The Structure of Paid and Unpaid Work

This Topic's central mechanism is a form of work that most measurement systems do not count.

It is enormous — plausibly comparable in value to a substantial fraction of the recorded economy. It is unequally distributed in a pattern that has narrowed and not closed. And its distribution inside households produces most of the inequality observed outside them, which is why a lesson about who loads the dishwasher belongs in a Part about inequality.

Two identical weeks, and what a time diary records.

Priya and Sam do the same job at the same firm, for the same salary, on the same contract. On any measure your employer collects, they are identical.

Here is what a time-use diary — the instrument that asks people to record what they were doing in ten-minute intervals through a day — records about their weeks.

Total hours of work, paid and unpaid combined, are close to equal. This is a real and frequently misreported finding: in most rich countries the total burden is similar. The composition is not.

Sam's unpaid work is largely deferrable. He does most of the household's repairs, assembles things, deals with the car, mows in summer, and does a substantial share of the enjoyable end of childcare — the park, the football, the bedtime story. Almost none of it has to happen at a particular hour, and none of it happens while he is doing something else.

Priya's unpaid work is largely non-deferrable and non-discretionary. Meals happen three times a day whether or not she is busy. Uniforms have to be clean by Monday. The child who is ill is ill now. Her diary shows far more entries with two activities recorded at once, which is what fragmentation looks like when you measure it.

And the diary misses the largest part.

Priya knows that the school shoes are nearly outgrown, that the form is due on Thursday, that the childminder is away next month, that the grandmother's prescription needs collecting, that the child who has gone quiet needs asking about. She anticipates, monitors, remembers and decides, and this is work — recent research separating the cognitive dimension of household labour finds that anticipation and monitoring are the most unequally distributed components of all, more so than the physical tasks.

None of it appears in a time diary, because you cannot record "carrying the schedule of a household" in a ten-minute box.

Now the consequence, which is the whole of this Topic.

Their employer announces a project requiring three months of unpredictable hours and travel. It is the route to promotion.

Sam's time is protected, because someone else absorbs the contingencies. Priya's time is interruptible, because she is the contingency absorber.

Neither of them is discriminated against. Neither wants a different arrangement strongly enough to have changed it. And five years later their salaries differ by a third, through a sequence in which no employer did anything and no rule was broken.

An activity that is essential, enormous, and excluded from the accounts by convention.

National accounts have a production boundary — a rule about what counts as economic production. Goods and services produced within a household for its own consumption fall outside it, with a few exceptions.

So if you pay a nursery, national output rises. If you look after the child yourself, it does not. If you employ a cleaner, output rises; if you clean, it does not. A society in which more of this work is marketised appears to grow, and one in which more of it moves back into households appears to shrink — with no change in the quantity of work done or the wellbeing produced.

How large is the excluded sector? Statistical offices produce satellite accounts that value it, and the estimates depend entirely on the method (see 7.2.2). Valuing at replacement cost — what it would cost to hire someone — gives one figure; valuing at opportunity cost — the wage the person forgoes — gives another, higher one for higher earners and a lower one for the poor.

Across methods and countries, the estimates run from something like a sixth of measured GDP to considerably more than half. The range itself is the point: the largest single sector of economic activity is measured with an uncertainty of several hundred per cent, because nobody has needed the number.

And the feminist critique of this is not that the accountants were malicious. It is 7.2.2's argument: what we measure shapes what we pursue and what counts as an economy — and a boundary drawn where this one is drawn makes the work done disproportionately by women definitionally not work.

What the evidence shows

Four findings from time-use data, which is the best instrument in this area.

One — the gap narrowed substantially, and mostly from one side.

Since the 1960s, the difference between men's and women's unpaid household work has fallen a great deal in rich countries. The larger part of that convergence came from women doing less rather than men doing much more — through smaller households, appliances, marketised services, lower standards, and reduced time on routine housework.

Men's contribution rose too, and the rise was concentrated in childcare rather than housework. Fathers' time with children has increased markedly across several decades; time on cleaning, laundry and cooking has moved much less.

Two — the composition differs more than the totals.

Routine, daily, non-postponable tasks are done disproportionately by women. Intermittent, deferrable, discretionary tasks disproportionately by men. This matters because the two have different consequences for the availability of a block of protected time — which is what a career requires.

Three — the cognitive component is the most unequal and the least measured.

Anticipating what will be needed, identifying options, deciding, and monitoring that it happened. Research separating these finds that the anticipation and monitoring components are distributed considerably more unequally than the execution — and that they are the components least visible to the person not doing them, which is why they generate a particular kind of household argument.

Four — the leisure gap is smaller than often claimed and the quality differs.

Total leisure time differs by less than people expect. What differs is whether it is contaminated — leisure taken while also supervising, listening for, or being available to someone. Diary data can capture this through secondary activities, and it shows the same pattern as everything else.

From the household to the labour market

Occupational segregation, and the finding that runs the wrong way.

Horizontal segregation — women and men concentrated in different occupations. Vertical segregation — concentrated at different levels within them.

