Two Greek thinkers examine an insurance fund, wage envelope, leave calendar and parent-and-child figures beside a medical bag.

Paid Family Leave

The United States guarantees twelve weeks of unpaid, job-protected leave to about half its workers and pays for none of it, which makes it the only high-income country without national paid maternity leave. States have filled part of the gap with their own programmes. Senator Kirsten Gillibrand and Representative Rosa DeLauro have proposed a national insurance in Congress for years ^[1]. Since July 2026 a second bill, from Gillibrand and the Republican Senator John Boozman, has been before the Senate: grants to states that run their own programmes ^[16].

1 in 3 people who take family or medical leave are paid nothing for it ^[2]^[11]
14 states already run their own paid leave programme ^[4]
7.4 million people needed leave and kept working through a serious illness instead ^[2]

The proposal

Workers get up to twelve weeks of paid family or medical leave from a national insurance, funded by a small contribution on earnings.

The debate

In Congress the argument is less about whether workers should have paid leave than about who runs it and how it is paid for. Democrats want a national insurance funded from every pay packet, modelled on the state programmes. Republicans warn against a new payroll tax, a federal mandate and public money replacing leave that employers already pay, and favour tax credits, flexible state programmes or grants to states.

“Without universal paid leave, millions of Americans are forced to make devastating choices between their livelihood and the health of themselves or their families.”

argues for: Kirsten Gillibrand, Democrat, New York Press release, 16 September 2025
“We’re not mandating a one-size-fits-all federal program, we’re building a flexible federal strategy that supports state innovation and addresses the real challenges families face today.”

argues against: Stephanie Bice, Republican, Oklahoma Press release, 5 May 2025
“But if all we are doing is supplanting that which is being paid by Google already, I am not sure that that is the best deal for the taxpayer.”

argues against: Bill Cassidy, Republican, Louisiana Senate Finance subcommittee hearing, 11 July 2018

The analysis

Paid Family Leave

55% of the weight is for, 45% against.

An argument's weight combines three questions: how big the effect is, how certain it is and how much the good at stake matters. The larger the slice, the more the argument weighs.

For

Against

Paid family leave is a good proposal on the current analysis.

Very likely better

Very likely better: Under the assumptions used, the proposal brings more than it costs in 96 of 100 model runs.

Overall effect: large

How strongly the proposal works overall, in five steps from very small to very large, comparable across all debates. The ring is always full, even when the overall effect is small; this step says how large it is.

The analysis sides with the supporters, and for the reason they give most often: the benefit arrives in the weeks a household has no income, and the contribution comes mostly from the best-paid. The strongest doubt gets far less attention than the payroll contribution. California's programme left first-time mothers less likely to be employed a decade later, and that finding is what could change the result.

In detail

All arguments

Arguments for4 arguments

Income in the weeks there is none

29points

About a third of Americans who take family or medical leave are paid nothing for it, and among workers earning under 15 dollars an hour it is three in five. The benefit arrives in the weeks a household has lost its earnings, and most of the new money reaches the lower and middle earners who have no paid leave today.

Value 5 · Household budgetsImpact 6.2Plausibility 9.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money arriving in a household during weeks when its earnings have stopped, and money is priced at the middle of the scale wherever it appears. What distinguishes this from an ordinary transfer is timing rather than class: the same household pays the contribution in the weeks it is earning and draws the benefit in the weeks it is not, so the gain is the difference between what a dollar is worth at those two moments. That difference is carried in the Impact, and the contribution appears against this argument at the ordinary weight. Nothing is counted here for the relief of not having to choose between a sick parent and a pay cheque, which is the same fact as the money under a different name. Nothing is counted for the leave itself, which is a separate argument. The value is the middle of the scale, and the whole case for this argument is that a dollar does not weigh the same in every week of the year.

The Labor Department's 2018 survey finds that 34 percent of those on leave are paid nothing and a further 24 percent only part of their wage; among workers earning under 15 dollars an hour, 61 percent are paid nothing, against 20 percent of others ^[2]^[11]. Twenty-seven percent of the workforce already lives in a state with its own insurance programme, so a federal one is new for about 119 million of the 163 million people in work ^[4]^[10]. Washington State's programme paid about 2.01 billion dollars in benefits in 2025, roughly 490 dollars per covered worker ^[7]; it is more generous than the bill, so 420 dollars is used, in a range from 350 to 575. On 119 million workers that is 50 billion dollars a year. Not all of it is new money for households: where an employer already paid for the leave, the insurance takes over the bill. Sixty percent is taken as new household income, in a range from 45 to 75 percent, which is 30 billion dollars a year. Who receives it decides its weight. In Washington the recipients spread across all five wage fifths, and the new money goes mostly to those unpaid today, who are disproportionately lower earners; at this site's weights for each fifth that gives 1.07. The money also arrives in the year a household loses weeks of earnings, which raises its worth by about a seventh: 1.2 in all, in a range from 1.0 to 1.4. The result is a Normalised Impact of 6.21 on the scale of this evaluation, in a range from 4.93 to 8.97. The Impact is the largest in this debate; set against the same money on the paying side, only the difference in where it lands counts.

