UK Labour productivity estimates for 184 industries

By Josh Martin

When I worked on the official UK productivity statistics at the Office for National Statistics (ONS), I would regularly hear from users that the statistics were not detailed enough. They would often have an interest in a particular industry, perhaps as someone working there, or as a researcher interested in the properties of that industry. I would explain the limitations of the data, and direct them to the best higher-level option.

But it left me thinking – might it be possible to produce productivity estimates for more detailed industries than are available in official statistics? And more importantly, would the results be any good?

This led to a project with my co-author Cliodhna Taylor, who also previously worked on the official productivity statistics. We set out to produce UK productivity estimates with far greater industry detail, and to test their reliability. The result of that work is labour productivity estimates for 184 industries – more than doubling the available detail compared to official statistics. The methods and results are described in a new paper with Economic Statistics Centre of Excellence (ESCoE) and The Productivity Institute (TPI). Since we were satisfied with the quality of the estimates, the full dataset and calculations are also available from the TPI’s Productivity Data Lab.

How did we approach it?

We set ourselves some basic principles:

1. Transparency and reproducibility: We would use only publicly available data sources, such that the methods would be transparent and the results could be re-produced and easily updated. If you’re interested, a full replication package is available alongside the dataset.
   
   
2. Consistency with the National Accounts: As far as possible given the use of only publicly available data, we would ensure the estimates were consistent with the data and methods used in the National Accounts. This meant that our estimates could be compared with other official data, and productivity estimates for higher-level industries on a like-for-like basis.
   
   
3. Meaningful industry breakdowns: We would use an industry breakdown that we felt was sensible. That meant scrutinising each industry time series, and testing the overall quality of the dataset.

Our productivity measures are Gross Value Added (GVA) per filled job, so we needed data on GVA and jobs for each detailed industry. We started from detailed official datasets published by the ONS, and split up the industries further from there:

• For GVA, we started with the GDP(o) Low-Level Aggregates published by ONS, and split it down further using data from the Annual Business Survey.
  
  
• For jobs, we started from Workforce jobs statistics, and split it down further using data from the Business Register Employment Survey.

We had to make some adjustments and fill in some gaps in some cases. See the paper for a comprehensive discussion of the methods.

Balancing detail and quality in the estimates

An important question for us was whether the estimates would be of sufficient quality. We expected some drop in quality compared to official statistics, since we were working with smaller and more detailed industries – at this detailed level, the data can be more volatile, and the scope for error is larger. But what would be an acceptable level of quality?

Inspired by an approach used by researchers in the US, we considered the trade-off between granularity (the level of detail of the dataset) and the quality of the estimates. In official productivity statistics, this trade-off is clear: the smaller the industry, the more volatile the year-to-year estimates. When we lined up our new detailed estimates, we found the same pattern. In other words, the decline in quality (increase in volatility) was consistent with what we would expect, given the additional detail we included. We did a similar test looking at revisions, and found the same result.

What did we learn from productivity estimates for detailed industries?

It is well known that some industries are more productive than others. But the insights are limited when you have a smaller number of industries to compare. Using the 184 industries in our dataset, we found a very wide range of productivity levels.

Higher productivity industries

The top 10 most productive industries in 2023 were mostly capital-intensive industries, such as oil & gas extraction, energy generation and distribution, and water transport. Some detailed industries also appear near the top, such as leasing of intellectual property, renting and leasing of motor vehicles, and construction of electricity and telecoms utility projects.

Lower productivity industries

We also learnt which industries were less productive. While manufacturing is often regarded as a high productivity industry, there is a lot of variation within manufacturing which our dataset sheds light on. We found that about 10% of jobs in manufacturing are industries with below-average productivity.

Using this level of industry detail, we were also able to make estimates of the productivity levels for industries defined in the UK Government’s Industrial Strategy, which mostly wouldn’t have been possible from existing productivity statistics. Most of the Industrial Strategy sectors which we were able to estimate had above-average productivity.

What next?

We’re thinking about how we can improve the estimates in future. One option is to use data sources which aren’t publicly available, which might help with some limitations of our current methods. We also want to explore estimates in constant prices (i.e. adjusted for price inflation over time) so that we can study productivity growth for detailed industries too – at the moment, our estimates are only in current prices (i.e. nominal terms), so shouldn’t be used to calculate productivity growth.

We hope the dataset will be useful for others to use in their work. For instance, it would allow researchers to control for the level of industry productivity at a more detailed level, which might enable better estimates of other economic relationships. We would be excited to hear from anyone who uses the dataset.

Summary

This blog introduces a new dataset of labour productivity estimates for 184 industries in the UK, more than doubling the available detail compared to official statistics. The estimates appear to be of reasonable quality, given the increase in industry detail. The estimates shed new insights on which industries are high and low productivity, and can also be used for policy analysis.

The views expressed here are those of the authors, and should not be taken as the views of the ONS, the Bank of England or any of its committees.

This blog has been reproduced with agreement from the Economic Statistics Centre of Excellence. The original blog post can be found on ESCoE’s website.