Changes in the composition of the employee workforce and the Coronavirus Job Retention Scheme (CJRS), or furlough scheme, both explain a large part of the growth in average earnings for employees during the coronavirus (COVID-19) pandemic.
The effect of employee workforce composition on earnings growth increased very rapidly at the start of the coronavirus pandemic; estimates using Labour Force Survey (LFS) data suggest that changes in the composition of the workforce contributed 0.5 percentage points of earnings growth in Quarter 1 (Jan to Mar) 2020, increasing to 3.4 percentage points of earnings growth in Quarter 2 (Apr to June) 2020.
The effect of furlough on earnings growth was greatest in Quarter 2 2020, where it contributed negative 3.2 percentage points to earnings growth (4.3%).
The effect of employee workforce composition on earnings growth was at its highest in Quarter 1 2021, where it contributed 5.6 percentage points to earnings growth (7.8%).
Decreases in the proportion of part-time employees and increases in the overall level of qualifications held by employees were the largest contributors to this effect, and most of the compositional effects for individual characteristics during the coronavirus pandemic were positive.
During the global financial crisis of 2008 to 2009, the impact of changes in the employee workforce on earnings growth was far smaller with a mix of positive and negative compositional effects from individual characteristics, peaking at 0.7 percentage points in Quarter 4 (Oct to Dec) 2008.
The coronavirus (COVID-19) pandemic and the responses to it have altered the UK labour market and average earnings. The Office for National Statistics (ONS) has been reporting and monitoring these coronavirus pandemic-related effects on average earnings in our monthly Average Weekly Earnings in Great Britain bulletin since we first observed them. In July 2021, we published a blog post explaining the challenges in interpreting average earnings data during the coronavirus pandemic and how changes in the composition of the workforce and base effects were affecting our lead measure of average earnings, which is our Average Weekly Earnings (AWE) bulletin.
In this article, we provide an estimate of the underlying average wage growth for employees during the coronavirus pandemic and up to October to December 2021. We separate this estimate from the wage growth that can be attributed to changes in the employee workforce and the wage growth that can be attributed to the government's Coronavirus Job Retention Scheme (CJRS) or furlough scheme. The CJRS scheme supported businesses to pay their employees on furlough during the coronavirus pandemic. We also identify which changes in the composition of the workforce have had the biggest impact on average earnings. Lastly, we make a comparison with the global financial crisis, where changes in the composition of the employee workforce were smaller and thus had a smaller impact on earnings growth.
The estimates presented in this article are experimental and based on Labour Force Survey (LFS) data. We use the LFS because it provides detailed information on the characteristics of the employee workforce, including whether an employee was on furlough. The earnings growth estimates in this article should not be interpreted as the official source of information on earnings growth. For those, please refer to our headline measure of AWE.Back to table of contents
This article provides a measure of the underlying wage growth, separating it from the wage growth attributed to the compositional and furlough effects just described.
We use data from the Labour Force Survey (LFS), as opposed to data from our headline measure of Average Weekly Earnings (AWE), because the LFS contains detailed information on the characteristics of workers, jobs and furlough, which we use in this analysis. Our measure of earnings is the LFS’ "gross weekly pay in main job". For a discussion of the advantages and limitations of using LFS data, see the Data sources and quality section.
To estimate the compositional and furlough effects, we use the Oaxaca decomposition method, following Abel, Burnham and Corder (2016) (PDF, 188.47KB) and Blundell, Crawford and Jin (2014), and measure year-on-year earnings growth by comparing average weekly earnings between one quarter and the same quarter a year earlier.
The Oaxaca decomposition method allows us to decompose earnings growth as:
Earnings growth (%) = underlying earnings growth + compositional effect + furlough effect,
where the underlying earnings growth is the part of earnings growth that we cannot attribute to the composition of the workforce (compositional effect) or furlough (furlough effect).
More details on the data and method, including a detailed econometric specification of the Oaxaca model used, are available in the Data sources and quality section.Back to table of contents
The global financial crisis took place in the UK between Quarter 2 (Apr to June) 2008 and Quarter 2 2009. The compositional effect during the global financial crisis was much smaller than during the coronavirus (COVID-19) pandemic. Between Quarter 2 2008 and Quarter 2 2009, the compositional effect increased average earnings growth by a maximum of 0.7 percentage points, whereas from Quarter 2 2020 to Quarter 2 2021, in each quarter the compositional effect increased earnings growth by at least 2.8 percentage points (Figure 4a).
Figure 4b: Part-time status, occupation and qualifications also made notable contributions to the compositional effect between 2008 and 2011
Compositional effect of average weekly earnings annual growth rates (LFS), not seasonally adjusted, contributions by characteristics of the employee workforce, UK, January to March 2008 to October to December 2011
Download the data
The global financial crisis was different to the coronavirus pandemic. Although all recessions can affect industries, jobs, and types of workers differently, the global financial crisis did not produce the rapid, large impact on trade and jobs in some industries, occupations and types of workers that the coronavirus pandemic did. The radical changes in the composition of the workforce observed during the coronavirus pandemic explain why the compositional effects were much larger than in the global financial crisis.
During the financial crisis, the individual contributions of the different characteristics in the employee workforce to the compositional effect were smaller than during the coronavirus pandemic, and they also contributed to it in different directions (see Figure 4b). Changes in the proportion of part-time workers were the largest negative contributors to the compositional effect (contributing -0.8 percentage points to earnings growth in Quarter 3 (July to Sept) 2009), suggesting that an increase in part-time workers contributed to a fall in average earnings. However, this was balanced by other factors such as qualifications (contributing 0.9 percentage points to earnings growth in Quarter 4 (Oct to Dec) 2009).Back to table of contents
In this article, we estimate how furlough and the employee workforce composition have affected earnings growth. We call these the “furlough effect” and the “compositional effect”.
