Speculations about whether strategic voting made a difference to the outcome of an election regularly whip up the passions of pundits, party strategists, electoral reformers and scholars alike. Yet, research on strategic voting's political effect has been hampered by the scarcity of data on district level party preferences. We propose the use of Bayesian small area estimation to predict district level preferences from just a handful of survey responses per district and comparing these predictions against election results to estimate how many voters switched sides in each district. We apply the approach to estimate how many seats changed hands as a result of strategic voting at the 1997 and 2001 UK general elections. Despite similar rates of strategic voting in both elections, the number of seats that were affected was markedly greater in 1997. Interestingly, the Liberal Democrats turn out to win the most seats because of strategic voting. We also estimate how many votes went in the ‘wrong’ direction—away from otherwise viable candidates. We validate our results by using journalistic sources and compare them with previous published estimates.

We discuss the problem of random measurement error in two variables when using a cross-lagged panel design. We apply the problem to the question of the causal direction between socio-economic status and subjective health, known also as health selection *versus* social causation. We plot the bias of the ratio between the social causation and the health selection coefficient as a function of the degree of measurement error in subjective health and socio-economic status for different scenarios which might occur in practice. Using simulated data we give an example of a Bayesian model for the treatment of measurement error that relies on external information about the degree of measurement error.

This is a comparative study of the multiple ways of measuring dissimilarities between state sequences. The originality of the study is the focus put on the differences between sequences that are sociologically important when studying life courses such as family life trajectories or professional careers. These differences essentially concern the sequencing (the order in which successive states appear), the timing and the duration of the spells in successive states. The study examines the sensitivity of the measures to these three aspects analytically and empirically by means of simulations. Even if some distance measures underperform, the study shows that there is no universally optimal distance index, and that the choice of a measure depends on which aspect we want to focus on. From the review and simulation results, the paper derives guidelines to help the end user to choose the right dissimilarity measure for her or his research objectives. This study also introduces novel ways of measuring dissimilarities that overcome some flaws in existing measures.

Multilevel modelling is a popular approach for longitudinal data analysis. Statistical models conventionally target a parameter at the centre of a distribution. However, when the distribution of the data is asymmetric, modelling other location parameters, e.g. percentiles, may be more informative. We present a new approach, *M*-quantile random-effects regression, for modelling multilevel data. The proposed method is used for modelling location parameters of the distribution of the strengths and difficulties questionnaire scores of children in England who participate in the Millennium Cohort Study. Quantile mixed models are also considered. The analyses offer insights to child psychologists about the differential effects of risk factors on children's outcomes.

We use data from the four sweeps of the UK Millennium Cohort Study of children born at the turn of the 21st century to document the effect that poverty, and in particular persistent poverty, has on their cognitive development in their early years. Using structural equation modelling, we show that children born into poverty have significantly lower test scores at age 3, age 5 and age 7 years, and that continually living in poverty in their early years has a cumulative negative effect on their cognitive development. For children who are persistently in poverty throughout their early years, their cognitive development test scores at age 7 years are almost 20 percentile ranks lower than children who have never experienced poverty, even after controlling for a wide range of background characteristics and parental investment.

The paper uses data from the consumer expenditure surveys to demonstrate that the mode of collection is important for the analysis of consumption data. We first show that population figures obtained with diaries markedly differ from figures obtained by using recall questions. We then exploit multiple measurements of food expenditure to identify the effects of the mode of collection on the distribution of reported consumption. Finally, we show how to combine information from multiple reports to obtain a single measure of total expenditure in consumer expenditure surveys. The paper concludes by offering guidelines for empirical analyses based on these data, and by providing an application of the methods proposed to the measurement of inequality and wellbeing.

A new semiparametric and robust approach to small area estimation for discrete outcomes is proposed. The methodology represents an efficient and easily computed alternative to prediction by using a generalized linear mixed model and is based on an extension of *M*-quantile regression. In addition, two estimators of the prediction mean-squared error are described: one based on Taylor linearization and another based on the block bootstrap. The methodology proposed is applied to UK annual Labour Force Survey data for estimating the proportion of the unemployed in local authorities in the UK. The properties of estimators are further empirically assessed in model-based simulations.

