The Pipeline into Routine Work Thins
With the French stock-flow frame in place, the decisive question is which channel changed after 1994. This chapter tests three possibilities: routine workers may have been pushed into manual work, the ladder from routine to abstract work may have accelerated, or fewer people may have entered routine from non-employment. The distinction matters because each explanation identifies a different group as the main bearer of structural change. The French evidence points most strongly to the third mechanism, while also showing why the cause of that shift must remain a qualified question.
The upgrading ladder did not suddenly switch on
The first result is easy to miss because it runs against the most familiar story. Routine-to-abstract upgrading was already a large flow before the polarization break, and it remained large afterward. In the paper's decomposition, the routine destination loses roughly 0.54 percentage points of employment per year through its net flow with abstract work before 1994, and roughly 0.52 points afterward. The sign is a drain from routine and a gain for abstract work, but its stability is the important fact.
If the post-1994 routine decline were primarily caused by an acceleration of incumbent routine workers moving into abstract jobs, the routine-to-abstract drain should become much larger at the break. It does not. The ladder continues to operate at about the same aggregate rate. Workers already in routine occupations are still moving upward, and the flow is not a new event that appears only after polarization begins.
A second comparison is more revealing. Before 1994, the net flow from manual into routine added about 0.20 percentage points per year to routine's share. During the polarization period, that buffer falls to about 0.08. Manual-to-routine movement still helps routine, but it offsets less of the routine-to-abstract drain. The direct job-to-job component therefore becomes more negative after 1994, not because upgrading to abstract suddenly accelerates, but because a supporting flow from manual work weakens.
The third change occurs outside the employed occupations. Routine's excess non-employment flow is positive before 1994, about +0.25 percentage points per year, and negative afterward, about −0.09. Before the break, non-employment acts as a net feeder into routine relative to the amount needed to preserve its share. Afterward, it no longer replenishes routine. The combination is decisive: a stable drain upward, a smaller buffer from manual work, and a non-employment channel that stops feeding the middle.
This is why a stock chart alone is misleading. The routine share can fall even while a routine worker's route upward remains available. The composition changes because the balance of flows around that worker's occupation changes. The next step is to separate the two directions within the non-employment channel rather than treating its net value as a single event.
Entry, not exit, drives the non-employment change
Routine can lose its non-employment buffer for two broad reasons. More routine workers might leave for non-employment, or fewer people in non-employment might enter routine. Both would reduce routine's net flow, but they describe different labour-market problems. The first is an exit problem among people already in routine work. The second is an entry problem affecting people trying to obtain or regain a foothold in routine work.
The study compares both flows with share-neutral benchmarks. For entry, the benchmark is the amount of inflow routine would receive if non-employed entrants were allocated in proportion to routine's existing employment share. For exit, the benchmark is the corresponding share of departures that would be expected if routine workers left at a share-neutral rate. The comparison asks whether routine is over- or under-represented in each channel, rather than simply whether the raw number of entries or exits is high.
The entry deviation becomes more negative across the break: it moves from about −0.03 before 1994 to −0.06 afterward. Routine receives a smaller fraction of the non-employed inflow than its share-neutral benchmark would imply. The exit deviation, by contrast, remains about −0.06 in both periods. Routine is not newly overrepresented among exits to non-employment at the break; its relative exit pattern is essentially unchanged.
The paper decomposes the change in the net non-employment channel into a benchmark effect and a deviation effect. Because routine's share and overall entry and exit rates both fall, the changing size of the benchmark has offsetting effects on the entry and exit sides. Those effects nearly cancel in the net. The remaining change is a widening entry deviation, not a new exit deviation.
The counterfactual makes the diagnosis more intuitive. In the polarization period, routine's employment share declines at about −0.52 percentage points per year. If the exit deviation were held at its pre-1994 value while the other changes were left in place, the decline would be almost unchanged, around −0.53. If the entry deviation were held at its pre-1994 value instead, the decline would be less than half as large, around −0.21. Removing the change in entry selection preserves a replenishing non-employment channel; removing the change in exit selection does not.
The paper therefore attributes about four-fifths of the post-1994 acceleration in the routine decline to the entry side of the non-employment channel. This is a French accounting result, not a claim that no incumbent routine worker lost a job. Individual displacement can occur without being the dominant net explanation for the changing occupational share. The evidence says that the aggregate mechanism is more consistent with routine being passed over by entrants and re-entrants than with a mass of routine incumbents being pushed downward.
The state diagram shows why the word “polarization” should not be allowed to imply one universal route. The routine-cognitive channel loses its replenishment from non-employment, while the routine-manual channel was already contracting for a different reason. At the same time, the main upward ladder remains active. The arrows describe net patterns across periods, not a deterministic path for every individual worker.
