Where We Left Off
Before we begin
Five parts are now behind us. Part One separated the three claims welded into every defence of these rules. Part Two found their origin in property and inheritance. Part Three found the Indian machine, which runs on marriage rather than paternity. Part Four measured every difference between men and women and found most of them close to a coin toss. Part Five examined the finding that differences appear to grow when constraints lift, and concluded that the finding is real in some measures, shaky in others, and unexplained in all of them.
Each of those parts ended by asking the same question: does any of this justify the rules? Each time the answer was no.
A reader could be forgiven for concluding that the traditional position has been dealt with. It has not. It has been examined in pieces, on my terms, in a sequence I chose.
So this part does something different. It argues for the other side.
Not a version built to lose. Not a summary of what traditionalists say, with the weak bits included. The strongest case that can be constructed from the best available evidence, assembled by somebody trying to win — and in several places it will be a better case than the one its actual advocates make, because they tend to reach for arguments that fail while leaving the ones that work unused.
Then, in Chapter Nine, I score it. Honestly, including the parts that survive.
I want to say plainly what this part will contain, because some of it will be uncomfortable for readers who have found the previous five congenial. It contains real evidence that children raised in stable two-parent households do better on average. It contains real evidence that marriage changes men’s behaviour rather than merely selecting for it. It contains the fact that partnership is collapsing across the developed world and that the collapse is measurable. And it contains the strongest argument of all, which is made most often by women and dismissed most often by men: that the old arrangement was a guarantee, that the replacement was promised and only partly built, and that the entire risk of the transition falls on the person being asked to make it.
Every one of those is true. What Chapter Nine asks is what they actually justify.
How to Read the Boxes
The notation, in case this is where you started
Six kinds of box run through the series. One live example of each.
A steelman is the strongest possible version of a position you disagree with — the version its best advocate would give, with the weak arguments removed and the good ones stated at full force.
The opposite is a strawman: a version built to be easy to knock down.
Why it matters here: this whole part is a steelman, and Chapter Nine argues that steelmanning is not the neutral act it appears to be. Building somebody’s best case gives them something they did not have.
A Word Box appears the first time a hard word does — never later, never only in a glossary.
Before any argument, one number, because it does more work than any argument in this part.
In a society without antibiotics, without safe surgery and without blood transfusion, a woman’s chance of dying in any given childbirth was somewhere around one in a hundred. Over a reproductive life of six or eight births, her cumulative chance of dying in the process ran into the high single figures or above.
Put that in a village. Take a hundred women who reach childbearing age. Somewhere between five and ten of them will die producing the next generation — not from war, not from famine, from an ordinary event that happens in a room.
Every rule in this series was written by and for people living with that number. It does not make the rules right. It makes them a response to something, and Chapter Two is about what.
Any number too big to picture gets a body attached.
Much of this part rests on studies comparing children, adults or households in different family arrangements. That design has one dominant weakness and it has to be named before any number is quoted.
The problem: nobody is randomly assigned to a marriage. People who marry and stay married differ from those who do not — in income, education, health, temperament and a hundred other things — before the marriage exists. Any difference in their children could be caused by those prior differences rather than by the marriage.
What better designs do: compare siblings who experienced different family arrangements, follow the same individuals before and after a change, or use a change in law that affected some people and not others. These strip out much of the selection.
What they find, consistently: the effects are real and smaller than the raw comparisons suggest. Both halves of that sentence matter, and each camp reliably quotes one of them.
That box names the evidence and its limits. The next one appears wherever competent people genuinely disagree.
Worth settling before eight chapters of doing it.
An argument you have only met in its weakest form is an argument you have not met. Defeating a bad version of a position tells you nothing about the position, and it leaves you unprepared for the good version when somebody finally makes it. It is also the only honest way to find out whether you are right.
Real positions are held by real people for real reasons, and a philosopher’s improved version is not what anybody is actually defending. Building it lends respectability to a set of practices whose actual justification is worse, and then the improved version gets quoted by people who hold the unimproved one.
An Argument box appears where people who have studied something genuinely disagree.
Both sides above assume that there is a position here to be steelmanned — one traditionalist case, which can be made better or worse.
There is not. What follows is at least four separate arguments, and they do not support each other. An argument about children. An argument about partnership. An argument about male behaviour. An argument about women’s economic security.
They point at different problems, imply different remedies, and in places contradict each other outright. The argument about male behaviour says the difficulty is men and the intervention belongs there. The argument about women’s security says women need protection. Those are not the same claim and a policy following one will not follow the other.
The general form: treating a coalition of arguments as a single position. Coalitions look strong from outside because every member’s weakness is covered by somebody else’s strength — and they cannot be defeated or adopted as a unit, because they were never one thing.
This is the signature box. There are five in this part.
Every chapter closes with one of these, restating it in the plainest words available, key terms in bold.
Read only these and you should still finish holding the whole argument.
A Note on This Part
Read this before Chapter One
I am arguing for something I do not hold, and I am going to do it properly. Chapters Two through Eight contain no hedging, no undercutting asides, and no signals to the sympathetic reader that I do not mean it. That would be cowardice dressed as balance. The scoring happens in Chapter Nine, in one place, where you can see it.
If you find yourself persuaded somewhere in the middle of this part, that is the point. Part One said to notice which findings you hope are true, because that marks where you are least reliable. This part is designed to locate that spot for a reader who has enjoyed the previous five.
And a warning about what this part is not. It is a case for stable families, for partnership, and for taking the costs of transition seriously. It is not a case for caste endogamy, enforced widowhood, dowry, seclusion, or a rumour published about a woman who stood in a street. Chapter Nine shows that the strongest traditionalist arguments do not reach those things — and that this is a much more serious problem for the position than any objection I could raise from outside it.
