How reliable is the British labor market data?
Unfortunately, the answer to that question is “not very”!
Let’s start with the traditional labor market data that is typically featured in the news — like unemployment, employment, and wage growth. Since it would be impossible to question every single person living in the U.K. on a regular basis, statistical agencies like the ONS (Office for National Statistics) rely on survey data that is sent to thousands of households. With the help of some statistical adjustments, statisticians then try to ensure that the data they have collected is a representative subsample of the population. This is how the unemployment rate is measured.
The problem that the ONS has faced in recent years is that the response rate to U.K. Labour Force Survey has completely plunged. While other agencies like the Bureau of Labor Statistics (BLS) in the U.S. are struggling with similar issues, the British experience has been the most dire. Response rates have declined to 20%, which led to the ONS temporarily ceasing the publication of its monthly labor market data in the fall of 2023.
While there has been slight improvement since then, response rates remain low. Meanwhile, a methodological overhaul was promised but has been postponed several times and was pushed out to 2026.
Given all these problems, the Bank of England has publicly questioned the accuracy of the ONS statistics. The implications of misleading labor market data are consequential as central bankers rely on those statistics to conduct monetary policy. A mistake can have far-reaching consequences, potentially even causing a recession. For example, central bankers might keep interest rates too high for too long because they overestimate the strength of the labor market. Or they might lower them prematurely and spark inflation. Either way, unreliable data makes big policy mistakes more likely to happen.
Three employment series tell a different story
To complement the household survey, the ONS publishes two other employment series, which are based on different data collection methods and have different scopes. As a reminder, the employment series based on the household survey captures every single person employed in the U.K., including self-employment and gig work (even as the latter is hard to estimate).
Payroll employment data is based on HMRC records (tax records) and captures employees only. One reason why these two different series might diverge, for example, is because of countercyclical trends in self-employment. We know that the recent loss of payroll jobs, concentrated in hospitality and retail, has probably led to more people moving into self-employment, including gig work, to offset income losses.
The third series, called “workforce jobs,”, is only released on a quarterly basis instead. It is collected through a combination of employer and household surveys as well as administrative sources. Furthermore, it does not count employed people but the number of jobs instead. Therefore, one person having two jobs would be counted twice.
The divergence between the workforce job series and payroll employment since Q2 of last year is somewhat puzzling. While workforce jobs are up by 1%, payroll employment is down by 0.5%. Theoretically, the shift into self-employment could play a role here, but it could also point towards deeper problems such as data unreliability.
The payroll employment data also has flaws
Because of the previously discussed problems with the household survey, analysts have increasingly focused their attention on the payroll employment data. Alas, that dataset also suffers from significant problems.
Revisions to economic data are very common and just a natural process when dealing with data collection. There is a natural tradeoff between accuracy and speed of release. We need economic data to be timely, but that also means that early estimates are not as accurate. The problem with the U.K. payroll data is that the bias is predictable beforehand. The first data release is consistently inaccurate in one direction, meaning that statistical adjustments to the process could improve the data if the ONS was willing to make the investment.
Let me illustrate the issue. The graph below shows the first release and final release of U.K. payroll jobs growth. As one can see, months with large job losses have frequently been revised up and months with large job gains have frequently been revised down.
This is extremely problematic because news outlets are focusing on the first data release to write attention-grabbing headlines while paying little attention to subsequent revisions that make for much less scary headlines!
Turns out that the actual amount of job losses since October 2024 was only 160,000 after revisions came in—a sizeable difference.
The initial payroll employment release is systematically biased
Unfortunately, the bias in the U.K. payroll data is quite enormous. The graph below illustrates the statistical relationship between the first job growth number and subsequent revisions.
Based on the historical relationship, an initial monthly job growth number of 60,000 leads to a downward revision of about 15,000 jobs while a monthly job loss number of 30,000 leads to an upward revision of about 50,000.
Given that we know beforehand in which direction subsequent revisions are likely to go, the initial data release could be improved or should be withheld altogether until better data is available.
While the BLS (Bureau of Labor Statistics) in the U.S. has recently come under fire from President Trump, who asserted that the recent downward revisions to the jobs data meant that the data was rigged from the start, American payroll data does not suffer from the same problem. Even as revisions tend to be correlated during the business cycle—periods of economic weakness tend to have more downward revisions while strong growth is typically associated with more upward revisions—there is zero correlation in the U.S. data between the initial jobs number and subsequent revisions. Therefore, there is no obvious way of knowing in which way later revisions will go—exactly as it should be!
What does that mean for recruiters?
Even though we live in a time of big data, labor market statistics have ironically become less reliable over time, which is obviously a massive problem. Recruiters should know that the initial headline grabbing numbers do not necessarily reflect the reality of what’s going on. They should also remember the limitations and flaws of the labor market data that the ONS is currently releasing. Better to disregard the monthly ups and downs and focus on quarterly trends and a variety of measures, which more accurately reflect the dynamics of the labor market. While economic and labor market data from private data providers can provide supplementary information, national statistics remain the gold standard for now despite their flaws.
