The difference between occupations can amount to years of life, not just percentages. French life tables and our own calculations from Spanish mortality records reveal the gap, and why it is harder to explain than it looks.
Imagine two groups of 100 men in France, all age 35. One group consists of blue-collar workers. The other includes managers and senior professionals.
If the death rates recorded at each age in France during 2020–2022 stayed the same, about 85 blue-collar workers would reach age 65, compared with 94 managers and senior professionals.
Out of 100 people age 35, how many would reach 65?
A longer dark segment means more people surviving. The terracotta segment shows how many would not reach 65 if age-specific death rates stayed the same.
Men
Women
Rounded, conditional estimates from period life tables, not a 30-year follow-up of actual groups of 100. The period includes the COVID-19 pandemic and elevated mortality. Source: INSEE.
The occupational survival gap is nine in 100 among men, but four in 100 among women. Why the difference? These period life tables cannot isolate its causes.
The numbers invite an obvious question: If a blue-collar worker became a manager tomorrow, would he live longer? Unfortunately, a chart of occupations cannot tell us.
But there is a second surprise. In our analysis of Spanish death and employment records for 2022–2023, age-adjusted annual mortality in elementary occupations was 1.99 times that of scientific and intellectual occupations among men, but only 1.30 times among women. The comparison concerns employed groups, not the years of life caused or prevented by their jobs. Explore the underlying data, uncertainty estimates and calculation code.
A five-year gap. But what does it mean?
French statisticians also estimated how many years people who had already reached age 35 could expect to live if age-specific death rates remained at their 2020–2022 levels.
For men in managerial and senior professional occupations, the answer was 48.9 more years. For blue-collar workers, 43.6 more years. That’s a 5.3-year gap. For women in the corresponding groups, the gap was 3.4 years.
Here’s the crucial distinction: 48.9 years remaining at age 35 means a life expectancy of roughly age 84, not death at age 49. Nor is it a prediction of when someone who is 35 today will die.
France’s national statistics agency, INSEE, provides a finer-grained picture as well. The figure below displays 15 occupational categories for men and 12 for women, with no hidden menus. Each number is the estimated years remaining after 35. A point farther right means a longer expected life. Pale dots show the pre-pandemic period of 2017–2019; darker dots show 2020–2022.
How many years remain after age 35?
48.9 years remaining at 35 means a life expectancy of about age 84 in this period life table, not death at 49. Larger numbers and points farther to the right indicate more years of expected life.
Men
Women
The dot scale starts at 40, not zero. Values are displayed to one decimal place, but differences of a few months between nearby groups may be unstable; uncertainty intervals for these detailed categories are not shown. Source: INSEE.
A few comparisons stand out. Among men, primary-school and vocational teachers averaged about 48.1 more years after 35, while transport drivers averaged 44.4. That’s a 3.7-year difference, but the chart does not tell us what caused it. Lower-skilled manual workers sit near the bottom of both the men’s and women’s distributions. Across the detailed categories, the gap between the highest and lowest estimates reaches about 6.0 years for men and 4.4 for women.
Those are not precise league-table positions. Smaller occupational groups carry statistical uncertainty, and a difference of a few months between neighboring rows may not be meaningful. The categories also do not establish that someone spent an entire career in the same occupation. For some people, the statistics use the last occupation recorded.
Education complicates any simple job explanation. In the same French data, men with postsecondary education had 8.0 more expected years after age 35 than men without a formal diploma (48.5 versus 40.5), exceeding the 5.3-year occupational gap. Education, income, early-life health and career opportunities overlap; these gaps cannot be subtracted to isolate the effect of work. Source: INSEE.
Postsecondary education vs. no formal diploma
Managers and senior professionals vs. blue-collar workers
France, 2020–2022. Differences in expected years remaining at age 35 (INSEE). These contrasts overlap and cannot be subtracted or interpreted as causal effects.
Spain offers a different test
The French numbers describe remaining years of life. Would occupational differences also appear if we looked instead at deaths among people who were employed?