▸ Show calculation ▾ Hide calculation

	Workers not already covered by a state programme [4][10]	163 million employed, 27 percent already covered	119 million workers
×	Benefits paid per covered worker a year Setting, range 350 to 575 dollars: Washington's programme paid about 2.01 billion dollars in 2025, about 490 dollars per covered worker; it replaces 90 percent of a low wage and runs to 16 weeks, more generous than the bill, so a sixth is taken off [7]	420 dollars	50 billion dollars a year
÷	In euro-equivalent the rate used throughout this evaluation	1.16 dollars to the euro	43.1 billion euro-equivalent a year
×	Share that is new money for the household Setting, range 45 to 75 percent, without a source of its own: where an employer already paid for the leave the insurance takes over the bill and the household gains nothing	60 %	25.9 billion euro-equivalent a year
×	Weight of a dollar for the households receiving it Setting, range 1.0 to 1.4. Who gains: in Washington's programme 8, 19, 27, 27 and 18 percent of recipients come from the five wage fifths, lowest first, and the new money goes disproportionately to lower earners, of whom 61 percent are unpaid on leave today against 20 percent of others; at this site's weights for the five fifths that gives 1.07. The benefit also arrives in the year a household loses earnings, which raises what it is worth by about a seventh more [7][11]	1.2	31.06 billion euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	6.21

Score 6.21 Impact × 5 Value × 9.5 Plausibility ÷ 10 = 29 of 100

That the benefit is paid follows from the bill once it is law: anyone who meets the conditions draws it, so the argument does not depend on a behavioural response for its occurrence. The counterfactual is the same workers under the leave law as it stands, unpaid unless their employer chooses otherwise. How much is paid does depend on behaviour — how many people take leave and for how long — and that rests on a controlled comparison with the states that already run such insurance, which differ in industry mix, unionisation and female labour force participation; that doubt sits in the range of the amount rather than here. Reverse causation is not a concern for a benefit set by statute. What remains is the risk that a new federal programme is administered badly in its first years, which is why the value is 9.5 rather than 10. The Plausibility is high: the payment follows from the law, and only its size is estimated.

Evidence: Mechanism · plausibility at most 9.5 Evidence of effect: Definitional

Leave nobody could afford

2.7points

In the Labor Department's 2018 survey about 7 percent of employees needed leave and did not take it, and two thirds of them said they could not afford to go unpaid — on today's workforce about 7.4 million people. They worked through a serious illness, their own or someone else's.

Value 9.5 · HealthImpact 0.6Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is what happens to people who work through something they should not work through: an operation postponed, a treatment attended between shifts, a dying parent visited at weekends. It is placed in the class this site uses for life and health, one step below the top because almost all of it is strain and illness rather than death. It is distinct from the money above, and the test is the usual one — a worker who takes the unpaid leave anyway gets nothing from this argument, only from that one. Nothing is counted for the family member being cared for: whether a relative's presence changes the course of their illness has not been measured, and that gain, if it exists, is left out rather than denied. Nothing is counted for the employer, who in most of these cases would rather the person took the leave. The value sits just below the maximum: the stream is health, mostly health that is recovered.

The Labor Department's 2018 survey, the most recent it has run, finds that about 7 percent of employees needed leave in that year and did not take it, two thirds of them because they could not afford it ^[2]^[3]; on today's workforce that is about 11.3 million and 7.4 million people, a projection rather than a reported count. Of those, about 5.4 million are in states without their own programme. Not all of them would take leave under a federal one: some fear the consequences at work anyway, and two thirds of wages is still a cut. Sixty-five percent is used, in a range from 45 to 80 percent — higher than it would be for a benefit alone, because the bill also gives those who draw it the right to return to their job — all except workers in their first 90 days with an employer — and fear of losing the job is the second most common reason for staying at work ^[1]^[3]. That gives about 3.5 million leaves a year that do not happen today. What each of those is worth is the weakest number in this evaluation: 0.02 quality-adjusted years each, in a range from 0.005 to 0.06 — about a week of good health, which is low for someone attending a parent's final illness and high for a routine recovery. That gives about 70,000 quality-adjusted years a year. The Impact is a twelfth of the money in this debate, which is what happens when a large number of people each gain something small and hard to price.

