identification for the sample microdata. This new approach was developed by the two authors and is published here for the first time. We demonstrate this approach in an application study based on UK census data where we can compare the estimated risk measures to the known truth.

Web surveys are becoming increasingly popular but tend to have more breakoffs compared to the interviewer-administered surveys. Survey breakoffs occur when respondents quit the survey partway through. The Cox survival model is commonly used to understand patterns of breakoffs. Nevertheless, there is a trend to using more data-driven models when the purpose is prediction, such as classification machine learning models. It is unclear in the breakoff literature what are the best statistical models for predicting question-level breakoffs. Additionally, there is no consensus about the treatment of time-varying question-level predictors, such as question response time and question word count. While some researchers use the current values, others aggregate the value from the beginning of the survey. This study develops and compares both survival models and classification models along with different treatments of time-varying variables. Based on the level of agreement between the predicted and actual breakoff, we find that the Cox model and gradient boosting outperform other survival models and classification models respectively. We also find that using the values of time-varying predictors concurrent to the breakoff status is more predictive of breakoff, compared to aggregating their values from the beginning of the survey, implying that respondents’ breakoff behaviour is more driven by the current response burden.

This paper investigated the extent to which parental socioeconomic status was associated with life course socioeconomic status heterogeneity between adult cohort members of the 1958 National Child Development Study and how this association differed depending on methods used to address longitudinal missing data. We compared three variants of the full information maximum likelihood approach, namely available case, complete case and partially observed case and two methods designed to compensate for missing at random data, namely multilevel multiple imputation and multiple imputation chained equations. Our results highlighted the important contribution of parental socioeconomic status in explaining the divergence in achieved socioeconomic status over the adult life course, how the available case approach increasingly overestimated socioeconomic attainment as age increased and survey sample size decreased and how the complete case approach downwardly biased the effect of parental socioeconomic status on adult socioeconomic status.

For decades, criminologists have been aware of the severe consequences of the dark figure of police records for crime prevention strategies. Crime surveys are developed to address the limitations of police statistics as crime data sources, and estimates produced from surveys can mitigate biases in police data. This paper produces small area estimates of crimes unknown to the police at local and neighbourhood levels from the Crime Survey for England and Wales to explore the geographical inequality of the dark figure of crime. The dark figure of crime is larger not only in small cities that are deprived but also in wealthy municipalities. The dark figure is also larger in suburban, low-housing neighbourhoods with large concentrations of unqualified citizens, immigrants and non-Asian minorities. ESRC Data CDRC 260 ES/L011840/1 ES/L011891/1 UK Data Service 108913 SN 7280

The requirement to anonymise data sets that are to be released for secondary analysis should be balanced by the need to allow their analysis to provide efficient and consistent parameter estimates. The proposal in this article is to integrate the process of anonymisation and data analysis. The first stage uses the addition of random noise with known distributional properties to some or all variables in a released (already pseudonymised) data set, in which the values of some identifying and sensitive variables for data subjects of interest are also available to an external ‘attacker’ who wishes to identify those data subjects in order to interrogate their records in the data set. The second stage of the analysis consists of specifying the model of interest so that parameter estimation accounts for the added noise. Where the characteristics of the noise are made available to the analyst by the data provider, we propose a new method that allows a valid analysis. This is formally a measurement error model and we describe a Bayesian MCMC algorithm that recovers consistent estimates of the true model parameters. A new method for handling categorical data is presented. The article shows how an appropriate noise distribution can be determined.

Collecting biological data in representative surveys is becoming more common due to their potential to inform research and policy. Nevertheless, using nurses to collect these data can lead to unintended effects. In this paper, we investigate how nurses influence the non-response process by looking at five waves of data coming from two surveys in the UK: Understanding Society and the English Longitudinal Study of Ageing. We find that nurses explain between 5 and 14 percent of the variance in non-response to biological data collection. We also find that older nurses are more successful in obtaining cooperation and consent to biological data collection and nurses with more survey experience are more likely to successfully collect blood samples. Finally, we show that including nurse characteristics in non-response weighting models leads to modest changes in population estimates of biological markers.

