What the Productivity Lab learned at the ESCoE 2025 Conference

By William Sarsfield

From the 21st to the 23rd of May, I had the opportunity to attend the annual Economic Statistics Centre of Excellence (ESCoE) conference, hosted in partnership with the Office for National Statistics (ONS) at King’s College London. As someone with a computer science background who now spends most of their time working with regional productivity data, the event offered an incredibly rich and interdisciplinary look into how economic measurement is evolving – often with data at its core.

Here are some of the key takeaways and highlights from my time at the conference.

Pensions, Doctors, and Labour Supply

The opening plenary session was led by Carol Propper, who presented fascinating research on how pension reforms affected labour supply among senior doctors in the NHS. Using granular Electronic Staff Records payroll data, she showed that the 2015 pension reform – despite raising the retirement age and moving to a career-average model – led to a 17-19% increase in average pension entitlements, which induced a positive labour supply response.

It was a masterclass in combining policy change with administrative microdata to assess real-world outcomes. For those of us looking at regional productivity, the lesson was clear: labour market incentives matter deeply and can be quantified with the right data.

Lunch Conversations and Data Discoveries

At lunch, I had the chance to speak with Adam Haunch from ESCoE, who recommended checking out CompNet – a comparative productivity dataset for 17 European countries. While it doesn’t include the UK, the level of detail in firm-level productivity and performance indicators could provide helpful benchmarks or modelling inspiration for UK-focused work.

Data-Rich Explorations of the UK Economy

Across multiple sessions, ONS and ESCoE researchers unveiled a host of innovative new datasets and tools. A few that stood out:

Decision Maker Panel: A unique longitudinal survey of ~2,500 UK firms per month, capturing perceptions on inflation, uncertainty, and economic conditions. Several papers used this data to examine how firms are reacting to inflation and the media’s role in shaping expectations.
Public Sector Management Practices Survey (2023): A pilot comparing management quality between the public and private sectors, potentially useful for linking organisational productivity with outcomes in regional settings.
ONS’s Input-Output Analysis Tool: Marianthi Dunn introduced a new online interface that simplifies traditionally complex I/O tables, making them accessible for broader policy modelling – inspirational work and worth exploring further for anyone working with sectoral or regional economic data.
Environmental and Water Accounting: Multiple sessions, including work by Chris Jones, Paul Ekins, Alice Bartolini, and Silvia Ferrini explored how ecosystem services and water use can be integrated into national accounting frameworks using the System of Environmental-Economic Accounting (SEEA) Ecosystem Accounting (EA) framework. These efforts could help us better understand the regional economic value of natural capital, especially for areas with high environmental amenity.

Microdata Innovations: Geography, Jobs, and the Green Economy

Some of the most exciting contributions came from applied work using novel microdata sources:

Card spending data was used to cluster geographic markets based on consumer transaction patterns – a project led by researchers from the Competition and Markets Authority (CMA), ESCoE, and Swansea University. It’s not yet public, but the methodology could help redefine local retail markets beyond traditional boundaries.
Online job ads are now being scraped and categorised by occupation and region (ITL1) by ONS. This has big implications for understanding labour demand trends at a regional level – this kind of data would be valuable in the Productivity Data Lab as it is a similar indicator to those used in our ITL1 Scorecards.
Vehicle and housing responses to pollution pricing, especially London’s Ultra Low Emissions Zone (ULEZ), were modelled at postcode level using detailed microdata on registrations and transport use by the CMA. It was a powerful example of how policy effects ripple through housing markets, commuting and environmental outcomes.
Green Economy Measurement was nicely laid out by Chris Jones and Neil Wilson from the ONS. They introduced the Low Carbon and Renewable Energy Economy (LCREE) survey which tracks firm activity across 17 sustainability sectors, giving insight into how firms contribute to environmental goals along the product lifecycle.

From GDP to GDP-B

The second day’s plenary by Erik Brynjolfsson introduced GDP-B, a new approach to economic measurement that factors in the benefits of digital products and free services, moving beyond conventional GDP. It was encouraging to see serious progress toward capturing the value of the digital economy in official statistics – something that aligns closely with how I think about digital systems, user experience and data-driven innovation.

The TPI Stand – Evening Poster Session

A special thanks to everyone I had the opportunity to speak to during the evening poster session at the end of the 2nd day of the conference. Here I presented the Productivity Lab’s ITL3 scorecards and dashboards, giving insights into the data and discussing different facets of regional productivity. This was paired with a live demo of the regional productivity growth tool to give a deeper understanding of the differences in productivity between different regions across the UK.

Climate in the National Accounts

The final day began with Sébastien Roux’s talk on augmenting national accounts to better incorporate climate data, presented by France’s Insee. With growing pressure to integrate environmental externalities into official statistics, this work offered a glimpse into the future of economic measurement.

Machine Learning and Missing Data

Lastly, I was intrigued by work from Miriam Steurer and Sabrina Spiegel using random forest algorithms to fill gaps in real estate data for property price indices. As we face similar issues with missing regional productivity indicators, techniques like MICE (Multiple Imputation by Chained Equations) and tree-based methods could become part of our own toolkit. As someone who has studied a variety of machine learning techniques, it was exciting to see random forest algorithms being applied in a practical context to solve real-world data challenges.

Final Thoughts

For a data scientist immersed in regional productivity metrics, the 2025 ESCoE conference was an excellent experience. I came away not just with datasets and methods to explore, but with a clearer sense of how cross-disciplinary the field of economic measurement is and how valuable a technical background can be within it. Whether you’re measuring pensions, pollution, or productivity, the effective use of data is paramount – and this conference has given numerous examples of this.