▸ Show calculation ▾ Hide calculation

	Workers who needed leave and could not afford to take it [3]	of 11.3 million who needed leave and did not take it	7.4 million workers
×	Share in states without a programme of their own [4]	73 %	5.4 million workers
×	Share who would take the leave once it is paid Setting, range 45 to 80 percent: the bill's right to return to the job removes the formal fear of taking leave for workers past their first 90 days with an employer, and two thirds of a wage is still a cut	65 %	3.51 million leaves a year
×	Quality-adjusted years gained each Setting, range 0.005 to 0.06: about a week of good health, low for a parent's final illness and high for a routine recovery	0.02	70,200 quality-adjusted years
×	Value of the years the value of a healthy life year used across this site	40,000 euro-equivalent each	2.81 billion euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	0.56

Score 0.56 Impact × 9.5 Value × 5 Plausibility ÷ 10 = 2.7 of 100

The number of people who say they could not afford leave is measured; what that costs them is not. The counterfactual is the same workers under unpaid leave, which is what the survey asks about directly. The chain is named — no wage replacement, leave not taken, condition managed worse — and every link of it is plausible, but the health consequence at the end has never been measured for this population; the studies of paid leave measure birth outcomes and maternal depression, not the untreated conditions of people who stayed at work. The confounder that matters is that a stated inability to afford leave may stand in for a job the worker fears losing; the bill's right to return answers the formal part of that fear, except for workers in their first 90 days with an employer, but not the informal part, and the survey cannot separate the two, which is why take-up is set well short of everyone. Reverse causation does not arise. Since the direction is not in doubt but nothing measures the size, the number rests on a chain that is complete rather than on evidence. The Plausibility is at the middle: the unmet need is counted precisely and what it costs people is assumed.

Evidence: Projection · plausibility at most 6 Evidence of effect: Mechanistic · at most 6

Babies born heavier

0.86points

When the five states that ran disability insurance were required in 1978 to cover pregnancy, and so to pay for maternity leave, they saw fewer low-birthweight and early births than the other states, with the largest effects among unmarried and Black mothers. A small share of infant deaths follows the same pattern.

Value 9.5 · Life and healthImpact 0.1Plausibility 6

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the health of newborns: fewer born under two and a half kilograms, fewer born early, and a small number who live rather than die in their first year. It is placed just below the top of the scale because most of what is counted is morbidity that is largely recovered from, with a minority of deaths inside it. The mother's own health is a separate stream and is not counted here, and neither is the medical spending avoided, which would be the same event priced twice. The children concerned are not the ones whose mothers already had paid leave, since for them nothing changes. What is priced is the difference in outcomes for babies whose mothers can now stop working before and after the birth. The value sits just below the maximum: the stream mixes lives with illness that is mostly outgrown, weighted towards the illness.

About 3.6 million babies are born in the United States each year, of whom roughly 2.6 million to mothers in states with no programme of their own. Low birthweight affects 8.6 percent of American births, so about 226,000 of those. Stearns found that paid maternity leave through the disability insurance system reduced the share of low-birthweight births by 3.2 percent across all births in the covered states, and by more than ten percent among the mothers who actually drew the benefit; the population figure is applied here, which is about 7,200 births a year ^[6]. Infant mortality runs at 5.4 per thousand, so about 14,200 deaths among the same births; a three percent reduction, in a range from one to six percent, is about 430 deaths. Valuing the deaths at the figure this site uses for a life and putting half a quality-adjusted year on each avoided low-birthweight birth gives a Normalised Impact of 0.15 on the scale of this evaluation. Nothing is counted for the longer-run schooling and earnings differences that follow birthweight, which are real and would enlarge this. The Impact is small beside the money in this debate and it carries the highest value per unit, which is why it survives at a fiftieth of the size.

▸ Show calculation ▾ Hide calculation

	Births to mothers in states without a programme of their own [8]	3.6 million births, 73 percent	2.63 million births a year
×	Low-birthweight births among them [8]	8.6 %	226,000 births a year
×	Reduction in low birthweight measured across all births in the five states that paid for maternity leave through disability insurance after 1978; over ten percent among the mothers who drew it [6]	3.2 %	7,232 fewer low-birthweight births
×	Value of the avoided low-birthweight births half a quality-adjusted year on each avoided low-birthweight birth, at the value of a healthy life year used across this site	0.5 quality-adjusted years each at 40,000 euro-equivalent	145 million euro-equivalent a year
+	Infant deaths avoided Setting, range 1 to 6 percent: birth outcomes improve on the same designs, infant mortality less clearly [8]	14,200 infant deaths, 3 percent fewer (about 430), at 1.4 million euro-equivalent each	602 million euro-equivalent a year
=	Both together	145 + 602, about 745	747 million euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	0.15

Score 0.15 Impact × 9.5 Value × 6 Plausibility ÷ 10 = 0.9 of 100

The counterfactual is the five states that ran temporary disability insurance — California, New Jersey, New York, Rhode Island and Hawaii — before 1978, when a federal amendment required them to cover pregnancy and so to pay for maternity leave, compared with the states that never had such insurance over the same years. The design is quasi-experimental: a difference-in-differences around one legislated date, with the non-insurance states absorbing whatever changed nationally in those years. The confounder that matters is that the insurance states could have moved differently from the rest for reasons of their own — prenatal care and health coverage expanded unevenly in that period; the study addresses it partly by finding the largest effects among unmarried and Black mothers, which is where a leave effect should be concentrated and where a general coverage expansion would not be. Reverse causation is not plausible on this timescale. What is not resolved is the transfer: the programmes studied were narrower, older and less generous than the one proposed, and the American birth cohort has changed. The Plausibility is at the middle to upper part: a clean design on the right question, in an era and under programmes that are not quite this one.