Research that compares those who do and do not participate in protest over time purports that protesters are becoming increasingly similar to the non-protesting population. Using a protest survey that includes the frequency of protest participation, we consider the extent to which those who protest to different degrees are similar to non-protesters. Selection bias in non-probability protest survey data is compensated for by combining the data with random reference samples from the European Social Survey under a quasi-randomisation approach. We test hypotheses on the normalization of protesters and compare two methods for compensating for selection bias: a proportional weighting method and a propensity score adjustment method. The propensity score adjustment method is more effective in mitigating selection bias by balancing on variables that explain the selection and outcome, and enables the comparison of groups of protesters to non-protesters. We find that protesters become increasingly differentiated from non-protesters according to their left-wing self-placement and education as their extent of protest participation increases.

We review two approaches for improving the response in longitudinal (birth cohort) studies based on response propensity models: strategies for sample maintenance in longitudinal studies and improving the representativeness of the respondents over time through interventions. Based on estimated response propensities, we examine the effectiveness of different re-issuing strategies using Representativity Indicators (R-indicators). We also combine information from the Receiver Operating Characteristic (ROC) curve with a cost function to determine an optimal cut point for the propensity not to respond in order to target interventions efficiently at cases least likely to respond. We use the first four waves of the UK Millennium Cohort Study to illustrate these methods. Our results suggest that it is worth re-issuing to the field nonresponding cases from previous waves although re-issuing refusals might not be the best use of resources. Adapting the sample to target subgroups for re-issuing from wave to wave will improve the representativeness of response. However, in situations where discrimination between respondents and nonrespondents is not strong, it is doubtful whether specific interventions to reduce nonresponse will be cost effective.

Background Although social inequalities in health are consistently observed across the lifecourse in different populations, there are methodological issues in describing whether these inequalities decrease in later life. The association between socioeconomic adversity and systemic inflammation is well documented in cross-sectional studies, however, the association between living in socioeconomic disadvantage and repeated systemic inflammation in older adults has not been examined in detail, particularly taking into account longitudinal patterns of missingness. Inference from longitudinal analyses of ageing populations is susceptible to biases arising from attrition and non-random dropout. The accumulation of drop-outs over several waves reduce the representativeness of the study population and certain subpopulations can be over or under-represented in the sample. Methods 4,574 men and women aged 52 years and older from the English Longitudinal Study of Ageing (ELSA) wave 2 onwards were analysed. ELSA is a prospective cohort study that is representative of the English population. C - reactive protein levels were measured in waves 2, 4, 6, and 8 (2004–2016). Latent growth curve models estimated the relationship between different measures of socioeconomic position (education, wealth, and social class) and C - reactive protein, compensating for missing data under different assumptions: complete case analysis, full information maximum likelihood, multiple imputation, Diggle-Kenward selection model, and pattern-mixture model. All models were adjusted for gender, age, ethnicity, and marital status. Results At baseline in wave 2, we found differences between the most and least affluent categories of socioeconomic position. Participants with foreign or no qualifications (0.16 log(mg/l), 95% CI 0.09–0.23), participants in the lowest wealth tertile (0.24 log(mg/l), 95% CI 0.16–0.32), and participants in manual occupations (0.13 log(mg/l), 95% CI 0.07–0.19) had increased levels of C-Reactive protein compared to the most advantaged categories of education, wealth, and social class. Although, C - reactive protein levels decreased in later waves, the differences between the most and least socioeconomic advantaged groups remained large. Furthermore, differences between the Diggle-Kenward and other methods for compensating for missing data suggest that the missing completely at random and missing at random analyses underestimated socioeconomic differences in C-Reactive protein. Conclusion This study demonstrates that living in socioeconomic disadvantage is associated with higher C - reactive protein levels over time and that the social discrepancies in health between the most and least affluent socioeconomic groups persist at older ages. It also highlights the importance of compensating for missingness in longitudinal studies with ageing participants who are susceptible to non-random drop out.