The standard measure is an index of dissimilarity: the proportion of one group who would have to change occupation for the distribution to match the other's. It has fallen in most rich countries and it has fallen slowly, with much of the decline concentrated in earlier decades and in professional occupations, while manual and care occupations remain heavily segregated.

And here is the finding that changes how segregation should be understood.

The intuitive account is that women enter lower-paying occupations. The alternative account is devaluation: that occupations lose relative pay after they become female-dominated, because work done by women is valued less.

These are opposite causal directions and they are separable with panel data on occupations over time. Analyses tracking American occupations across several decades find evidence consistent with devaluation — occupations that shifted towards female majority experienced relative pay decline afterwards, with a lag, net of skill and other characteristics.

The mechanism is not that anyone reduced wages on the day the composition changed. It is that pay in an occupation reflects an assessment of its worth, that assessment is socially made, and it moves with who does the work. Which is Fraser's misrecognition and maldistribution operating together (see 8.1.3), and it is measurable.

Greedy work: why the gap is large in some occupations and small in others.

One of the most useful recent accounts is deliberately narrow, and it explains a great deal.

Some jobs pay disproportionately for long, particular and unpredictable hours. Working sixty hours pays more than one and a half times what forty pays. Being reachable at short notice, being the person who holds the client relationship, being present for the crisis — these are rewarded steeply. Call these greedy jobs. Law, finance, consulting, senior management and several medical specialties are the standard examples.

Other jobs pay roughly linearly in hours, and workers are substitutable — one qualified person can hand over to another with a good record and no loss to the client.

The prediction is precise: the gender earnings gap should be large in the first kind and small in the second, because in the first, the person who cannot guarantee unbroken availability is heavily penalised, and in the second they are not.

And the prediction holds. The canonical case is pharmacy — a highly paid, highly qualified profession with an unusually small gender earnings gap, and the explanation offered is structural: pharmacists became substitutable as independent pharmacies gave way to chains and hospitals, records became standardised, and part-time work stopped carrying a large hourly penalty.

Why this matters more than it looks. It relocates the problem from individuals to the structure of jobs. A couple in which one person takes the greedy job and the other takes the flexible one is making a jointly rational decision that maximises household income — and it produces an unequal outcome. The lever is not the couple's preferences. It is the convexity of the pay-hours relationship, which is a feature of how work is organised and is not fixed by nature.

The child penalty, which is the most important quantitative finding in this Topic.

A design worth admiring (see 7.4.3). Take administrative records for a whole population, identify the birth of a first child, and plot earnings for mothers and fathers in the years before and after — an event study in which each parent's own pre-birth trajectory is the counterfactual.

The result is stark and has replicated in every country where it has been run. Before the first birth, mothers' and fathers' earnings trajectories are close to parallel. At the birth, mothers' earnings drop sharply. Fathers' do not move. And the mothers' path does not recover to the counterfactual — it flattens permanently, a decade or more later.

In Denmark, where the analysis was first done at scale, the long-run child penalty was estimated at around a fifth of earnings — and, more strikingly, it accounted for roughly four-fifths of the remaining gender earnings gap by the 2010s, up from around two-fifths three decades earlier. The other components of the gap closed. This one did not, so it grew as a share of what remained.

Cross-country replication shows very large variation — penalties several times larger in some countries than others, with German-speaking countries at the high end and the Nordic countries lower, and no country at zero.

The penalty is composed of hours, participation, occupation switching, employer switching towards more flexible and lower-paying firms, and forgone promotion. All five are documented.

And on the causes, one finding is worth carrying because it is a norm claim with evidence behind it. The size of a woman's child penalty is associated with her own mother's labour supply when she was growing up — women whose mothers worked more experience smaller penalties, net of other characteristics. That is intergenerational transmission of a norm, measured, and it is not something a story about biology or about pure economic optimisation predicts.

And the policy evidence is more mixed than the advocacy on either side.

Long parental leave available to mothers can worsen women's careers, by extending time out of employment and weakening attachment to the employer — several evaluations find this, and the effect is worst for the longest leaves.

Father quotas — leave reserved for the second parent on a use-it-or-lose-it basis — do increase fathers' take-up substantially, because that is what they are designed to do. Their effects on the longer-run division of household labour and on mothers' earnings are real but modest in most evaluations, and smaller than their advocates predicted.

Childcare availability and cost show the more consistent effects on maternal employment, particularly for lower-earning mothers, and particularly where provision is affordable and full-time.

And no country has eliminated the penalty, including those with the most extensive policies — which is the fact that any confident policy claim has to accommodate.

The honest summary: policy moves the size of the penalty substantially — the cross-national variation proves that — and no existing package removes it. Which suggests that the remaining component is in the structure of jobs and in norms, both of which are harder to legislate than leave entitlements.

And a note on the household, which 8.1.2 flagged and this lesson can now settle.

Measurement treats the household as a unit that pools its resources. The evidence says otherwise, and there is a clean natural experiment.

When a child benefit was shifted from being paid through the father's pay packet to being paid directly to the mother — a change in who received the same money, with no change in the amount — household spending patterns changed, with increased expenditure on children's and women's clothing relative to men's.

