This item is archived. Information presented here may be out of date.

Measuring errors in data

Natasha Bance

The Methodology Division leads on the development and support of statistical and survey methods for Office for National Statistics (ONS) and provides support for the wider Government Statitisical Service (GSS) and beyond.

As Principal Statistical Methodologists we jointly lead on demographic methods for the Census and Population Statistics Hub in Methodology. Our work is varied and includes, providing advice on and developing statistical methods to support the Population and Migration Statistics Transformation programme, mortality projections, and uncertainty measures for the mid-year population Estimates (MYEs).

We developed a framework to help us understand the complexity of Home Office administrative data. This has been used to evaluate other administrative datasets within ONS including the admin-based population estimates and RAPID data from the Department for Work and Pensions. This is a part of the standard administrative data toolkit in the Census and Population Statistics Hub in Methodology.

Our framework has been included in the book Measurement Error in Longitudinal Data (Cernat and Sakshaug, 2021) which examines how to understand and analyse change over time, as reflected in longitudinal data,  the context of data errors.  As contributors, we wrote the chapter “Error Framework for Longitudinally Linked Administrative Data”.

The chapter begins by describing ONS’ statistical transformation ambitions and the importance of understanding the statistical quality of administrative data. We then explain how operational and statistical quality are not the same thing and can conflict. We include a typology for classifying different types of longitudinal data.  We then introduce two frameworks for understanding data error; one for single data sources and one for multiple sources that are longitudinally linked. The chapter then discusses the different types of data error in turn, including ‘edge effects’ which are a particular feature of longitudinal administrative data.

We published Measurement Error in Longitudinal Data on 18 March 2021, it is available from Oxford University Press.

If you would like to know more about the research methods the Methodology Division are exploring or you are interested in working with us please contact .

We are especially interested in research relating to admin data and alternative sources.

Nicky Rogers and Dr Louisa Blackwell
Natasha Bance
Nicola is a Principal Statistical Methodologist in ONS Methodology, leading on demographic methods. Currently she is leading on developing new methods for measuring international migration during the COVID-19 pandemic, together with associated measures of statistical uncertainty.
Louisa is Project Lead for Integrated Statistical Design within ONS. Her role includes ensuring coherent design in the transformation of population and social statistics. The error framework was developed whilst working as joint head of demographic methods with Nicola.