Evidence: Study · plausibility at most 7 Evidence of effect: Quasi-experimental · at most 8

Mothers' health after birth

0.11points

Where states pay for leave, fewer mothers report depressive symptoms in the months after birth, and mothers and parents report better mental health. The effect is largest for single, younger and lower-income mothers.

Value 9.5 · HealthImpact 0.02Plausibility 6

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is health: fewer weeks of depression for mothers after birth. It is priced at the level this site uses for health, just below life itself. Postpartum depression is an illness, not a mood, and it weighs on the mother, on the care of the child and on the family around them. What is counted is the illness avoided, not the treatment costs, which are money and are left out. The value of the child's own health is counted separately, in the argument on birth outcomes. The value sits near the top of the scale, at the level this site uses for health.

About 2.63 million babies a year are born to mothers in states without a programme of their own. A study of six state programmes using the national survey of new mothers finds postpartum depressive symptoms 0.93 percentage points less common after the programmes started, against a baseline of about 11 percent ^[13]; applied to those births that is about 24,500 mothers a year, in a range from 12,000 to 60,000, the upper end allowing for the larger reductions in psychological distress found for California ^[14]. Each episode avoided is put at a tenth of a quality-adjusted year, in a range from 0.05 to 0.2 — a few months of moderate depression. That is about 2,450 quality-adjusted years a year. Nothing is counted for better self-rated child health or fewer infant hospital stays, which other studies find ^[15], because their size in health terms is not established and part of it overlaps with the birth outcomes counted separately. The Impact is the smallest in this debate, but it rests on measured effects.

▸ Show calculation ▾ Hide calculation

	Births to mothers in states without a programme of their own [8]	3.6 million births, 73 percent	2.63 million births a year
×	Fewer mothers with postpartum depressive symptoms Setting, range 12,000 to 60,000 mothers: measured across six state programmes against a baseline of about 11 percent; California's synthetic-control study finds larger reductions in distress [13][14]	0.93 percentage points	24,500 mothers a year
×	Quality-adjusted years gained each Setting, range 0.05 to 0.2: a few months of moderate depression avoided	0.1	2,450 quality-adjusted years
×	Value of the years the value of a healthy life year used across this site	40,000 euro-equivalent each	98 million euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	0.02

Score 0.02 Impact × 9.5 Value × 6 Plausibility ÷ 10 = 0.1 of 100

The counterfactual is mothers in the same states before the programmes started and in states without one. The design is quasi-experimental: a staggered difference-in-differences across six programmes with an estimator robust to differing start dates ^[13], supported by a synthetic-control study of California that finds psychological distress among mothers of infants down by about a quarter ^[14]. The confounder that matters is that states adopting paid leave also expanded other support for new mothers in the same years; the study compares states with and without programmes over the same period, which absorbs national trends but not state-specific ones. The same study finds small unexpected increases in low birthweight, which its authors attribute to selection or data problems; that is a reason for caution rather than a finding against. Reverse causation is not plausible at a legislated date. What is not resolved is the transfer to a federal programme. The Plausibility is a little above the middle: two independent designs agree on direction, and the size carries over from state programmes.

Evidence: Converging studies · plausibility at most 7 Evidence of effect: Quasi-experimental · at most 8

Arguments against4 arguments

A contribution on every pay packet

18points

The benefits are paid out of a levy on earnings, which every worker pays in every week whether or not they ever draw on it. Because it is a share of pay, most of it comes from the households that earn most. Part of it replaces leave employers used to fund themselves, and only the remainder is a new burden.

Value 5 · Household budgetsImpact 3.8Plausibility 9.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money leaving pay packets, priced at the middle of the scale exactly as the benefit is. This argument exists so that the benefit above is not counted as if it fell from the sky: it is the same money, taken in the weeks people are earning. The contribution is nominally split between worker and employer, but a levy on wages is borne by wages within a few years whichever side writes the cheque, so it is booked at the ordinary weight for household money across the whole earnings distribution. What is netted out here rather than booked separately is the leave employers used to pay for and now do not — that saving belongs to the same mechanism and appears once, as a reduction of this cost, not as a gain of its own. Nothing is counted for the administrative burden on employers of a new deduction, which is small and recurring. The value is the middle of the scale, and the contribution is booked net of what it replaces so that the same dollar is not subtracted twice.