To investigate, we analyzed two publicly available datasets from Spain’s national statistics institute, INE. One records deaths; the other records the number of people employed, broken down by occupation, sex, and age. In these statistics, occupational status is assigned using records from the start of the year, rather than a full work history. We compared nine occupational groups among workers ages 35–64 in 2022–2023, adjusting for differences in their age distributions.
Among Spanish men ages 35–64, the age-standardized annual mortality rate was 101.6 per 100,000 in scientific and intellectual occupations and 201.7 in elementary occupations. Among women the corresponding rates were 60.4 and 78.6. Elementary occupations is Spain’s term for mainly routine, lower-skill work, not elementary-school teaching. Some other female occupational groups had higher recorded rates, so no single ranking fits both sexes.
The approximate 95% confidence intervals for the age-adjusted rate ratios were 1.88–2.10 for men and 1.22–1.39 for women. Their difference exceeds the random variation expected under the Poisson death-count model (p < 0.001), but this does not rule out misclassification, employment selection or confounding.
The pattern also appears in each year separately: the elementary-to-professional rate ratio was 2.07 for men and 1.44 for women in 2022, versus 1.90 and 1.18 in 2023. The magnitude changed, especially for women; these are sensitivity checks, not proof of a stable causal effect. Year-by-year figures and code are included in the reproducibility package.
How much do annual death rates differ?
For every 100,000 employed people ages 35–64, after adjusting for differences in age. The “Difference” column shows how much higher the age-standardized annual rate was than in scientific and intellectual occupations of the same sex.
Men
deaths
deaths
deaths
deaths
deaths
deaths
deaths
deaths
Women
deaths
deaths
deaths
deaths
deaths
deaths
deaths
deaths
In the men’s panels, for example, 202 versus 102 represents a difference between age-standardized rates of about 100 deaths per 100,000 employed people per year. It is not a count of deaths proved to be caused by the jobs. Numbers are rounded; full results and approximate uncertainty intervals appear in the research workbook. Sources and Mike’s Balance calculations: INE deaths and employment.
View the Spanish data and methods18 occupational groups · uncertainty ranges · Excel, CSV and code
Download the data and calculation code
Download complete reproducibility package (ZIP, approximately 47 KB)
Download summary CSV (18 rows) · Download calculations workbook (Excel)
The ZIP contains age-specific calculations, a comparison workbook, annual sensitivity checks, the original analysis scripts, and standalone scripts for the sex-interaction and year-by-year checks. To rebuild the analysis from scratch, download the large source CSV files directly from INE using the links below. Source files are not redistributed in the ZIP.
Main results by occupational category
All figures are age-standardized annual deaths per 100,000 people employed, using a common distribution across ages 35–64. Approximate 95% uncertainty intervals reflect recorded death counts; they do not cover misclassification or unmeasured differences between groups. Values are shown to one decimal place.