A mismatch between the timescale of a structural vector auto-regressive model and that of the time series data used for its estimation can have serious consequences for identification, estimation and interpretation of the impulse response functions. However, the use of mixed frequency data, combined with a proper estimation approach, can alleviate the temporal aggregation bias, mitigate the identification issues and yield more reliable responses to shocks. The problems and possible remedy are illustrated analytically and with both simulated and actual data.

Numerous studies have investigated the relationship between the built environment and physical activity. However, these studies assume that these relationships are invariant over space. In this study, we introduce a novel method to analyse the association between access to recreational facilities and exercise allowing for spatial heterogeneity. In addition, this association is studied before and after controlling for crime, which is a variable that could explain spatial heterogeneity of associations. We use data from the Chicago site of the ‘Multi-ethnic study of atherosclerosis’ of 781 adults aged 46 years and over. A spatially varying coefficient tobit regression model is implemented in the Bayesian setting to allow for the association of interest to vary over space. The relationship is shown to vary over Chicago, being positive in the south but negative or null in the north. Controlling for crime weakens the association in the south with little change observed in northern Chicago. The results of this study indicate that spatial heterogeneity in associations of environmental factors with health may vary over space and deserve further exploration.

The paper reports results from the first ever study of the effect of short-term weather and long-term climate on self-reported life satisfaction that uses longitudinal data. We find robust evidence that day-to-day weather variation impacts self-reported life satisfaction. Utilizing two sources of variation in the cognitive complexity of satisfaction questions, we present evidence that weather effects arise because of the cognitive challenge of reporting life satisfaction. We do not detect a relationship between long-term climate and self-reported life satisfaction by using an individual fixed effects specification, which identifies climate impacts through individuals moving location.

A common problem is to compare two cross-sectional estimates for the same study variable taken on two different waves or occasions, and to judge whether the change observed is statistically significant. This involves the estimation of the sampling variance of the estimator of change. The estimation of this variance would be relatively straightforward if cross-sectional estimates were based on the same sample. Unfortunately, samples are not completely overlapping, because of rotations used in repeated surveys. We propose a simple approach based on a multivariate (general) linear regression model. The variance estimator proposed is not a model-based estimator. We show that the estimator proposed is design consistent when the sampling fractions are negligible. It can accommodate stratified and two-stage sampling designs. The main advantage of the approach proposed is its simplicity and flexibility. It can be applied to a wide class of sampling designs and can be implemented with standard statistical regression techniques. Because of its flexibility, the approach proposed is well suited for the estimation of variance for the European Union Statistics on Income and Living Conditions surveys. It allows us to use a common approach for variance estimation for the different types of design. The approach proposed is a useful tool, because it involves only modelling skills and requires limited knowledge of survey sampling theory.

We consider the estimation of the number of severely disabled people by using data from the Italian survey on ‘Health conditions and appeal to Medicare’. In this survey, disability is indirectly measured by using a set of categorical items, which consider a set of functions concerning the ability of a person to accomplish everyday tasks. Latent class models can be employed to classify the population according to different levels of a latent variable connected with disability. The survey is designed to provide reliable estimates at the level of administrative regions ‘*Nomenclature des unités territoriales statistiques*’, level 2), whereas local authorities are interested in quantifying the number of people who belong to each latent class at a subregional level. Therefore, small area estimation techniques should be used. The challenge is that the variable of interest is not observed. Adopting a full Bayesian approach, we base small area estimation on a latent class model in which the probability of belonging to each latent class changes with covariates and the influence of age is learnt from the data by using penalized splines. Demmler–Reinsch bases are shown to improve speed and mixing of Markov chain Monte Carlo chains used to simulate posteriors.

Large sporting events affect criminal behaviour via three channels: fan concentration, self-incapacitation and police displacement. I exploit information on football matches for London teams linked to detailed recorded crime data at the area level to estimate these effects empirically. I find that only property crime increases in the communities hosting matches but not violent offences. There is a negative away game attendance effect on crime which is due to voluntary incapacitation of potential offenders attending a match. Police displacement during home games increases property crime by 7 percentage points for every extra 10000 supporters in areas that are left underprotected.

