1. Main points

  • Based on new experimental methods, total public service productivity grew by an average of 0.2% per year between 1997 and 2019, with variation by service area, and periods of faster and slower growth over the period.  
  • Service area productivity growth rates ranged from negative 1.4% (public order and safety) to 0.9% (healthcare) per annum over the same period.  
  • Based on new experimental modelled nowcasts, total public service productivity is estimated to have been around 0.3% lower than pre-coronavirus (COVID-19) pandemic levels in 2022, although this varies by service area. 
  • An Office for National Statistics (ONS) work programme will update and monitor how best to improve estimates of timelier measures of annual public service productivity.

!

These are Experimental Statistics. These nowcast estimates will be subject to revision as modelling methods are refined and more up-to-date data becomes available. We advise caution when using the data.

Back to table of contents

2. Overview of productivity review

In June 2023, the Chancellor of the Exchequer asked Professor Sir Ian Diamond, the National Statistician, to undertake a Review of the measurement of public service productivity in England. To improve measures, the Office for National Statistics (ONS) is now partnering with government departments, academics, and expert users to help develop and further improve methodology and data sources.

This is the first publication from the Public Services Productivity Review (PSPR), focussed on improving short-term estimates. This utilises two experimental components. The first is an analysis of the baseline trend taken from existing annual total public service quality-adjusted productivity data, which is currently available for the period 1997 to 2020, to understand the long-term potential of productivity growth. This aims to provide insight into current underlying trends in public service productivity to improve policymakers’ ability to measure future performance.

Given the lag in official statistics, the second component is a nowcast estimate for public service productivity in 2021 and 2022, based on existing annual data and the path of quarterly public service productivity data. Quarterly data are currently available for the period Quarter 1 (Jan to Mar) 1997 to Quarter 2 (Apr to June) 2023 and are without quality adjustment. This nowcast estimate brings quality into the short-term measures to make them more comparable with the annual data, which is fully quality adjusted. Nevertheless, they should be considered experimental and subject to updating and revision. Our work on the Public Services Productivity Review will continue to evaluate other nowcasting methods and identify new data sources.

Further details on the concepts and methods used in our public service productivity statistics can be found in our Sources and methods for public service productivity estimates article. Further details on the annual data series can be found in our Public service productivity: total, UK, 2020 article, and on the quarterly series in our Public service productivity, quarterly, UK: April to June 2023 bulletin.

Back to table of contents

4. Public service productivity nowcasts for 2021 and 2022: a dynamic regression approach

Given the timeliness of our source data, which come from administrative sources, annual public service productivity estimates are currently produced at a two-year time lag. To give indicative estimates of timelier measures, we produce nowcast estimates of annual public service productivity each quarter, most recently in our Public service productivity overview, UK: April to June 2023 article. This nowcasting approach uses annualised quarterly public service productivity growth rates for the most recent periods and applies them to the annual series.

This approach has limitations. Although both the annual data and annualised quarterly estimates track changes in the quantity of services delivered, only the annual series applies a quality adjustment (see our Guide to quality adjustment in public service productivity measures article). Moreover, our quarterly output indices have lower coverage and are less granular than our annual estimates.

Between 1997 and 2019, our annual productivity measures perform better on average than quarterly productivity measures because of the inclusion of quality adjustment. Therefore, basing nowcasts on just the quarterly series may result in lower productivity growth estimates in 2021 and 2022.

As an alternative nowcasting approach, we produced experimental modelled estimates using dynamic regression. Unlike forecasting, which relies heavily on projections and assumptions about the future economic situation, our alternative approach uses data on annual quality-adjusted productivity trends for previous years (before 2020), and information on current quarterly public service productivity (up to Quarter 4 (Oct to Dec) 2022), to estimate annual public service productivity in 2021 and 2022. These nowcasts are a product of the observed annualised quarterly series in 2021 and 2022 and the relationship between the observed annual series and annualised quarterly series in 1997 to 2019 (excluding 2020 because of the impact of the coronavirus (COVID-19) pandemic). More information on the regression methods used in this article can be found in Section 8: Data sources and quality.

We acknowledge, as experimental statistics, that our alternative nowcasts presented for the first time in this article have limitations. As they use quarterly annualised data as part of the estimation process, the strength of the relationship between the two data series will affect the robustness of the estimates produced. To increase the reliability of our experimental nowcast method, we sought to incorporate relevant leading indicators for service area output. To reflect this, we have applied an adjustment to reduce healthcare output growth in 2021 by a quarter and then apply this to our nowcasts for healthcare and total productivity.

We will continue to evaluate other nowcasting methods and will monitor and improve experimental estimates of public service productivity where appropriate. This includes undertaking cross validation of estimates used when new data becomes available in early 2024. Changes to methods, quality adjustments and revisions to data will affect the accuracy of our current nowcast estimates and will lead to future revisions.