| Code | Occupation | Rate | Approx. 95% CI | Deaths |
|---|---|---|---|---|
| Men | ||||
| 1 | Managers | 120.9 | 113.9–127.9 | 1,297 |
| 2 | Scientific and intellectual professionals | 101.6 | 97.4–105.7 | 2,301 |
| 3 | Technicians and associate professionals | 131.7 | 126.2–137.1 | 2,223 |
| 4 | Office and administrative staff | 139.9 | 132.3–147.5 | 1,315 |
| 5 | Service and sales workers | 172.8 | 167.5–178.1 | 4,117 |
| 6 | Skilled agricultural workers | 188.7 | 176.2–201.2 | 918 |
| 7 | Craft and construction workers | 161.1 | 156.3–165.9 | 4,351 |
| 8 | Plant and machine operators | 171.1 | 165.0–177.3 | 2,974 |
| 9 | Elementary occupations | 201.7 | 194.5–208.8 | 3,098 |
| Women | ||||
| 1 | Managers | 69.7 | 61.7–77.7 | 302 |
| 2 | Scientific and intellectual professionals | 60.4 | 57.7–63.2 | 1,827 |
| 3 | Technicians and associate professionals | 66.5 | 61.7–71.2 | 749 |
| 4 | Office and administrative staff | 68.1 | 64.2–72.0 | 1,211 |
| 5 | Service and sales workers | 73.8 | 70.8–76.7 | 2,447 |
| 6 | Skilled agricultural workers | 70.6 | 58.9–82.2 | 165 |
| 7 | Craft and construction workers | 69.5 | 61.0–78.1 | 254 |
| 8 | Plant and machine operators | 83.9 | 71.8–95.9 | 187 |
| 9 | Elementary occupations | 78.6 | 75.0–82.2 | 1,902 |
Why the comparison does not measure the effect of changing jobs
The mortality file classifies occupation using administrative information available at the start of the year. People who exit employment, change occupations, or have an unrecorded occupation are not represented in the same way as people who remain employed. The estimates are not adjusted for education, income, health before employment, or other possible confounders. Calculated rates should not be interpreted as the probability that a particular individual will die in the next year.
Exact sources and analytic method
The numerator is the number of deaths in INE Table 32458. The denominator is the number employed at the start of each year, from Table 69961. The nine occupation codes are matched across tables and separated by sex and five-year age groups. Deaths in 2022 and 2023 are summed, as are populations measured at each year’s start; rates are multiplied by 100,000 and directly standardized to the pooled age distribution of employed people. Approximate intervals use Poisson death-count uncertainty. The full workbook reports individual age-band contributions.
Statistical check of the male–female difference. Using the approximate 95% Poisson count-based intervals for each standardized rate reported in the accompanying CSV, we estimated the standard error of each rate as (upper bound − lower bound)/(2 × 1.96), converted it to the log-rate scale by the delta method, and compared the two log-rate ratios. With death counts treated as independent and age-standardization weights held fixed, the male rate ratio is 1.99 (approximate 95% CI 1.88–2.10), the female ratio is 1.30 (1.22–1.39), and the ratio of those ratios is 1.53 (1.40–1.66; two-sided p < 0.001). This establishes an observed difference beyond estimated random death-count variation under the model; it does not test bias from how occupations or employment were recorded, nor does it identify a causal effect of the jobs. These are derived uncertainty estimates, not independently replicated source calculations.
Year-by-year sensitivity. The original workbook reports the following elementary-to-professional age-standardized mortality rate ratios: 2021, 1.77 (men) and 1.38 (women); 2022, 2.07 and 1.44; 2023, 1.90 and 1.18. The comparison remains directionally consistent for each sex across these three years, but varies in magnitude. The 2021 result is shown as an additional check rather than included in the article’s primary 2022–2023 pooled estimate. Reproducible checks are supplied in the ZIP.
Publisher: Mike’s Balance. Version: 1.2, October 12, 2026 (editorial, uncertainty and annual sensitivity updates; original 2022–2023 calculations unchanged). Attribution requested when reproducing derived results: “Mike’s Balance calculations using INE Tables 32458 and 69961 (2022–2023).” Original Spanish public statistics remain credited to INE.
Keep the scale in mind. An age-standardized annual rate of 202 deaths per 100,000 employed people is about 0.2% on a rate scale, not two deaths in every 100 people. Compared with 102 per 100,000, the observed difference is about 100 deaths per 100,000 workers per year. These population rates are neither individual predictions nor estimates of years lost.
The French and Spanish numbers cannot be placed on a single scale. France reports life expectancy for broad social and occupational groups, sometimes using a person’s last recorded occupation. Spain reports annual mortality among people employed at the start of the year. They are two different ways to see occupational inequality, not two estimates of the same causal effect.
What can the job itself do to your health?
Picture two people starting their careers. One completes college and takes a well-paid position in a relatively safe office. The other starts earning straight out of school and spends decades in a factory. Their health at 60 may differ for reasons that began long before either of them arrived at work.
