Skip to main content

Social statistics

Social statistics is the branch of statistics devoted to the application and development of quantitative methods tailored to the social and human sciences. Focusing on human behaviour, social structures, and societal trends, it aims to answer questions about populations, relationships, and collective patterns in societies. At the heart of social statistics is the recognition that social-science data often come from complex, messy real-world sources — surveys, longitudinal studies, administrative records, or observational data across time and groups.

Members of our social statistics research group have experience in a range of social science disciplines, including demography, education, epidemiology, psychology and sociology. We work on a range of statistical methods for answering questions across the social and human sciences, such as those relating to out-of-sample prediction based on complex data, description of population relationships using data from surveys and other sources, or causal inference from experimental or observational studies using approaches such as regression discontinuity, interrupted time series and synthetic and negative control designs. Data in these applications are often complex, high-dimensional and challenging to analyse. We develop methods that can cope with this complexity, such as:

· multivariate analysis of high-dimensional data;

· analysis of clustered data with complex correlation structures such as multivariate longitudinal data and multiprocess survival data; detection of outliers;

· analysis of problems with missing data, drop-out, misclassification and measurement error;

· and dealing with non-informative missingness in the presence of time-varying confounding in causal inference; and combining data from multiple different sources.

We develop and employ various statistical frameworks, models, methods of estimation, and computational algorithms such as:

· different types of latent variable, mixture and random effects models for continuous and categorical variables;

· Gaussian processes;

· interpretable machine learning methods;

· marginal modelling;

· composite likelihood methods;

· models for dependence using reproducing kernel Hilbert space methods;

· Markov decision and reinforcement learning methods;

· Bayesian methods;

· and computationally efficient Markov Chain Monte Carlo and sequential Monte Carlo computational techniques to facilitate parameter estimation, statistical inference, model choice, and prediction.

Faculty

Wicher Bergsma - Professor

Yunxiao Chen - Assistant Professor

Sara Geneletti - Associate Professor

Kostas Kalogeropoulos - Associate Professor

Jouni Kuha - Professor

Irini Moustaki - Professor and Deputy Head of Department (Teaching)

Fiona Steele - Professor

Research students

Motonori Oka
Research interests: Statistics and machine/deep learning for the education and social sciences