The Office for National Statistics (ONS) has published estimates of the compositional effect in our Average weekly earnings (AWE) in Great Britain bulletins. These estimates are based on age, occupation and part-time status. From March 2020 to February 2021, these estimates have steadily increased, from 0.8% to 2.9%.
In this analysis, we provide a refined method to estimate compositional and furlough effects using the Oaxaca decomposition. We use data from the Labour Force Survey (LFS) rather than AWE to make use of its detailed information about the characteristics of workers, jobs and furlough.
The method introduced in this release remains experimental. When further refinements and improvements are made in the coming months, we will consider incorporating additional analysis as part of our AWE in Great Britain bulletins.Back to table of contents
How furlough and changes in the employee workforce have affected earnings growth: experimental results from an Oaxaca decomposition using Labour Force Survey (LFS) data
Dataset | Released 29 April 2022
Experimental estimates of compositional and furlough effects using quarterly Labour Force Survey earnings data, from 2008 to 2021.
Coronavirus pandemic period
The phrases "during the coronavirus pandemic" and the "coronavirus pandemic period" refer to the period March 2020 to present.
Coronavirus Job Retention Scheme
The government announced the Coronavirus Job Retention Scheme (CJRS) on 20 March 2020. It was introduced to support employers through the coronavirus (COVID-19) period. This has commonly been referred to as the furlough scheme.
The scheme was based around Her Majesty’s Revenue and Customs’ (HMRC's) Pay As You Earn (PAYE) system. It worked by providing grants to employers of up to a maximum 80% of salary, to a maximum value of £2,500 per employee. Up to the end of July 2020, the scheme also met some of the cost of employer pension contributions and employer National Insurance contributions.
Furlough is defined as a temporary absence from work allowing employees to keep their job while the coronavirus pandemic continues.
Our identification of furloughed employees in the Labour Force Survey (LFS) data is not perfect and does not match the official number of furloughed employees reported in the Coronavirus Job Retention Scheme statistics. However, our estimates are the best approximation we have to these numbers. Although the LFS undercounts the number of employees on furlough, the trends observed in the LFS are similar to those in the CJRS statistics.
For details on how we identify furloughed employees, see the Data sources and quality section.
Average Weekly Earnings (AWE)
Average Weekly Earnings (AWE) is the lead monthly measure of average weekly earnings per employee. It is calculated using information based on the Monthly Wages and Salaries Survey (MWSS), which samples around 9,000 employers in Great Britain. The estimates do not include earnings of self-employed workers.
The estimates are not just a measure of pay rises. They do not, for example, adjust for changes in the proportion of the workforce who work full time or part time, or other compositional changes within the workforce.Back to table of contents
This analysis uses Labour Force Survey (LFS) data. We use data from the LFS instead of data from Average Weekly Earnings (AWE) because the LFS contains rich information on individual and job characteristics, including if workers reported being on furlough.
Estimates of gross weekly and hourly earnings from the LFS are based on employees and on two-fifths of the quarterly sample and are therefore subject to high sampling variability. The data on individual’s earnings captured by the LFS are thought to be of a lower quality than Annual Survey of Hours and Earnings (ASHE) or AWE data because LFS information is self-reported by employees. ASHE and AWE earnings data are based only on employees but are collected from the employer, which is thought to be more accurate as employers can consult payroll records. Individuals may not have such records to hand, and their responses may therefore be subject to higher levels of recall error. Furthermore, LFS responses can be given by proxy (by other individuals in the same household) when an individual is unavailable for interview. This gives further scope for recall error from respondents. For these reasons, the Office for National Statistics (ONS) recommends that any short-term measurement of change be made with caution.
“Full-time” in the LFS is based on respondents' self-assessment. The estimates relate to an individual's main job only.
The performance and quality monitoring report provides data on response rates and other quality measures.
Sample and methodology
We use the cross-sectional LFS, for each quarter from Quarter 1 (Jan to Mar) 2007 to Quarter 4 (Oct to Dec) 2021. Individual observations are dropped if any response to any of the characteristics in the model is missing, or if a refusal to respond has been recorded.
The variables included in our model are described in the accompanying dataset, along with brief notes on their specification.
Our identification in the data of which employees were on furlough is our best approximation. Our estimate of furloughed employees undercounts those on furlough compared with the Coronavirus Job Retention Scheme (CJRS) statistics, which implies that the furlough effect we estimate is likely to be understated.
In particular, we identify furloughed employees from Quarter 3 (July to Sept) 2020 onwards as those employees who have kept their job but have worked reduced hours or not worked because of being on furlough (LFS variable CORO20B2). In Quarter 2 (Apr to June) 2020, the LFS did not collect this specific information, so we use a combination of variables and classify an employee in furlough if they:
have been temporarily away from paid work (JBAWAY = 1)
have worked fewer hours than usual (“laid off/short time/work interrupted”) “due to economic or other causes or other reasons” (YLESS20 = 15 OR 19)
have stated that the reason they worked fewer hours was linked to coronavirus (CORO20A2 = 1)
For the employee to be put in the “furlough” category, all these criteria must apply to them.
To calculate the compositional effect in year-on-year average weekly earnings growth, we use a pooled Oaxaca decomposition, following Abel, Burnham and Corder (2016) (PDF, 188.47KB).
To produce this decomposition, we first estimate wage equations for each quarter. The detailed form of this wage equation is:
We then calculate the decomposition for each wage equation. The difference in the average wage between the wage equation in one period and the wage equation one period earlier (in this analysis, the same quarter one year earlier) can then be expressed as:
Back to table of contents
We would like to thank the Bank of England and Professor Ana Galvao for their input into this work.Back to table of contents
Contact details for this Article
Telephone: +44 20 7112 0119