The Lesson of the Mismatch: Read Labour-Market and Benefit Data Together
The UK’s post-pandemic rise in health-related economic inactivity is real, but the evidence does not support treating it as a one-for-one forecast of benefit claims. The final task of this series is to connect the aggregate trend, the relative experience of clinically vulnerable people, and the separate institutional routes that produce disability or incapacity records. The practical judgment is to read these figures together without forcing them into one number: ask what population, outcome, comparison, and time point each measure represents, then attach the conclusion to that level of evidence.
One country, two levels of evidence
The official trend and the longitudinal study describe the same broad UK problem at different levels. The ONS series is an aggregate percentage for people aged 16–64 recorded as economically inactive because of long-term sickness. Its annual figure was 23.7% in 2019, then 25.8% in 2020, 26.2% in 2021, 27.6% in 2022, 29.8% in 2023, 29.9% in 2024, and 30.7% in 2025. From 2019 to 2025, the category rose by seven percentage points. The longer series shows that this followed a broad decline from the late 1990s into the late 2010s, so the post-2020 movement is a reversal rather than a routine fluctuation around an unchanged line.
The series establishes a population-level change in a labour-market category. It does not establish why the category grew, whether the same people remained in it, or how many of those people applied for a benefit. The IZA paper reports that the number of working-age adults inactive because of long-term sickness had exceeded 2.8 million by early 2024, but it also warns that difficulties affecting the Labour Force Survey make the exact size and timing of the change difficult to quantify. That qualification is part of the evidence, not a reason to replace the trend with a more confident story.
The longitudinal study asks a narrower question. Using the UK Household Longitudinal Study linked to the Understanding Society COVID-19 Study, it compares people classified as clinically vulnerable to COVID-19 with people classified as low risk. Its main estimates ask whether the post-2020 change in outcomes was larger for the vulnerable group. The preferred Model 1 estimates for sickness-related inactivity are 2.1 percentage points in the pooled specification and 1.3 points with individual and year fixed effects. Both are statistically significant at the 5% level. The aligned-period estimates are smaller—0.7 points in the pooled model and 0.5 points with fixed effects—but remain positive, with the former significant at the 10% level and the latter at the 5% level.
These are not alternative estimates of the seven-point movement in the ONS series. They are estimates of a difference between changes in two groups within a selected longitudinal sample. A 1.3-point result means that the post-pandemic change for the clinically vulnerable group exceeded the change for the low-risk group by 1.3 points under that specification. It does not mean that 1.3% of the national population newly became inactive, and it does not identify the share of the ONS increase caused by vulnerability.
The result is nevertheless substantively important for the group it describes. The paper places the pre-pandemic prevalence of health-related inactivity among clinically vulnerable respondents at around 5–6%. Against that baseline, a 1.3–2.1 percentage-point relative difference corresponds to roughly a 20–40% increase. Absolute and relative magnitude answer different questions here. The percentage points keep the estimand visible; the relative comparison shows why a seemingly modest point difference can represent a serious deterioration for a group that began with a relatively low prevalence.
The benefit findings complete the contrast. For PIP/DLA receipt, the largest principal estimate is 2.2 percentage points in the pooled Model 1 specification, but its confidence interval includes zero and the p-value is 0.11. The other estimates are positive but statistically insignificant: 1.4 points in Model 1 with fixed effects, 1.6 points in the pooled aligned-period model, and 0.4 points in its fixed-effects version. For incapacity benefits, the four principal estimates are also positive but statistically insignificant: 1.3, 0.5, 1.5, and 0.6 percentage points across the corresponding specifications. The study therefore finds a robust relative pattern for sickness-related inactivity, but no statistically robust matching concentration in disability or incapacity receipt.