Why the manual rise is not a hidden routine downgrade
The manual share rises after 1994, which can appear to confirm a story in which routine workers are displaced downward. But the flow decomposition does not show that as the main net channel. Manual employment receives a positive excess flow from non-employment. Direct net flows from manual toward the other employed occupations are negative for manual, meaning that direct job-to-job movement does not account for its rise. The manual increase is therefore largely an entry-side phenomenon as well: people entering employment are disproportionately absorbed into manual work.
A routine worker can still move into manual work. The claim is not that the transition is impossible or that every person remains on an upward path. The claim is narrower: direct routine-to-manual downgrading is not the dominant aggregate explanation for the post-1994 change. That distinction is strengthened by the study's treatment of indirect paths. A worker who leaves routine for non-employment and later returns in manual work would not appear as a direct routine-to-manual move in a one-year matrix. The researchers therefore examine two-step paths through non-employment.
Those indirect flows are net upgrading across the bilateral pairs they examine, including the routine–manual pair. In other words, the non-employment channel does not reveal a concealed pattern in which routine workers systematically pass through a jobless spell and reappear in manual work. The evidence cannot rule out individual experiences of that kind, but it does not support them as the dominant net route in the French data.
This matters for interpretation of the lower tail. Manual employment is growing, but “manual growth” does not automatically mean “routine workers were downgraded into manual jobs.” The same stock outcome can result from a larger inflow of people who were previously outside observed employment. The entry margin changes the social meaning of the chart: the question becomes not only who left routine, but where new and returning workers were absorbed.
The break lives on the cognitive side of routine
The broad routine category combines two groups with different histories. Routine-manual occupations—blue-collar production and related work—were already declining before 1994 and continued to decline afterward. Routine-cognitive occupations—white-collar clerical, administrative, and supervisory work—grew before the break and then turned downward. The pre-1994 stability of total routine employment was therefore a cancellation: routine-cognitive growth offset routine-manual contraction. After 1994, the two components moved in the same direction, so the aggregate middle began to shrink persistently.
The entry-side contrast is large. Routine-cognitive excess non-employment flows fall from about +0.80 percentage points per year before 1994 to roughly +0.04 afterward. The replenishing flow almost disappears. Routine-manual excess non-employment is already strongly negative before 1994, around −0.56, and becomes less negative, around −0.12, afterward. The aggregate routine result is thus driven by a collapse on the cognitive side partly offset by an improvement in the manual side.
The upward ladder also has a cognitive centre. Roughly three-quarters of the routine-to-abstract drain runs through routine-cognitive work. The paper's more detailed occupation pairs point to intermediate administrative occupations moving into higher-level abstract occupations, while the manual-to-routine buffer is also concentrated in movement from service work into routine-cognitive jobs. The occupational structure is therefore not a single ladder with a uniform middle. It contains a cognitive route that historically served both as an entry point and as a stepping stone, and that route loses its ability to absorb new workers after 1994.
That specificity makes the evidence more informative, but it also limits the claim. Routine-cognitive and routine-manual are groupings based on French occupation codes. They are useful for identifying different patterns of task exposure and movement, not proof that every clerical, supervisory, or production job experienced the same technology or demand shock. The data reveal where the aggregate break is concentrated; they do not observe every firm's adoption decision or every worker's reason for moving.
ICT prices fit the timing, but timing is not causation
The paper discusses falling information and communication technology prices as a plausible trigger for the routine-cognitive break. Communications prices in France peaked around 1994 and then fell as the telecommunications market opened to competition. Computer equipment had already been getting cheaper for two decades, but the decline accelerated after 1994. The timing matches the change in routine-cognitive employment, and the tasks most directly exposed—spreadsheets, word processing, databases, and accounting systems—fit the occupational specificity of the break better than a story about production work alone.
That is a coherent hypothesis. It is not a completed causal identification strategy. A price series that changes when employment changes can be a trigger, a response, or a marker for several institutional changes happening together. The paper also notes French employer social-contribution relief on low wages beginning in 1993. That policy could have contributed to the rise of manual employment and may have affected which low-wage jobs were attractive to entrants. Its effects need not compete with the ICT story: one mechanism could reduce demand for certain routine-cognitive tasks while another changes the relative cost of low-wage manual hiring.
The authors leave the two explanations to future work rather than treating one as settled. A stronger causal test would need to distinguish the smooth, technology-related price trend from the discrete timing of policy-schedule changes and show why the effects appear where the flow decomposition says they do. It would also need to confront other changes in education, labour-force composition, trade, and firm organisation.
The secure conclusion is therefore layered. The French data show a post-1994 polarization pattern. The flow accounting shows that the changing routine share is driven primarily by the collapse of the non-employment-to-routine entry channel, with a smaller contribution from the weakened manual-to-routine buffer, while routine-to-abstract upgrading remains stable. The concentration in routine-cognitive work makes ICT a plausible explanation, but plausibility is not proof. The next task is to consolidate this evidence before asking how worker characteristics and local labour demand determine who can use the ladders that remain.