Our Data Science Campus has previously published Outputs exploring economic nowcasting methods. Likewise, the Economic Statistics Centre of Excellence has recently published a Discussion paper on nowcasting unpaid production activity and quality adjustments in public service productivity.

Total public service productivity

Our experimental regression-based measure of annual total public service productivity estimates growth of around 15.1% and 2.2% in 2021 and 2022, respectively. This large increase in 2021 reflects a post-coronavirus (COVID-19) pandemic rebound, as services recovered from the decrease seen during the pandemic. The level of total public service productivity is estimated to be around 0.3% lower by 2022 than their pre-pandemic peak.

We have included confidence intervals at the 95% level around our central modelled estimate. These provide an indication of the degree of uncertainty for total public service productivity estimates for 2021 and 2022. The width of the confidence intervals are determined by the uncertainty there is in the estimate produced (for instance, wider confidence intervals reflect greater uncertainty).

Our growth estimates in 2021 range between the lower bound (12.3%) and the upper bound (17.8%). Meanwhile in 2022, our estimates suggest productivity growth between 1.2% and 3.1%. In any particular year of the nowcast period, public service productivity growth is expected to lie somewhere within these growth rates on 95 out of 100 occasions.

If we apply our CAGR of 0.2% from 2019 onwards, the level of annual total public service productivity for 2022 would be 0.9% higher than our regression-based nowcast estimate.

Compared with the current nowcast estimates in our Public service productivity overview, UK: April to June 2023 article (that assume annual growth rates mirror those in the annualised quarterly series), our experimental nowcast approach assumes public service productivity grew faster post-pandemic. This is because of the inclusion of a quality adjustment element and methodological differences in our data. Figure 5 demonstrates the difference between the two nowcast measures.

Our regression-based estimates for 2021 and 2022, unlike the method used in our Public service productivity overview, UK: April to June 2023 article, account for the relationship between the observed annual series and annualised quarterly series in 1997 to 2019. This might account for differences in growth rate estimates.

We do not have timely data on public service delivery outcomes. Therefore, we cannot test the extent to which quality has improved or deteriorated over the nowcast period. If quality was to improve slower than pre-pandemic or deteriorated between 2021 and 2022 this nowcast could be an over-estimate of annual public service productivity growth in 2021 and 2022.

Back to table of contents

5. Productivity by public service area for 2021 and 2022: a dynamic regression approach

We have also produced modelled estimates for 2021 to 2022 for the following service areas: healthcare, education, and public order and safety. We exclude estimates for service areas which follow the “output-equals-input” output measure (more details can be found in Section 6 of our Public service productivity: total, UK Quality and Methodology Information (QMI)), these include: defence, social security administration, police and other services. We also exclude adult social care and children’s social care, given data limitations.

To nowcast estimates for 2021 to 2022, we require annualised quarterly data for each service area, as described in Section 4: Public sector productivity for 2021 to 2022: a dynamic regression approach. This requires us to select the most appropriate quarterly regressors to estimate the 2021 to 2022 annual values for these series. Each service area uses the most appropriate corresponding regressor; for example, public order and safety uses justice and fire.

Modelled estimates take into consideration past observations of the annual series, and the more up to date annualised quarterly series data for each service area. This means it is not possible to weight and aggregate individual service area estimates in line with estimates of total public service productivity described in Section 4: Public sector productivity for 2021 to 2022: a dynamic regression approach.

Healthcare

Healthcare represents the largest service area included in public service productivity (around approximately 40% expenditure share of total public service provision). The coronavirus (COVID-19) pandemic caused widespread cost pressures and disruption to healthcare inputs and outputs. Consequently, annual healthcare productivity fell by 23.0% in 2020 on a quality-adjusted basis.

During the coronavirus pandemic period, there were several fundamental changes to the delivery of healthcare services (for instance, some non-urgent services were stopped to reduce the spread of COVID-19) and challenges in data collection.

Because of the exceptional impact of the pandemic on healthcare services, we conducted an additional quality assurance of our new experimental nowcasts, reaching out to external and other government departmental stakeholders. The universal message we received is that the post-pandemic environment the NHS is facing is still materially different from the pre-pandemic environment. Moreover, the NHS, while receiving additional resources, has had to heavily focus on returning quantity of outputs to pre-existing levels. The opportunity to consider improvements to quality of services appears very restricted. Therefore, we recommend that the lower bound confidence interval in our experimental methodology for healthcare is likely to better reflect growth in healthcare productivity across this period.

Using our experimental nowcast approach, we assume annual healthcare productivity on a quality-adjusted basis has grown by 26.6% in 2021 and 1.1% in 2022. This represents a recovery from much of the previous falls seen during the pandemic.

Our estimates of growth in 2021 range between the lower bound (22.2%) and the upper bound (30.9%), while in 2022, our estimates suggest productivity growth between 0.3% and 1.8%. Therefore, in any particular year of the nowcast period, healthcare productivity growth is expected to lie somewhere within these growth rates on 95 out of 100 occasions.

