ADR UK Research Fellows: Methodological developments within administrative data research

Dr Katie Harron (UCL), Professor Pam Sonnenberg (UCL) , Professor Bianca Lucia De Stavola (UCL), Professor James Carpenter (London School of Hygiene & Tropical Medicine) and Professor Andrew Copas (UCL) aim to provide guidance on appropriate use of synthetic administrative data. Synthetic data – or artificial data that looks like the original data source without containing information on any ‘real’ individuals – preserves the statistical properties of the original data sources.

If widely shared, synthetic data could allow researchers to understand the structure of data sources, develop analysis plans and algorithms, and test different statistical models. This could be done in parallel to applying for access to linked administrative datasets, streamlining the research process. Final refinements and analyses could then be conducted on the real data.

This project will assess the feasibility, performance and comparability of a range of methods for generating synthetic data. The findings will demonstrate ways of making existing data more widely useable and improving access to new data linkages through advanced methods. Relevant guidelines for the generation of synthetic data in different settings (for example, for training or exploring data prior to applying for the real data) will also be developed.

UCL researchers Dr Richard Silverwood, Professor Lisa Calderwood, Professor Bianca De Stavola and Professor George Ploubidis are analysing linked datasets to better understand the methods for addressing missing data, data quality and residual confounding (incorrect findings that occur as result of variables that have not been measured).

Research-ready administrative datasets tend to be population-representative and large in size, but as the data is not collected specifically for research purposes, it often lacks information in key domains.

Meanwhile, the UK has a long and successful history of national longitudinal cohort studies in which the same group of individuals are followed over time, potentially for several decades. These cohort studies aim to be population-representative at initiation and collect rich information across a wide range of research areas. However, they inevitably suffer from missing data, particularly due to decreasing levels of participant response over time. Linkages between existing cohort studies and administrative data, with the potential to draw on the different strengths of each data source, have therefore become increasingly common over recent years.

The overall aim of this project is to investigate ways to leverage the strengths of each data source in mutually beneficial ways to address important questions around missing data, data quality and residual confounding. This will be done using existing data linkages in health and education with UCL Centre for Longitudinal Studies (CLS) cohorts.

Professor Imran Rasul (UCL) and Dr Monica Costa-Dias and Dr Sarah Cattan of the Institute of Fiscal Studies are using machine learning techniques to study young people’s life course trajectories in education and criminal behaviour.

They aim to understand how these trajectories start and evolve to help inform how policies can be best designed and targeted to prevent young people engaging in crime. To do so, they will identify different types of trajectory, describe the characteristics of individuals with different types of trajectory, and map their journeys through the education and criminal justice systems.

This project will then aim to advance the use of machine learning techniques to analyse UK administrative data. They will apply a type of methods called ‘sequence analysis’, which have mostly been applied to survey data, to cluster individuals in terms of their educational and offending trajectories. They will test the viability and usefulness of using these techniques in administrative data research. If successful, this will provide researchers with a tool to explore administrative data as a rich source of longitudinal information about individuals’ lives and contacts with public services.