Childhood circumstances, income, access to medical care, and health-related habits all matter. The French education figures show just how tightly social circumstances and health are linked. The relationship also runs in reverse: health influences who enters a job and who can stay in it. Someone with a chronic illness may leave physically demanding work earlier. As a result, the people still doing that work can be healthier, on average, than those who left. Researchers call this the healthy worker effect. A similar selection problem appears in studies of elite athletes: their lower observed mortality does not tell us how much comes from training and how much from the characteristics of people who reach elite sport.
None of this means workplace hazards are imaginary. For some exposures, the evidence is much stronger than anything an occupation ranking can show.
Asbestos. Breathing asbestos fibers can cause mesothelioma and lung cancer. What matters is not simply whether someone is labeled a construction worker. It is whether they encounter asbestos, how much they inhale, and how effectively they are protected.
Excessively long workweeks. Reviews by the World Health Organization and the International Labour Organization associate working 55 hours or more per week with an approximately 35% higher relative risk of having a stroke, compared with working 35–40 hours. That’s an estimate for a particular medical event, not a claim that long hours shorten life by 35%.
Demanding careers and healthy years. A European study estimated remaining life expectancy from age 50 onward. Rather than measuring each person’s working conditions directly, the researchers used descriptions of occupations from the U.S. O*NET database to characterize past European careers. In the resulting model, people with the most demanding work histories were estimated to have roughly four fewer remaining years of life than those with the least demanding. The estimated gap in years of very favorable self-rated health was roughly seven years for men and nine for women.
That last finding needs a translation of its own. The model classified health as favorable only for the self-reported responses “excellent” or “very good.” Even the response “good” fell into the other category. These are modeled expected years, not years of health observed one by one. So nine years does not mean nine additional years with a diagnosed illness or disability. Nor does the study establish that changing jobs would give an individual nine extra healthy years.
Look past the job title
An occupation can conceal very different working lives. One driver keeps predictable hours and gets enough sleep. Another spends excessive hours behind the wheel and rarely feels rested. Two people working in construction can have entirely different exposure to dangerous dust.
If you want to understand the health risks of your own work, the more useful questions are concrete ones: Are hazardous substances present, and is protection effective? How many hours are you actually working? Can you influence your schedule? Is there enough time to recover between shifts?
The mortality gap is measurable. Its cause is not settled by a job title. National statistics show who dies at higher rates across occupational groups, but they cannot tell us how many years the work itself took away. Two people with the same occupation can face very different hazards, hours and opportunities to recover.
That distinction changes the practical question. Instead of asking “Which job will give me five extra years?”, ask “Which exposures and working conditions can be changed?” Reducing dangerous dust or exhausting schedules might protect health without changing anyone’s occupational label. A ranking describes inequality; identifying preventable risks is what could help people live longer.
Sources and methods
- INSEE: Life expectancy by occupational category; INSEE: Differences between managers and blue-collar workers; detailed mortality tables. The 2020–2022 period includes the COVID-19 pandemic. Period life tables describe what would happen under the observed age-specific death rates, not individual forecasts.
- INE: Deaths by occupation, sex, and age and number of employed people. Mike’s Balance calculations: employed people ages 35–64, 2022–2023, with direct age standardization in five-year age bands against a common pooled employed-age distribution. Death counts refer to people classified as employed at the start of each year; the denominator is employed people counted at the same reference date. People can change work status during the year, and some death records have no identifiable occupation. Estimates are not adjusted for education, income, or pre-existing health. The approximate statistical uncertainty does not include possible errors in the administrative records.
- WHO: Asbestos; WHO/ILO: Long working hours.
- Vandenberghe, Career Arduousness and (Healthy) Life Expectancy in Europe, published online in 2025; 2026 journal issue. This observational analysis uses SHARE career histories and U.S. O*NET occupational descriptors. The estimates do not establish causation.
- Vinni and Hakama: The healthy worker effect.