The 30 billion dollars a year of new household money has to be raised, and the bill raises it as a contribution on earnings shared by workers and employers; within a few years a levy on wages is borne by wages whichever side pays it in. The 40 percent of benefits that replaces leave employers were already paying for is not a new burden: the firm stops paying and the contribution pays instead, and the money stays with the same pay packets. Running the programme is counted as its own argument. Who pays decides the weight. A contribution on wages up to the Social Security limit falls on earnings, and earnings are concentrated: in 2024 the top fifth of households received about 54 percent of all wages and the lowest two fifths about 7 percent ^[12], and after the limit the top fifth still carries about half. At this site's weights for each fifth that gives 0.73, in a range from 0.65 to 0.85. The result is a Normalised Impact of 3.78 on the scale of this evaluation, in a range from 3.11 to 5.56. The Impact is the second largest here: the same money as the benefit, weighing less on the side that pays it, because most of it comes from the top of the earnings scale.

▸ Show calculation ▾ Hide calculation

	Workers not already covered by a state programme [4][10]	163 million employed, 27 percent already covered	119 million workers
×	Benefits paid per covered worker a year Setting, range 350 to 575 dollars: Washington's programme paid about 2.01 billion dollars in 2025, about 490 dollars per covered worker; it replaces 90 percent of a low wage and runs to 16 weeks, more generous than the bill, so a sixth is taken off [7]	420 dollars	50 billion dollars a year
÷	In euro-equivalent the rate used throughout this evaluation	1.16 dollars to the euro	43.1 billion euro-equivalent a year
×	New household money the contribution has to raise Setting, range 45 to 75 percent: the other 40 percent replaces leave employers already paid for, so the same pay packets carry it before and after	60 %	25.9 billion euro-equivalent a year
×	Weight of a dollar for the households paying it Setting, range 0.65 to 0.85: a contribution on wages up to the Social Security limit falls on earnings, and the top fifth of households earns about half of all wages, the lowest two fifths about 8 percent; at this site's weights for the five fifths that gives 0.73 [12]	0.73	18.89 billion euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	3.78

Score 3.78 Impact × 5 Value × 9.5 Plausibility ÷ 10 = 18 of 100

That the contribution is levied follows from the bill: it is a legal consequence, not a behavioural claim. The counterfactual is the existing patchwork of employer policies and state programmes. Its size is the size of the benefits it funds and carries the same doubt, which sits in the range; the share of employer-paid leave that disappears when the public benefit arrives has been observed in California and Washington but not measured cleanly, and it is part of that range too. Reverse causation does not arise. What remains is the same implementation risk as on the receiving side, so the value is the same, 9.5. The Plausibility is high: the levy follows from the law, and only its size is estimated.

Evidence: Mechanism · plausibility at most 9.5 Evidence of effect: Definitional

First-time mothers earn less

4points

The best-identified study of California's programme follows first-time mothers for twelve years in tax records and finds them about 4 percentage points (6 percent) less likely to be employed and about 20,400 dollars poorer over years nine to twelve.

Value 5 · Household budgetsImpact 1.6Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is earnings that do not arrive, priced at the middle of the scale like every other money stream in this evaluation. It is not given a higher class for being about careers or about women: what is measured in the tax records is income, and income is income. Part of the reduction is a choice — some of these mothers work less because they want to, and time at home has a value that is not zero — which is handled by taking a figure well below the measured one rather than by lowering the value class. Nothing is counted for the effect on the gender pay gap as such, which is this same money under an aggregate name. Nothing is counted for the children, who appear in their own argument on the other side. The value is the middle of the scale, and the part of the reduction that is chosen rather than imposed is taken out of the quantity, not out of the value.

Bailey, Byker, Patel and Ramnath follow first-time mothers in tax records around California's 2004 threshold and find them about 3.8 percentage points less likely to be employed, 6.2 percent below the comparison group, and about 20,400 dollars poorer in cumulative earnings in the years around a decade after the birth, against a control mean of 153,000 dollars ^[5]. That is about 5,100 dollars a year in the years the loss is measured. The figure used here is about 4,100 dollars, in a range from 1,700 to 8,100: lower than the measurement because some of the reduction is chosen time at home rather than a loss, and not lower still because the programme proposed is twice as long and pays more than California's did. About 461,000 first-time mothers a year would newly take the leave — first births, employed before the birth, outside the existing state programmes, at seventy percent take-up. The loss appears late and is taken to last about five years, so each year's cohort of mothers carries about 20,300 dollars in total; that is booked against the year the cohort gives birth, the same way the birth outcomes on the other side are booked against the year of birth, which comes to about 9.4 billion dollars a year — the same as five cohorts inside the affected years at once. Inside the first ten years only the earliest cohorts have reached those years, and the amount is not cut to that window because nothing on either side is discounted or cut at the horizon. This is the argument that would grow most if the dose mattered proportionally. The Impact is a quarter of the benefit it accompanies, and it is the only cost here that a supporter of the measure would not have predicted.