The study of temporal and spatial trends in large databases, such as behavioural risk factor surveillance data, can be a great challenge, especially when the intent is to study the time-related effects of multiple independent variables; this is an issue which is not usually addressed in trend analysis in epidemiological studies. This study demonstrates the use of varying coefficient models using non-parametric techniques, which can show how coefficients vary in time or space; it is a useful statistical tool that is applied for the first time to health surveillance data. Using the US ‘Behavioral risk factor surveillance system’, a varying coefficient model is constructed using obesity as an outcome measure. Odds ratio plots and probability maps illustrate the temporal or spatial changes in coefficients of the independent variables; these results can be used to identify changes in at-risk subgroups of the population for the odds of obesity.

Statistics Netherlands applies a design-based estimation procedure to produce road transportation figures. Frequent survey redesigns caused discontinuities in these series which obstruct the comparability of figures over time. Reductions in the sample size and changes in the sample design resulted in variance breaks and unacceptably large sampling errors in the recent part of the series. Both problems are addressed and solved simultaneously. Discontinuities and small sample sizes are accounted for by using a multivariate structural time series model that borrows strength over time and space. The paper illustrates an increased precision when we move from univariate models to a multivariate model where the domains are jointly modelled. This increase is especially significant in the most recent period when sample sizes become smaller, with standard errors of the design-based estimator of the target variables being reduced by 40 & #x2013;70 & #x0025; with the model-based approach.

Passing the ball is one of the key skills of a football player yet the metrics commonly used to evaluate passing ability are crude and largely limited to various forms of a pass completion rate. These metrics can be misleading for two general reasons: they do not account for the difficulty of the attempted pass nor the various levels of uncertainty involved in empirical observations based on different numbers of passes per player. We address both these deficiencies by building a statistical model in which the success of a pass depends on the skill of the executing player as well as other factors including the origin and destination of the pass, the skill of his teammates and the opponents, and proxies for the defensive pressure put on the executing player as well as random chance. We fit the model by using data from the 2006–2007 season of the English Premier League provided by Opta, estimate each player's passing skill and make predictions for the next season. The model predictions considerably outperform a naive method of simply using the previous season's completion rate as a predictor of the following season's completion rate. In particular, we show how a change in the difficulty of passes attempted in both seasons explains a significant proportion of the shift in the observed performance of some players—a fact that is ignored if the raw completion rate is used to evaluate player skill.

We examine the average cost function for property and casualty insurers. The cost function describes the relationship between a firm's minimum production cost and outputs. A comparison of cost functions could shed light on the relative cost efficiency of individual firms, which is of interest to many market participants and has been given extensive attention in the insurance industry. To identify and to compare the cost function, current practice is to assume a common functional form between costs and outputs across insurers and then to rank insurers according to the centre of the cost distribution. However, the assumption of a common cost–output relationship could be misleading because insurers tend to adopt different technologies that are reflected by the cost function in their production process. The centre-based comparison could also lead to biased inference especially when the cost distribution is skewed with a heavy tail. To address these issues, we model the average production cost of insurers by using a Bayesian quantile regression approach. Quantile regression enables the modelling of different quantiles of the cost distribution as opposed to just the centre. The Bayesian approach helps to estimate the cost-to-output functional relationship at a firm level by borrowing information across firms. In the analysis of US property–casualty insurers, we show that better insights into efficiency are gained by comparing different quantiles of the cost distribution.

During redesigns of repeated surveys, the old and new approaches are often conducted in parallel to quantify discontinuities that are initiated by modifications in the survey process. For budget limitations, the sample size allocated to the alternative approach is often considerably smaller compared with the regular survey that is used for official publication. In this paper, small area estimation techniques are considered to improve the accuracy of domain estimates obtained under the alternative approach. Besides auxiliary information that is available from administrations, direct domain estimates available from the regular survey are useful auxiliary variables to construct model-based small area estimators. These methods are applied to a redesign of the Dutch Crime Victimization Survey.