As noted previously, an adjustment has been applied to our nowcast for healthcare productivity to reflect more accurately the most up-to-date indicators of service area outputs.

Education

Education represents the second largest service area in public service productivity by expenditure share. As with healthcare, annual education productivity on a quality adjusted basis fell sharply (26.1%) in 2020. The coronavirus pandemic affected the delivery of education services through remote learning on teaching hours, teaching materials provided, and increased sickness rates. Educational attainment in 2020 was also affected by the pandemic. More information on how the coronavirus pandemic affected education productivity can be found in our Public service productivity: total, UK, 2020 article.

Using our experimental nowcast approach, annual education productivity on a quality-adjusted basis is assumed to have grown by around 26.7% in 2021 and 7.4% in 2022. This represents a recovery from much of the previous falls seen during the pandemic.

Confidence intervals for annual education productivity nowcasts are wider than those for healthcare and public order and safety, highlighting a higher degree of uncertainty with these estimates.

Our estimates of growth in 2021 range between the lower bound (17.9%) and the upper bound (35.6%), while in 2022, our estimates suggest productivity growth between 5.4% and 9.2%. Therefore, in any particular year of the nowcast period, education productivity growth is expected to lie somewhere within these growth rates on 95 out of 100 occasions.

Public order and safety

Within the public order and safety (POS) service area, there are four main components: fire, courts, probation, and prisons. Police is measured separately to POS and is, therefore, excluded from these measurements.

As explained in Section 3: Trends in public service productivity, annual POS productivity on a quality-adjusted basis has fallen steadily between 1997 and 2019. Restrictions to deal with the coronavirus pandemic contributed to a further fall of 13.2% in 2020.

Our nowcast method estimates that annual POS productivity almost recovers to the pre-coronavirus (COVID-19) pandemic trend level in 2021 and 2022 with annual growth of around 0.4% and 6.8%, respectively.

As with healthcare and education, we have provided confidence intervals around our central estimate. Our estimates in 2021 range between the lower bound (negative 8.8%) and the upper bound (9.7%), while in 2022, our estimates suggest productivity growth between 5.2% and 8.1%. Therefore, in any particular year of the nowcast period, POS productivity growth is expected to lie somewhere within these growth rates on 95 out of 100 occasions.

Back to table of contents

6. Public service productivity data

Public service productivity nowcasts, UK
Dataset | Released 17 November 2023
Experimental dynamic regression nowcast estimates of public service productivity.

Back to table of contents

7. Glossary

Classification of the functions of government

The Classification of the functions of government (COFOG) is the structure used to classify government activities. It is defined by the United Nations Statistics Division.

Confidence interval

Confidence intervals use the standard error to derive a range in which we think the true value is likely to lie.

Direct output measurement

Using a cost-weighted activity index to estimate the non-quality-adjusted of a service provided, such as the number of students in state schools, adjusted for attendance to produce an estimate of total hours of schooling delivered each year. Differs from indirect output measurement, where output is assumed equal to inputs.

Public services

These are services delivered by or paid for by government (central or local). If paid for by the government, they may be delivered by a private body. For example, the provision of nursery places by the private sector, where these places were funded by the government.

Quality adjustment

A statistical estimate of the change in the quality of a public service, using an appropriate metric, such as safety in prisons as part of the public order and safety adjustment.

Service area

The way we refer to the breakdown of public services into nine areas, closely following COFOG.

Back to table of contents

8. Data sources and quality

Regression methods

The nowcasts are produced using dynamic regression and the relevant annualised quarterly series. Dynamic regression is a time series analysis method which allows nowcast to use information from past observations of the series and also information from predictor variables. The predictor variables used were the relevant annualised quarterly series. For example, total productivity (annual) is regressed on total productivity (annualised quarterly). Therefore, the nowcasts are a consequence of the observed annualised quarterly series in 2021 and 2022 and the relationship between the observed annual series and annualised quarterly series in 1997 to 2019 (2020 was excluded because of the effect of the coronavirus (COVID-19) pandemic). Dynamic regression is an extension of autoregressive integrated moving average (ARIMA) modelling; the extension allows inclusion of predictor variables. For example, a standard regression would be:


Where the error is assumed to be uncorrelated. However, in dynamic regression, the error term is assumed to follow an ARIMA process. For example, if the error term followed an ARIMA process the model would be:


This is where the final error term (epsilon) is assumed to be white noise.

Acknowledgements

Mark Hogan, Dimitrios Nikolakis, Francis Dunnett.

Back to table of contents

10. Cite this article

Office for National Statistics (ONS), released 17 November 2023, ONS website, article, Public service productivity, UK: 1997 to 2022

Back to table of contents

Contact details for this Article

Ed Bailey, A Blunden, Nick Chapman, Rebecca McAlpine
psp.review@ons.gov.uk
Telephone: +44 1633 580075