▸ Show calculation ▾ Hide calculation

	First-time mothers newly taking the leave each year [8]	3.6 million births, 38 percent first births, 66 percent employed beforehand, 73 percent outside existing programmes, 70 percent take-up	461,000 women a year
×	Years each cohort carries the loss, booked against its birth year Setting, range three to eight years: the loss is measured around a decade after the birth and is taken to last about five years; each cohort's loss is booked against its birth year, like the birth outcomes, which equals five cohorts inside the affected years at once [5]	5	2.31 million women
×	Earnings not received each year Setting, range 1,500 to 7,000 euro-equivalent (about 1,700 to 8,100 dollars): measured at about 5,100 dollars a year in California, reduced because part of the reduction is chosen time at home and raised by nothing for the longer and more generous leave proposed [5]	3,500 euro-equivalent (about 4,100 dollars at 1.16 dollars to the euro)	8.07 billion euro-equivalent a year
×	Weight of a dollar at these earnings the women concerned were employed before the birth and sit around the middle of the distribution	1.0	8.07 billion euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	1.61

Score 1.61 Impact × 5 Value × 5 Plausibility ÷ 10 = 4 of 100

The counterfactual is first-time mothers who gave birth just before the programme took effect. The study compares births shortly before and shortly after the start date in July 2004, leaving out a window around the date itself, and uses each mother's recorded take-up of the benefit from tax records to estimate the effect on those who actually drew it ^[5]. That design is quasi-experimental, a discontinuity in time, and it is the strongest in the paid leave literature by some distance. The confounder that matters is that the discontinuity falls at a calendar date, so anything else changing in California at the same moment would be attributed to the programme; the study does not run a separate check against an unaffected group, and it reports mothers of later children only as a different, differently selected group, not as a control. Reverse causation does not arise at a legislated date. What is also unresolved is the disagreement with the rest of the literature: several earlier studies using survey data find the opposite sign in the first year after birth, and while their designs are weaker, a single contradicting result on a contested question does not carry a high plausibility. The Plausibility is at the middle: the design is the best available, the calendar date is not checked against an unaffected group, and the result stands against a body of weaker work pointing the other way.

Evidence: Study · plausibility at most 6 Evidence of effect: Quasi-experimental · at most 8

Covering the absences

3.4points

Paying for leave means more of it is taken, and somebody has to do the work. For a small employer that is overtime, an agency temp, or output that does not happen.

Value 5 · Company budgetsImpact 1.5Plausibility 4.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is what it costs a firm to keep running while somebody is away: overtime at a premium, an agency worker at a margin, a manager doing two jobs, or work that simply does not get done. It is priced at the middle of the scale like all money and it is a real cost rather than a transfer, because the hours and the premium are consumed. The wage of the absent worker is not part of it — that is paid by the insurance and is counted twice over in the first two arguments. Nothing is counted for the disruption to colleagues, which is the same hours under another name. Nothing is counted for the firms that would have had to cover the leave anyway under the existing unpaid statute, which is why only the additional leave is counted here. The value is the middle of the scale, and only the leave that would not otherwise have been taken is counted.

Two things add leave-weeks. About 3.5 million people take a leave they would not have taken, at roughly four weeks each, which is 14 million weeks. And the 10.9 million who take unpaid or partly paid leave today extend it once they are paid for it, by about a week each on average, in a range from half a week to three weeks; California's experience is that leave-taking roughly doubled for bonding leave, which would be more than this. That gives about 24.9 million additional leave-weeks a year, roughly 480,000 full-time equivalents or a quarter of one percent of all hours worked. What one such week costs a firm is put at about 350 dollars, in a range from 120 to 640: at the low end the work waits, at the high end it is covered by overtime or an agency. The firm does not pay the absent worker, because the insurance does, so the extra cost is at most the overtime premium of about half a weekly wage or an agency's margin, not a whole wage. That is about 8.7 billion dollars a year. The Impact is a fifth of the benefit paid, which is the ordinary proportion when an insurance makes people use something they were entitled to but could not afford.