© 2019 The Authors. International Statistical Review © 2019 International Statistical Institute Factor analysis models are used in data dimensionality reduction problems where the variability among observed variables can be described through a smaller number of unobserved latent variables. This approach is often used to estimate the multidimensionality of well-being. We employ factor analysis models and use multivariate empirical best linear unbiased predictor (EBLUP) under a unit-level small area estimation approach to predict a vector of means of factor scores representing well-being for small areas. We compare this approach with the standard approach whereby we use small area estimation (univariate and multivariate) to estimate a dashboard of EBLUPs of the means of the original variables and then averaged. Our simulation study shows that the use of factor scores provides estimates with lower variability than weighted and simple averages of standardised multivariate EBLUPs and univariate EBLUPs. Moreover, we find that when the correlation in the observed data is taken into account before small area estimates are computed, multivariate modelling does not provide large improvements in the precision of the estimates over the univariate modelling. We close with an application using the European Union Statistics on Income and Living Conditions data.

This research examines the geographical distribution of perceived neighbourhood disorder in Manchester, England, by using small area estimates. Sample surveys are the main source of information to analyse perceived disorder. However, most surveys are only representative of large areas, and direct estimates may be unreliable at small area level. Small area estimation techniques borrow strength from related areas to produce reliable small area estimates. This research produces Spatial Empirical Best Linear Unbiased Predictor (SEBLUP) estimates, which account for spatially correlated random area effects, of perceived neighbourhood disorder from the Manchester Resident Telephone Survey. The highest levels of perceived disorder are found in the city centre and some Northern and Central-Eastern areas. Perceived disorder is higher in areas with higher population churn, income deprivation and crime. Small area estimation techniques are a potential tool to map perceived disorder.

We develop a non-parametric imputation method for item non-response based on the well-known hot-deck approach. The proposed imputation method is developed for imputing numerical data that ensure that all record-level edit rules are satisfied and previously estimated or known totals are exactly preserved. We propose a sequential hot-deck imputation approach that takes into account survey weights. Original survey weights are not changed, rather the imputations themselves are calibrated so that weighted estimates will equal known or estimated population totals. Edit rules are preserved by integrating the sequential hot-deck imputation with Fourier-Motzkin elimination which defines the range of feasible values that can be used for imputation such that all record-level edits will be satisfied. We apply the proposed imputation method under different scenarios of random and nearest-neighbour hot-deck on two data sets: an annual structural business survey and a synthetically generated data set with a large proportion of missing data. We compare the proposed imputation methods to standard imputation methods based on a set of evaluation measures.

© The Author(s) 2019. Worry about crime is known to be higher in some European regions than others. However, cross-national surveys, which are the main source of information to map worry about crime across Europe, are designed to be representative of large areas (countries), and regions often suffer from small and unrepresentative sample sizes. This research produces reliable model-based small area estimates of worry about crime at regional level from European Social Survey data, in order to map the phenomenon and examine its macro-level predictors. Model-based small area estimation techniques borrow strength across areas to produce reliable estimates of parameters of interest. Estimates of worry about crime are higher in most South and East European regions, in contrast to Northern and Central Europe.

© The Author(s) 2019. Small area estimation (SAE) plays a crucial role in the social sciences due to the growing need for reliable and accurate estimates for small domains. In the study of well-being, for example, policy makers need detailed information about the geographical distribution of a range of social indicators. We investigate data dimensionality reduction using factor analysis models and implement SAE on the factor scores under the empirical best linear unbiased prediction approach. We contrast this approach with the standard approach of providing a dashboard of indicators or a weighted average of indicators at the local level. We demonstrate the approach in a simulation study and a real data application based on the European Union Statistics for Income and Living Conditions for the municipalities of Tuscany.

This chapter provides an overview of data linkage for exploiting and combining information about the same entities across data sources. Data linkage can be deterministic (exact), where each matching variable needs to agree exactly to determine a correct match, or probabilistic, where users allow for errors in the matching variables and assign a probability of a correct match. Through classic decision theory, the chapter determines the set of matches (and non-matches) and provides a linked dataset for further analysis. The chapter also describes some recent advances in record linkage and concludes with some initial research on compensating for linkage errors in the analysis of linked data.