Research · Labour Economics · 2026
Does Poor Health Lower
Your Hourly Pay?
Using 7,372 permanent employees from the UK's Q1 2026 Labour Force Survey, I
estimate how self-reported health conditions relate to wages, and whether that
gap is real or just a statistical artefact of who gets hired.
-8.5%
OLS wage penalty (limiting)
+2.2%
After selection correction
↓ Download Full Paper (PDF)
Who is in this study?
Permanent employees aged 21-60 with valid wage and education data, Q1 2026 QLFS.
Sample size
7,372
Permanent employees
Mean hourly wage
£22.78
Gross, Q1 2026
Limiting condition
9.9%
of sample
Work from home
30.2%
at least some of the time
The Wage Gap
Workers with a limiting health condition earn less. But how much of that gap survives once we control for education, experience, occupation, and industry?
Hourly wage distribution by health status (capped at 99th percentile)
Healthy: median wage
£19.24
Reference group
Non-limiting: median wage
£18.97
No significant penalty after controls
Limiting: median wage
£15.18
-8.5% after controlling for human capital
Raw median gap
£4.06
Healthy minus limiting (unadjusted)
What does the regression tell us?
After controlling for age, education, occupation, gender, part-time status, industry, firm size,
region, and WFH status, workers with a limiting health condition earn
8.5% less per hour than otherwise identical healthy workers. Workers with a non-limiting
condition show no statistically significant penalty at the 5% level.
But, crucially, this gap may not reflect discrimination in pay-setting. It may instead reflect
who gets hired into permanent work in the first place.
Do health groups follow the same wage trajectory?
A structural stability test (HC1-robust Wald) finds no significant slope differences. The experience and education returns are the same across health groups. The gap is a level shift, not a different trajectory.
Predicted wage vs experience, evaluated at sample means of other variables
The curves run parallel, confirming that the wage penalty for limiting conditions is a constant percentage gap at all experience levels, not a diverging trajectory. The slopes-only Wald test (F = 1.57, p = 0.068) fails to reject equal returns to experience across health groups.
What else drives wages?
Key coefficient estimates from the extended model with HC1 heteroscedasticity-robust standard errors. Error bars show 95% confidence intervals.
Extended model: selected coefficients with 95% CI (hover for % effect)
Is the wage gap real, or a hiring artefact?
Workers with health conditions face a much higher barrier to getting permanent employment, and those who do hold permanent jobs earn less than their characteristics alone would predict, as if trading pay for security. OLS rolls that selection effect into the health coefficient, making the in-work gap look worse than it really is.
1
Selection into employment
A probit model on 39,636 working-age adults shows a limiting condition reduces the probability of permanent employment (coeff −0.30, p < 0.001).
The predicted selection probability is converted to an inverse Mills ratio (λ̂) and added to the wage equation. λ̂ = −0.446 (p < 0.001) confirms significant selection bias.
Once selection is accounted for, the limiting-health wage penalty flips from -8.5% to +2.2%, statistically indistinguishable from zero. Validated by B = 200 pairs bootstrap.
OLS vs Heckman selection-corrected wage penalties (% effect, 95% CI)
The main finding
The OLS penalty of -8.5% for limiting health conditions
disappears entirely once we correct for the selective hiring of health-impaired workers into
permanent employment. The Heckman-corrected estimate is +2.2%,
statistically indistinguishable from zero.
This implies the wage gap is not discrimination in pay-setting. It is a barrier to entry: workers
with limiting conditions find it harder to get permanent jobs in the first place, and once that
selection is accounted for, those in permanent work earn no less than comparable healthy
colleagues.
Industry wage premiums
The 20 industry dummies have the largest joint significance of any variable group (LM = 468.2). Here is how each sector compares to the reference (Health & Social Work).
Industry wage premiums / penalties vs Health & Social Work (faded = not significant at 5%)
Read the full paper
The full academic paper includes econometric model specification, all diagnostic tests (RESET,
MWD, LM, Breusch-Pagan, White), structural stability analysis, and complete appendix tables.
↓ Download PDF
Data: UK Office for National Statistics, QLFS Q1 2026 · Tools: Python, Quarto, statsmodels