▸ Show calculation ▾ Hide calculation

	Leaves that do not happen today [3]	3.5 million a year at about four weeks each	14 million leave-weeks
+	Existing unpaid leaves extended once they are paid Setting, range half a week to three weeks: California's bonding leave roughly doubled, which would be more than this [9]	10.9 million leaves, about one week longer each	10.9 million leave-weeks
=	Additional leave-weeks a year	14 + 10.9	24.9 million leave-weeks
×	Cost of covering one leave-week Setting, range 100 to 550 euro-equivalent (about 120 to 640 dollars): at the low end the work waits, at the high end overtime or an agency covers it; the insurance pays the absent worker, so the firm bears at most the overtime premium of about half a weekly wage or an agency's margin	300 euro-equivalent (about 350 dollars at 1.16 dollars to the euro)	7.47 billion euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	1.49

Score 1.49 Impact × 5 Value × 4.5 Plausibility ÷ 10 = 3.4 of 100

The counterfactual is the same firms under unpaid leave, and the direction is not in dispute: paid leave is taken more than unpaid leave, which is the point of it. What has no measurement is the cost per week. The chain is complete — more leave taken, hours to be covered, overtime or agency or output foregone — and the counter-mechanism is named and real: surveys of employers in states with programmes report that most found the effect small or positive partly because a worker who returns is cheaper than one who quits. That counter-mechanism is not resolved, and it would cut this argument substantially if it holds outside the states that chose to legislate. Reverse causation does not arise. No study puts a price on the coverage of an additional leave-week, so what places this number is a complete chain rather than a finding. The Plausibility is below the middle: the additional leave is well evidenced and what covering it costs has never been measured.

Evidence: Mechanism · plausibility at most 6 Evidence of effect: Mechanistic · at most 6 Placement: Chain closed, unevidenced · plausibility 4–5

Open: State programmes hold the leave-weeks they pay for, and payroll data hold the overtime and agency hours of the same firms. Matching the two before and after a programme starts would price a covered week directly and could carry this above 6.

Running a new federal insurance

1.7points

A national leave insurance needs staff, systems and claims handling. Washington spends about four cents on administration for every dollar of benefit, and a new federal programme will not start out cheaper.

Value 5 · AdministrationImpact 0.3Plausibility 9.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is real resources: staff, computer systems and offices that a new insurance ties up and that could otherwise do something else. It is priced at the middle of the scale, the level this site uses for any cost of running public business. The pay of the staff is not a second cost on top of it; the outlay for them is the cost. Nothing is added for the burden on employers of reporting wages, which in states with programmes runs through the existing payroll systems. The value is the middle of the scale, the level this site uses for administrative cost.

Washington's programme spent about 77 million dollars on administration in 2025 against about 2.01 billion of benefits, close to 4 percent ^[7]. That includes the grants it pays to small employers. Four percent is used here, in a range from 3 to 8 percent, because a new federal programme has to build its systems and coordinate with fourteen state programmes before it runs at Washington's cost. The upper end allows for a start-up period in which claims handling is slower and dearer. On benefits of 50 billion dollars a year that is about 2.0 billion dollars. This cost was previously folded into the contribution; it stands here on its own because it is spent rather than passed from one household to another. The Impact is a twentieth of the benefit: what it costs to pay it out.

▸ Show calculation ▾ Hide calculation

	Benefits paid a year [7]	119 million workers × 420 dollars ÷ 1.16	43.1 billion euro-equivalent a year
×	Administration as a share of benefits Setting, range 3 to 8 percent: Washington spent about 77 million dollars against 2.01 billion of benefits in 2025 [7]	4 %	1.73 billion euro-equivalent a year
÷	Normalised Impact scale of this evaluation	5 billion euro-equivalent a point	0.35

Score 0.35 Impact × 5 Value × 9.5 Plausibility ÷ 10 = 1.7 of 100

A programme of this kind cannot run without administration, so the cost arises whenever the measure exists; how large it is depends on how well it is run, and that sits in the range. The counterfactual is no federal programme and no federal administration for one. The amount rests on Washington's reported spending, a single programme that is not a random sample of the country. Reverse causation does not arise. The Plausibility is high: the cost is certain, and only its size is estimated.

Evidence: Mechanism · plausibility at most 9.5 Evidence of effect: Definitional

How robust the result is

Under the assumptions used, the proposal brings more than it costs in 96 of 100 model runs.

gains more than it costscosts more than it gains

Each dot stands for 200 of the 20,000 model runs.

The two largest arguments are the same money seen twice: what the contribution takes out of pay packets and what the benefit puts into households in the weeks they have no earnings. What counts is where it lands: the contribution falls on wages and so mostly on the top of the earnings scale, the benefit mostly on lower and middle earners who are unpaid on leave today. Weighted by this site's rule for each income fifth, a dollar of benefit is counted at 1.2 and a dollar of contribution at 0.73. What covering one leave-week costs a firm is a setting without a measurement. So is the share of the benefit that is new money for the household rather than leave an employer already paid for, set at six in ten. Weighted so, the measure lies ahead in 96 of 100 runs.

Show the distribution of the model runs

20,000 simulated futures

Very likely better · 96 %

In 96 of 100 simulated futures the proposal brings more than it costs.

No effect is known exactly; every argument has a range. The calculation draws a possible size for each argument from its range 20,000 times and counts how often the benefits exceed the costs.

Net effect in points per future. Left of zero the drawbacks outweigh the benefits, right of zero the benefits win. Eight in ten futures fall between +1.9 and +11 points.