This should not happen if households pool. It is direct evidence that who receives income affects how it is spent, which is why the household is a site of distribution rather than a unit of it — and why every measure that equivalises household income and stops there is making a substantive assumption it has not tested (see 8.1.2).

Because "women choose differently" is where most public discussion of this stops, and it is where the analysis should start.

The statement is true and it explains nothing on its own, because every choice in the story was made within a structure: a pay-hours relationship that is steeply convex in some jobs and not others; a household in which one person's time was already the contingency reserve; a leave system with particular entitlements; a childcare market with particular prices; and a norm learned from the previous generation. Change any of those and the choice changes, which is what the cross-national variation demonstrates.

Four questions for anything you read about this.

Is the pay gap quoted raw or adjusted — and if adjusted, for what? Controlling for occupation and hours removes the two largest channels (see 8.4.2 on the same error).

Is unpaid work counted at all? If the analysis compares only paid hours, it has excluded the mechanism.

Does the evidence come from a design with a counterfactual? The event-study child penalty is worth more than a hundred cross-sectional decompositions, for exactly the reasons 7.4.3 gives.

And does the account explain the variation? The penalty differs several-fold between rich democracies with similar technology, similar education and similar attitudes. Any explanation that appeals only to biology or only to universal preferences has to explain that, and none does.

One last thing, which is the reason this lesson comes before the pay gap rather than after it. The pay gap is a symptom. The distribution of non-deferrable, non-discretionary, cognitively demanding unpaid work is the mechanism — and it is invisible in every dataset the labour market produces.

Total work is roughly equal between men and women in rich countries; the composition is not. Women's unpaid work is disproportionately routine, daily and non-postponable; men's is disproportionately deferrable and discretionary — and the difference is in whether a protected block of time exists, which is what a career requires. The cognitive dimension — anticipating, identifying, deciding and monitoring — is the most unequally distributed and the least measurable.

The convergence since the 1960s came mostly from women doing less, with men's increase concentrated in childcare rather than housework.

Unpaid household production falls outside the national accounts' production boundary by convention, so marketising it appears as growth. Valuations range from around a sixth of GDP to more than half depending on method — an uncertainty of several hundred per cent in the largest sector of economic activity.

Occupational segregation has fallen slowly. And the causal direction runs partly the unexpected way: panel evidence on occupations over decades is consistent with devaluation — occupations losing relative pay after becoming female-dominated, net of skill.

Greedy jobs pay disproportionately for long, particular and unpredictable hours and penalise substitutability; the gender earnings gap is large where hours are convex and small where they are linear and workers are interchangeable, with pharmacy the canonical low-gap high-pay case. The lever is the shape of the pay-hours relationship, not the couple's preferences.

The child penalty is the central quantitative finding. Event studies around first birth show mothers' earnings dropping sharply and never recovering while fathers' do not move; in Denmark it came to account for around four-fifths of the remaining gender earnings gap, because the other components closed and this one did not. It varies several-fold across countries and is nowhere zero. And its size is associated with the woman's own mother's labour supply — a measured norm effect.

Policy moves it and does not remove it: long maternal leaves can harm careers, father quotas raise take-up with modest longer-run effects, and childcare availability has the most consistent effects on maternal employment.

And households do not pool: when a child benefit was redirected from fathers to mothers with no change in amount, spending patterns changed.

Time-use diary — recording activities in short intervals, including simultaneous ones.

Non-deferrable work — tasks that must happen at a particular time, which fragment the available block of time.

Cognitive household labour — anticipating, identifying options, deciding and monitoring; the most unequally distributed component.

Production boundary — the national accounting rule excluding household production for own consumption.

Satellite account — a valuation of excluded production, sensitive to the replacement-cost or opportunity-cost method.

Horizontal / vertical segregation — concentration in different occupations; at different levels within them.

Index of dissimilarity — the proportion of a group who would have to move for two distributions to match.

Devaluation — the decline in an occupation's relative pay after it becomes female-dominated.

Greedy work — jobs with convex returns to long and unpredictable hours and low substitutability between workers.

Child penalty — the persistent drop in mothers' earnings following a first birth, absent for fathers; measured by event study against each parent's own pre-birth trajectory.

Father quota — leave reserved for the second parent on a use-it-or-lose-it basis.

Household pooling assumption — the untested premise that resources are shared within households.

One — keep a diary for two days. Record what you did in half-hour blocks, including anything you did simultaneously. Then mark each entry as deferrable or not, and see which column is longer.

Two — list the cognitive load. Write down everything you are currently carrying in your head that someone in your household would have to be told if you were away for a week. Then ask who else could produce that list.

Three — find the convexity. For a job you know, work out whether sixty hours pays more than one and a half times what forty pays, and whether one person can substitute for another. That answers most of what the gender gap will be in that occupation.

Four — look up the penalty. Find an event-study estimate of the child penalty for your country and one other. The difference between them is what policy and norms account for.

Five — test the pooling assumption. In any household you know well, work out who decides what about money. Then ask whether treating it as one purse would describe it accurately.