When the Middle Shrinks Before Workers Are Displaced
Occupational polarization is the subject of this series: the long-run expansion of high-skill abstract work and, to a lesser extent, low-skill manual work alongside the contraction of middle-skill routine work. The question is not simply why some occupations grow while others shrink. It is how that visible pattern is produced, who bears its costs, and which alternatives are actually available when work changes. The reader will gain a way to distinguish a changing occupational stock from the worker flows beneath it, then use skills and local vacancies to judge whether a proposed transition is real rather than merely plausible on paper.
That distinction matters because the same employment chart can describe very different lives. A shrinking middle might mean that established routine workers are being pushed downward into manual work. It might mean that they are moving upward into abstract work while new entrants are no longer being hired into routine jobs. It might reflect more exits into non-employment, or some combination of these processes. Those routes do not carry the same implications for wages, training, social mobility, or policy. Protecting incumbent workers from displacement is not the same task as opening a viable first job or re-entry route for people who can no longer enter the occupation that once absorbed them.
The pattern is visible before the mechanism
The broad pattern is often called a U-shape. In a labour market divided roughly into manual, routine, and abstract occupations, the routine share falls while the shares at the lower and higher ends rise. The pattern is intuitively associated with automation or technology, but the intuition is incomplete. A chart of employment shares tells us where workers are observed at two points in time. It does not tell us which workers moved, which workers never entered, or what happened between the two observations.
The French evidence examined later in this series makes the distinction concrete. In the DADS–EDP administrative panel, the within-employment routine share fell from 69.0% in 1984 to 53.0% in 2021. Over the same period, the abstract share rose from 13.7% to 26.4%, and the manual share from 17.2% to 20.7%. Those figures establish a large change in the composition of employment. They do not, by themselves, establish mass downward displacement from routine work.
The timing also matters. Before 1994, routine employment was broadly stable, manual employment declined slightly, and abstract employment grew at a moderate pace. After 1994, routine employment fell more quickly, manual employment changed direction, and abstract employment grew faster. The break is a useful clue because it asks us to identify which route changed. It is not yet an explanation. The same date could mark a change in hiring, separations, technology, education, policy, or the composition of people entering the labour market.
A useful mental model is to treat an occupational share as the water level in a tank. The level can fall because water flows out, because less water flows in, or because the other tanks grow faster. In a labour market, the equivalent flows include direct moves from one job to another, exits into non-employment, entries and re-entries from non-employment, and changes in the total amount of employment. Looking only at the level makes those routes indistinguishable.
One pattern, several worker experiences
Consider two simplified labour markets with the same decline in routine employment. In the first, long-established routine workers lose their jobs and move into lower-paid manual work. In the second, routine workers continue to move into abstract jobs at roughly the same rate as before, but fewer people entering or returning to employment are hired into routine work. The routine share can fall by a similar amount in both cases. Yet the workers exposed to the adjustment are different: incumbents in the first market, and prospective or returning workers in the second.
This is why the entry margin is not a statistical footnote. “Entry” here should be understood broadly. It includes people entering or re-entering observed employment from a non-employment state, not only young people taking their first job. If the composition of those entrants changes—because employers demand different capabilities, because a familiar occupational ladder narrows, or because other jobs become cheaper or more attractive—the occupational structure can polarize without a wave of incumbent downgrading.
The distinction also changes what counts as an informative policy. If incumbent displacement is the main mechanism, income support, job protection, and rapid re-employment may be central. If the pipeline into routine work is narrowing, those tools may not address the main bottleneck. Education, early-career preparation, re-entry support, counselling, and routes into adjacent occupations become more important. This does not make incumbent protection irrelevant; it means that a policy aimed at one flow should not be mistaken for a response to another.
The evidence must also be read at the right level. Worker-flow accounting can show which observed channels line up with the change in occupational shares. It does not automatically identify the ultimate cause. A stable upgrading flow may coexist with a changing hiring flow. A technology-price trend may coincide with a break in routine employment without proving that it caused the break. Good analysis keeps the measured mechanism, the candidate explanation, and the policy judgment separate.
The diagram captures the central reframe. The aggregate pattern is an outcome that several channels can produce. A proposed alternative job adds another layer of uncertainty: even when two occupations share relevant skills, the destination must have vacancies, and those vacancies must be accessible to workers who may be unable or unwilling to move region.
Why the question extends beyond one country
The entry-margin idea becomes more useful when placed beside a broader question: what happens when the jobs available in a local labour market change? Job-to-job mobility is a normal part of a dynamic economy. The OECD paper reports that hirings and separations together amount to about 22% of employment each year across OECD countries, with roughly half reflecting direct moves between jobs without an intervening period of unemployment. Such moves can improve the match between workers and firms, support career progression, and reallocate labour toward more productive employers.
But mobility is not frictionless. Across countries with available data, only around 2% of newly hired workers also change region on average, with substantial variation. The share is below 0.2% in Italy and close to 5% in Sweden. A worker may have a skill that is valuable elsewhere and still remain in the current local labour market because housing, family responsibilities, transport, information, or preference make relocation difficult. A national vacancy count can therefore exaggerate the opportunity available to a particular worker.
The green transition is a particularly clear setting in which these constraints meet. Emissions-intensive activity is often concentrated in particular regions and large plants. A sector can account for a modest share of national employment while representing a major source of work in one place. At the same time, new jobs involving green tasks may appear in different regions, may require a wider bundle of skills, or may be filled by workers coming from occupations that are not themselves vulnerable.
That means “move from a vulnerable job into a green job” is not a complete transition model. Some workers may stay with the same employer while their tasks change. Some may move across sectors into a non-green occupation whose skill requirements are closer. Some may reach a green-task occupation after targeted training. Others may face a local shortage of suitable vacancies even when the national economy appears to offer many alternatives. The relevant question is not whether a green occupation exists somewhere. It is whether a worker in a specific origin occupation can reach a suitable destination in the labour market that actually governs their choices.
The useful question is not only “where did the jobs go?”
The two research programmes provide complementary pieces of that question. The French worker-flow study asks how a changing occupational structure is assembled from job-to-job movement and flows through non-employment. Its central result is a qualified one about France: polarization after 1994 is associated mainly with altered entry into routine work, while routine-to-abstract upgrading remains active. The OECD study asks how to identify plausible destinations when skills, vacancies, and place all matter. Its answer is a skills-first, place-based approach, not a promise that every high-overlap occupation is accessible.
Neither source supports a universal law. The French study concerns salaried employment in one country and uses an observed/non-observed employment state with important measurement assumptions. The OECD framework draws on online job postings and labour-force data from Italy, the United Kingdom, and the United States; its occupational-proximity measure is a prediction tool, not a certificate of competence or a causal estimate of hiring. These boundaries are not minor qualifications. They determine what a reader can responsibly infer from the evidence.
The durable lesson is therefore narrower and more useful than the claim that technology simply destroys middle-skill work. When the middle shrinks, first ask which flow changed: incumbent upgrading, incumbent downgrading, exit, or entry and re-entry. Then ask whether the people affected have a skill-related destination, whether that destination has local demand, and what prevents the match from occurring. The next step is to make those flows visible in the French data, so that the apparent disappearance of routine work can be separated from the changing pipeline into it.