Scale and assessment

Scale of this evaluation: Normalised Impact, unitless and calibrated to this topic. For comparison, one point here corresponds to about 5 billion euro-equivalent a year.

Size class:large

Effect large (+31) · 63 % of the effect is positive

Areas show individual argument weights. The positive share of the effect nets out money transfers between the two sides; the probability of a positive net effect is determined separately.

How we score →

Review and sources

Last reviewed by Claude Opus 5.5 · October 28, 2026 · 9× AI, 2× human

October 28, 2026AI reviewClaude Opus 5.5record updated
Page texts for the new page layout (draft G) taken over from the Lab; numbers unchanged.
October 21, 2026AI reviewClaude Fable 5.1record updated
Money amounts in argument texts stated in dollars per style-en.md; scores, chains and balance unchanged
October 20, 2026AI reviewClaude Fable 5.1record updated
Summary and scale note rewritten to the style of 20 Oct 2026; numbers unchanged
October 16, 2026AI reviewClaude Opus 5.5record updated
Texts updated to Bilanz 2.2
October 15, 2026Assessed by a human, implemented by AIClaude Opus 5.5record updated
Upper bound of the cost of covering a leave-week corrected from 900 to 550 euro, because the insurance pays the absent worker; simulated balance 0.607 to 0.720, better.
October 15, 2026Assessed by a human, implemented by AIClaude Opus 5.5re-scored
Review of 15 October, second pass with Julian's decisions: Weights derived from who receives (1.2) and who pays (0.73), benefit and contribution booked as one pair, administration and maternal health as own arguments, cost per worker raised to 420 dollars; simulated balance 0.607, level.
October 15, 2026AI reviewClaude Opus 5.5re-scored
Review of 15 October: the California study now reported correctly (about 4 percentage points, no placebo test; plausibility 5.5 to 5), the 90-day exception to the right to return added, derivations itemised, ranges on all six arguments; simulated balance 0.515, level as before.
September 14, 2026AI reviewClaude Opus 5re-scored
Gesamtprüfung 08.09.: Scope an S. 2823 (85/69/50 %, Rückkehrrecht), pro-2 Take-up 55→65 % (i 2,38→2,81), con-3 i 6,83→7,47, Stearns-Link/Design 1978 und DOL-Datum 2018 korrigiert, con-2 Kohortenbuchung neu begründet; r 0,51→0,51.
September 6, 2026AI reviewClaude Opus 5First evaluation
First evaluation: programme cost taken from operating state programmes, career effect from the regression-discontinuity study of California's leave.

Evaluations are produced with AI support and reviewed on a schedule for new developments; human passes are marked separately.How we review →

Show all 16 sourcesHide sources

Congress.gov: S. 2823, Family and Medical Insurance Leave Act. congress.gov
U.S. Department of Labor: Employee and Worksite Perspectives of the Family and Medical Leave Act. dol.gov
National Partnership for Women and Families: New Department of Labor family and medical leave data illustrates gaps in coverage. nationalpartnership.org
Bipartisan Policy Center: State paid family leave laws across the United States. bipartisanpolicy.org
Bailey, Byker, Patel and Ramnath, American Economic Journal: Economic Policy: The Long-Run Effects of California's Paid Family Leave Act on Women's Careers and Childbearing. aeaweb.org
Stearns, Journal of Health Economics: The effects of paid maternity leave: evidence from Temporary Disability Insurance (Journal of Health Economics 43, 2015). sciencedirect.com
Washington State Employment Security Department: 2025 Annual Paid Family and Medical Leave Report. esd.wa.gov
Centers for Disease Control and Prevention: Births: final data, and infant mortality statistics. cdc.gov
Rossin-Slater, Ruhm and Waldfogel, Journal of Policy Analysis and Management: The Effects of California's Paid Family Leave Program on Mothers' Leave-Taking and Subsequent Labor Market Outcomes. onlinelibrary.wiley.com
U.S. Bureau of Labor Statistics: The Employment Situation — September 2026. bls.gov
Brown, Roy and Klerman (Abt Associates for the Department of Labor): Family and Medical Leave Experiences Vary by Gender and Wages. abtglobal.com
U.S. Bureau of Labor Statistics, Consumer Expenditure Survey 2024 (via FRED): Wages and salaries by quintile of income before taxes. fred.stlouisfed.org
Wells, White, Collin, Wang, Modrek and Hamad, American Journal of Epidemiology 195(1): State paid family leave policies and maternal and infant health. doi.org
Doran, Bartel, Ruhm and Waldfogel, Social Science & Medicine 256: California's paid family leave law improves maternal psychological health. doi.org
Bullinger, Journal of Health Economics 66: The effect of paid family leave on infant and parental health in the United States. doi.org
Congress.gov S. 5017: More Paid Leave for More Americans Act. congress.gov