In line with recent developments in the statistical analysis of functional data, we develop the semiparametric functional auto-regressive modelling approach to the density forecasting analysis of national rates of inflation by using sectoral inflation rates in the UK over the period January 1997–September 2013. The pseudo-out-of-sample forecasting evaluation and test results provide an overall support to superior performance of our proposed models over the aggregate auto-regressive models and their statistical validity. The fan chart analysis and the probability event forecasting exercise provide further support for our approach in a qualitative sense, revealing that the modified functional auto-regressive models can provide a complementary tool for generating the density forecast of inflation, and for analysing the performance of a central bank in achieving announced inflation targets. As inflation targeting monetary policies are usually set with recourse to the medium-term forecasts, our proposed work may provide policy makers with an invaluably enriched information set.

The statistical analysis of observational data for fair lending purposes relies on the assumption that, at the firm level, racial discrimination (or the lack thereof) is stable across time. Using data from a mortgage lender during the period 1998–2006, we examine this crucial assumption for the case of pricing differentials for black applicants in household mortgage lending, effectively evaluating possible dynamics in aggregate discrimination patterns. We offer evidence that these estimated pricing differentials may vary substantially across time.

The rise of data science could be seen as a potental threat to the long-term status of the statistics discipline. I first argue that, although there is a threat, there is also a much greater opportunity to re-emphasize the universal relevance of statistical method to the interpretation of data, and I give a short historical outline of the increasingly important links between statistics and information technology. The core of the paper is a summary of several recent research projects, through which I hope to demonstrate that statistics makes an essential, but incomplete, contribution to the emerging field of ‘electronic health’ research. Finally, I offer personal thoughts on how statistics might best be organized in a research-led university, on what we should teach our students and on some issues broadly related to data science where the Royal Statistical Society can take a lead.

We compare three major UK surveys, the British Household Panel Survey, Family Resources Survey and the English Longitudinal Study of Ageing, in terms of the picture that they give of the relationship between disability and receipt of the *attendance allowance* benefit. Using the different disability indicators that are available in each survey, we use a structural equation approach involving a latent concept of disability in which probabilities of receiving attendance allowance depend on disability. Despite major differences in design, once sample composition has been standardized through statistical matching, the surveys deliver similar results for the model of disability and receipt of attendance allowance. Provided that surveys offer a sufficiently wide range of disability indicators, the detail of disability measurement appears relatively unimportant.

The paper develops a method for producing current quarter forecasts of gross domestic product growth with a (possibly large) range of available within-the-quarter monthly observations of economic indicators, such as employment and industrial production, and financial indicators, such as stock prices and interest rates. In light of existing evidence of time variation in the variances of shocks to gross domestic product, we consider versions of the model with both constant variances and stochastic volatility. We use Bayesian methods to estimate the model, to facilitate providing shrinkage on the (possibly large) set of model parameters and conveniently generate predictive densities. We provide results on the accuracy of nowcasts of realtime gross domestic product growth in the USA from 1985 through 2011. In terms of point forecasts, our proposal improves significantly on auto-regressive models and performs comparably with survey forecasts. In addition, it provides reliable density forecasts, for which the stochastic volatility specification is quite useful.

The paper investigates the relationship between fertility and women's education in Italy, using data from the 2009 Household Multipurpose Survey of Family and Social Subjects. We use event history models, adopting a Bayesian approach for inference to study the association between fertility and women's education in the presence of a time varying unobserved component. Our analysis shows that either disregarding the unobserved component or assuming a time constant unobserved heterogeneity can lead to misleading results, at least in the context studied.

We analyse whether female athletes differ from male athletes in their risk-taking behaviour in a competitive setting. Data from high jump and pole vault competitions allow us to identify risky strategies. We estimate whether female athletes use risky strategies as often as male athletes and whether or not their returns to risky strategies differ. Returns to risky strategies are identified via an instrumental variable approach where we use competitive pressure to instrument individual risk taking. Female athletes take fewer risky decisions than men and could improve their outcomes by incurring more risk. We show that competitive pressure results in more risky decisions by both men and women; however, men react stronger to competitive pressure.

