Linked population data is advantageous for health and social research. In Wales, we are fortunate that the SAIL Databank allows researchers to combine a range of diverse datasets to support population-scale research for projects that provide public benefit.
Two of the most important sources of demographic information are the Welsh Demographic Service Dataset (WDSD) and Census 2021. In this blog Rhodri Johnson, Research Officer within the ADR Wales Health and Well-being research theme, and Mike Edwards who works in SAIL as Principal Data Linkage Architect, discuss the work that they’ve undertaken to better understand the two rich datasets, how linkage between them works and what users should look out for. The work also provides useful information for SAIL from a technical perspective, acting as a baseline for ongoing data linkage enhancements.
We’ve recently published work in the International Journal of Population Data Science (IJPDS), which examines how these two data sources link together within SAIL using an anonymised linking field (ALF), the approach used within SAIL for individual level data linkage. By understanding how complete and representative this linkage is, researchers can plan and build study cohorts with additional insight into expected linkage rates and understanding potential sources of bias.
So what are the two datasets, why would we use them together and why are they different? In short, WDSD provides longitudinal individual demographic and housing level address registration data, with Census providing a snapshot of data but with richer content relating to individuals. Either data source can be used as a proxy of a population spine.
Census data offers a rich source of individual and household level data, while WDSD provides continuously updated administrative healthcare-based information including GP address registration data. Combining such datasets is a common approach in SAIL, allowing researchers to create the most comprehensive possible picture of the Welsh population; collating the information to create rich study cohorts to attempt to answer numerous research questions. This work aims to allow researchers to understand how generalisable the data is when linked using the same point in time approach that we used.
Incomplete linkage can introduce bias if certain groups are less likely to be linked. It was with this in mind that we carried out our work using an observational, cross-sectional design within the SAIL Databank.
We compared the two datasets as at 21 March 2021, the official Census date, with an additional buffer (six months pre, and three months post census date in WDS) to allow for General Practice (GP) address registration updates. This aimed to allow a comparison of people recorded as living in Wales at the time of the Census.
It is important to note that these results represent linkage at a point in time. Researchers building longitudinal cohorts—using ever-linked data over longer periods—may observe higher linkage rates due to population mobility and ongoing updates in administrative systems.
Key findings at a glance
- The study examined how Census 2021 data linked with the Welsh Demographic Service Dataset (WDS) within the SAIL Databank.
- Some groups were less likely to link between WDSD and Census 2021, including people who were male, aged 75 or over, of Asian ethnicity, more recent migrants, people identifying as or not disclosing LGBTQ+ status, those separated/divorced/widowed, and individuals living in rented accommodation. The absence of these groups in the linkage has implications for cohort representativeness.
- Linkage was measured at a specific point in time (March 2021, allowing six months pre and post within WDS) — wider or longitudinal analyses may yield higher linkage rates.
- Future SAIL work will look at bespoke methods to enhance linkage quality and transparency.
These differences suggest that while overall linkage quality is high, some population subgroups remain under-represented. Researchers using these data should therefore be aware of possible bias when building study cohorts.
What next?
As part of ongoing improvement, Mike Edwards (and the data linkage) team are developing new methods to enhance linkage rates when creating cohorts.
Mike Edwards, Principal Data Linkage Architect from SAIL, said: “Our ongoing work includes improving individual- and address-matching approaches with both deterministic and probabilistic methodologies, qualitative and quantitative descriptors of quality, and methods for more transparency in the linkage pipeline.
“A key aspect to improving linkage quality and the confident use of linked data, is in opening dialogues between data owners, linkage practitioners, and the users of linked data, and we are always striving for inclusive collaboration to provide guidance and policy around best practices in producing and using linked datasets. This future work aims to support researchers by improving both linkage completeness and transparency, allowing for more accurate population analytics while maintaining the highest standards of privacy and governance.”
So what does this mean for SAIL?
This work provides practical guidance for anyone using Census 2021 and WDSD data in the SAIL Databank. It means that:
- When constructing cohorts, researchers can anticipate expected match rates between Census and administrative data.
- They can recognise when apparent “missingness” reflects expected linkage limitations rather than data quality problems.
- They can design analyses that account for differential linkage and interpret their findings with additional information available.
There are potential wider use cases of this work including improving general understanding of differences between routinely collected and survey data.
Ultimately, understanding where and why records do not link is important. Our work, and the study that we have published on it, offers researchers practical insights into the linkage between Census 2021 and WDSD, helping them to plan analyses and interpret findings with confidence. By making linkage patterns transparent, we are helping to ensure that population-wide analyses remain robust, representative, and trustworthy, which is central to the aims and mission of SAIL, and the ongoing work in Population Data Science at Swansea University to improve the usability and governance of administrative data for research.
Researchers can learn more about accessing Census 2021 and WDSD data within SAIL.
Read the full study and more about the work of the ADR Wales Health and Well-being research theme.