Abstract

Quality is a core consideration of any statistical project. We need to understand the quality of our data, methods, and outputs to ensure they are fit for purpose, enable sound decision making, and provide transparent communication around what we produce. Many metrics and methods exist for measuring this quality, and exciting developments are made year on year across the international statistical community.

Our team at the Office for National Statistics researches and develops new methods for measuring administrative data quality. Over the past few years, we have applied multiple cutting-edge methods and linked up with international statisticians and academics conducting similar work. We have learned a lot from this collaboration and hope that sharing our experiences more widely will help others in the international community with their methodological and quality considerations.

In this session we will discuss these methods in more depth. Our methods span multiple phases of the admin data journey. At the input level, we will cover latent class modelling for estimating measurement error, and representativity indicators for measuring representativeness. At the methods / processing level, we will discuss Dempster Shafer Theory for measuring the uncertainty of values and variables derived from administrative data using rule-based methods. At output level, we will share our experiences of applying multiple imputation latent class modelling for producing estimates of categorical variables by classification, and surrounding uncertainty measures for the outputs. As well as sharing our methodology and outputs, we will discuss the benefits and limitations of applying these methods. We will outline the challenges we faced, and considerations for wider implementation. We will also discuss our practical and methodological recommendations for anyone interested in applying them, and how we think they could be further developed in the future.

We have also developed a suite of user-friendly admin data quality tools to support others in assessing quality across the data journey. In our session we will also give a brief overview of these tools, what their purpose is, and how to use them.