New cervical cancer screening guidelines in the USA and many European countries recommend that women are tested for human papilloma virus (HPV). To inform decisions about screening intervals, we calculate the increase in precancer or cancer risk per year of continued HPV infection. However, both time to onset of precancer or cancer and time to HPV clearance are interval censored, and onset of precancer or cancer strongly informatively censors HPV clearance. We analyse these bivariate informatively interval-censored data by developing a novel joint model for time to clearance of HPV and time to precancer or cancer by using shared random effects, where the estimated mean duration of each woman's HPV infection is a covariate in the submodel for time to precancer or cancer. The model was fitted to data on 9553 HPV positive and negative women undergoing cervical cancer screening at Kaiser Permanente Northern California: data that were pivotal to the development of US screening guidelines. We compare the implications for screening intervals of this joint model with those from population-average marginal models of precancer or cancer risk. In particular, after 2 years the marginal population-average precancer or cancer risk was 5%, suggesting a 2-year interval to control *population-average* risk at 5%. In contrast, the joint model reveals that almost all women exceeding 5% individual risk in 2 years also exceeded 5% in 1 year, suggesting that a 1-year interval is better to control *individual* risk at 5%. The example suggests that sophisticated risk models that can predict individual risk may have implications that are different from those of population-average risk models that are currently used for informing medical guideline development.

Consistent negative correlations between sibship size and cognitive performance (as measured by intelligence quotient and other mental aptitude tests) have been observed in past empirical studies. However, parental decisions on family size may correlate with variables affecting child cognitive performance. The aim of this study is to demonstrate how selection bias in studies of sibship size effects can be adjusted for. We extend existing knowledge in two aspects: as factors affecting decisions to increase family size may vary across the number and composition of current family size, we propose a sequential probit model (as opposed to binary or ordered models) for the propensity to increase family size; to disentangle selection and causality we propose multilevel multiprocess modelling where a continuous model for performance is estimated jointly with a sequential probit model for family size decisions. This allows us to estimate and adjust for the correlation between unmeasured heterogeneity affecting both family size decisions and child cognitive performance. The issues are illustrated through analyses of scores on Peabody individual achievement tests among children of the US National Longitudinal Survey of Youth 1979. We find substantial between-family heterogeneity in the propensity to increase family size. Ignoring such selection led to overestimation of the negative effects of sibship size on cognitive performance for families with 1–3 children, when known sources of selection were accounted for. However, the multiprocess modelling proposed could efficiently identify and control for such bias due to adverse selection.

This study compares the extent of selection error (non-response and coverage error) evoked by the four major contemporary modes of data collection (face to face, telephone, mail and Web) and three sequential mixed mode designs (telephone, mail and Web with face-to-face follow-up) for the case of the Dutch Crime Victimization Survey. Sociodemographic characteristics and target variables from the survey serve as benchmark variables. A special two-wave experimental design allows studying design differences in selection error on Crime Victimization Survey variables independently from differences in measurement error. Despite large differences in response rates, only small or no differences in selection error between the four single-mode designs are found on both types of variable. We observe cases when the error is enlarged or mitigated in the mixed mode designs despite the fact that the designs yielded large response increases. Our results question the use of response rates to motivate the choice of mode and use of mixed mode surveys.

Organizations disseminate statistical summaries of administrative data via the Web for unrestricted public use. They balance the trade-off between protection of confidentiality and quality of inference. Recent developments in disclosure avoidance techniques include the incorporation of synthetic data, which capture the essential features of underlying data by releasing altered data generated from a posterior predictive distribution. The US Census Bureau collects millions of interrelated time series microdata that are hierarchical and contain many 0s and suppressions. Rule-based disclosure avoidance techniques often require the suppression of count data for small magnitudes and the modification of data based on a small number of entities. Motivated by this problem, we use zero-inflated extensions of Bayesian generalized linear mixed models with privacy-preserving prior distributions to develop methods for protecting and releasing synthetic data from time series about thousands of small groups of entities without suppression based on the magnitudes or number of entities. We find that, as the prior distributions of the variance components in the Bayesian generalized linear mixed model become more precise towards zero, protection of confidentiality increases and the quality of inference deteriorates. We evaluate our methodology by using a strict privacy measure, empirical differential privacy and a newly defined risk measure, the probability of range identification, which directly measures attribute disclosure risk. We illustrate our results with the US Census Bureau's quarterly workforce indicators.

The original version of Bayesian reconstruction, which is a method for estimating age-specific fertility, mortality, migration and population counts of the recent past with uncertainty, produced estimates for female-only populations. Here we show how two-sex populations can be similarly reconstructed and probabilistic estimates of various sex ratio quantities obtained. We demonstrate the method by reconstructing the populations of India from 1971 to 2001, Thailand from 1960 to 2000 and Laos from 1985 to 2005. We found evidence that, in India, the sex ratio at birth exceeded its conventional upper limit of 1.06, and, further, increased over the period of study, with posterior probability above 0.9. In addition, almost uniquely, we found evidence that life expectancy at birth, , was lower for females than for males in India (posterior probability for 1971–1976 equal to 0.79), although there was strong evidence for a reversal of the gap through to 2001. In both Thailand and Laos, we found strong evidence for the more usual result that was greater for females and, in Thailand, that the difference increased over the period of study.

Two-phase study designs are appealing since they allow for the oversampling of rare subpopulations, which improves efficiency. We describe a Bayesian hierarchical model for the analysis of two-phase data. Such a model is particularly appealing in a spatial setting in which random effects are introduced to model between-area variability. In such a situation, one may be interested in estimating regression coefficients or, in the context of small area estimation, in reconstructing the population totals by strata. The gains in efficiency of the two-phase sampling scheme are compared with standard approaches by using 2011 birth data from the research triangle area of North Carolina. We show that the method proposed can overcome small sample difficulties and improve on existing techniques. We conclude that the two-phase design is an attractive approach for small area estimation.

We propose a model-based strategy for ranking scientific journals starting from a set of observed bibliometric indicators that represent imperfect measures of the unobserved ‘value’ of a journal. After discretizing the available indicators, we estimate an extended latent class model for polytomous item response data and use the estimated model to cluster journals. We illustrate our approach by using the data from the Italian research evaluation exercise that was carried out for the period 2004–2010, focusing on the set of journals that are considered relevant for the subarea statistics and financial mathematics. Using four bibliometric indicators (IF, IF5, AIS and the *h*-index), some of which are not available for all journals, and the information contained in a set of covariates, we derive a complete ordering of these journals. We show that the methodology proposed is relatively simple to implement, even when the aim is to cluster journals into a small number of ordered groups of a fixed size. We also analyse the robustness of the obtained ranking with respect to different discretization rules.

Knowledge of the current state of the art in information and communication technology of businesses (ICTB) is an important issue for governments, markets and policy makers, because information technology improves access to information and plays an important role in firms' competitiveness. Statistical agencies use normalized surveys to provide harmonized statistics about the use of technology in enterprises. Classical design-based estimators are appropriate for large domains, because direct estimates are consistent and easy to obtain by using sampling weights. However, to supply estimates for unplanned domains, where the sample size is random, model-based estimators are usually required. In this paper, alternative logistic model-based estimators are suggested to derive small area estimates from ICTB surveys. Final estimates are benchmarked to achieve coherence with direct estimates in larger domains, and standard errors are given by using bootstrap techniques. A Monte Carlo simulation study is conducted to compare the performance of the small area estimators proposed and to evaluate the behaviour of the mean-squared error estimator. Results are illustrated with the 2010 ICTB survey of the Basque country (Spain).

More than 1100 abandoned mines, milling sites and waste piles from the uranium mining period are scattered across the Navajo Nation, resulting in exposures to environmental metals, including uranium. The Diné Network for Environmental Health project began in response to concerns regarding the community health effects of these environmental exposures on chronic disease. The paper presents the results of the initial Diné Network for Environmental Health survey of 1304 individuals living on the Navajo Nation. We examine the relationship between uranium mine waste exposure and kidney disease, diabetes and hypertension. These chronic diseases are found at high prevalences in the study population, present major public health risks and have been linked to metals exposures in other studies. We model the exposure–outcome relationship by using a multivariate model for the three binary responses. We implement a Bayesian multivariate *t*-model, which has marginal log-odds ratio parameter interpretations and is computationally efficient. In examining environmental exposures, appropriately adjusting for potential confounders is pivotal to obtaining policy relevant effect estimates. We use Bayesian model averaging to account for uncertainty in the functional form for confounding adjustment within a small set of measured confounders. Using this multivariate framework, we find evidence of associations between these chronic diseases and both historic mining era and legacy mining exposures.