1What A Steelman Requires
Building somebody else’s best argument has rules. The first is that you have to actually try. The second is that you have to understand what kind of thing you are defending — and the rules in this series are not a morality. They are a technology.
1.1 — The discipline
There is a test for whether you have understood a position you disagree with. Can you state it in a form its holders would recognise, and would want to sign?
Most people cannot, about most things they oppose. What they can produce is a version assembled from the worst thing anybody on that side has said, which is easy to defeat and tells you nothing.
So this part follows three rules, and they are worth stating because you should hold me to them.
No undercutting. Chapters Two to Eight contain no asides winking at the sympathetic reader, no “of course this is nonsense but”, no adjectives doing quiet damage. If an argument is strong, it is stated as strong.
Use the best available evidence, not the evidence the position’s advocates usually cite. In several places the strongest support for a traditionalist claim comes from research that traditionalists do not know about and would find surprising. It goes in anyway.
Drop the arguments that fail. A steelman is not a survey. Several things routinely said in defence of these rules are simply wrong, and including them would be a way of losing on purpose.
1.2 — Technology, not morality
Here is the frame that makes the whole case work, and it is the thing its advocates most often get wrong about their own position.
The rules in this series are usually defended as values — as an account of what is decent, pure, honourable, appropriate. Defended that way they are very hard to argue for, because a value has no evidence behind it and the person on the other side simply does not share it. The conversation ends immediately.
Read them instead as technology and everything changes.
A technology is a solution to a problem. It can be assessed. It can be compared with alternatives. It can be judged on whether it worked, at what cost, and to whom. And crucially, a technology can be obsolete without ever having been wrong — the way a well-designed oil lamp is not a mistake, merely superseded.
Part Two already established that these rules solved real problems. The traditionalist who says “these things exist for a reason” is making a technological claim and then, almost always, defending it in the language of morality — which is the weakest ground available and the ground they choose.
This part will not make that mistake. Every argument in it is of the form: here is a problem, here is what the old arrangement did about it, here is the evidence, and here is what happens where the arrangement is gone.
Everything above — and every version of this argument on both sides — assumes that we know what the rules were for.
We do not, in the sense that matters. What we have is a set of functions, worked out retrospectively by researchers looking at outcomes: inheritance security, boundary maintenance, disease avoidance, insurance. Those are real and Part Two established several of them.
But almost nobody who ever lived under these rules held them for those reasons. They held them because it was what was done, because their mother had, because the alternative was unthinkable, or because they were told. A functional account is a description of what a practice achieved, not of what anybody was trying to achieve.
Which means this entire part is a modern reconstruction. I am building a case that no traditional society ever made, out of evidence none of them had, and attributing it to people who would not recognise a word of it. The traditionalist reading this chapter with satisfaction should notice that the argument being made on their behalf is not their argument. It is mine.
The general form: reconstructing a rationale and attributing it to people who never had one. It is what makes a functional explanation so persuasive and so slippery, and Chapter Nine asks what building it may have done.
1.3 — What a fair test looks like
Finally, the standard. Part One, Chapter Nine set out what the traditionalist would have to show to win, and it is worth restating because this part is an attempt to meet it.
That the rule solves a real problem — not that it once did, but that the problem exists now.
That the harm from removing it exceeds the harm from keeping it — which requires counting both columns, and Part Twelve counts the second.
That no cheaper instrument exists.
That the burden belongs where it is placed.
Four conditions. Chapters Two to Eight assemble the strongest available case. Chapter Nine checks it against all four, and tells you honestly how many it clears.
A steelman has rules. No undercutting — no asides winking at the sympathetic reader. Use the best evidence, including evidence the position’s advocates do not know about. Drop the arguments that fail, because including them is a way of losing on purpose.
The frame that makes the case work: read these rules as technology, not morality. A value has no evidence behind it and the conversation ends. A technology solves a problem, can be compared with alternatives, and can be obsolete without ever having been wrong — the way an oil lamp is not a mistake, merely superseded.
Traditionalists make a technological claim and then defend it in moral language, which is the weakest ground available and the ground they choose.
But we do not know what the rules were for. We know what they achieved — worked out retrospectively by researchers. Almost nobody who lived under them held them for those reasons. So this part is a modern reconstruction, built from evidence none of them had, and the traditionalist enjoying it should notice that the argument being made on their behalf is not theirs. It is mine.
2The World The Rules Were Built For
Before you can judge a set of instructions you have to know what they were instructions about. The world these rules were written for had no antibiotics, no contraception, no paternity test, no welfare and no law a woman could reach. Take those five away and read the rulebook again.
2.1 — Childbirth
Start with the thing modern readers most consistently fail to imagine, because it has been solved so completely in rich countries that it has left ordinary memory.
Childbirth killed women. Not rarely, not in unlucky cases — as a normal feature of life. Before antisepsis, safe surgery and transfusion, the chance of dying in any single delivery ran at somewhere around one in a hundred, and a woman had six or eight of them.
The causes were haemorrhage, obstructed labour, and infection — the last of which was frequently introduced by the person attending. It took decades for the medical profession to accept that doctors moving between patients without washing their hands were killing them.
And a pregnancy could not be prevented. Reliable contraception did not exist. Abortion existed, was dangerous, and frequently killed the woman as well.
So set out the situation as it presented itself to somebody actually living in it. A sexual encounter carried, for a woman, a meaningful probability of a pregnancy she could not prevent and could not end, and that pregnancy carried a meaningful probability of killing her.
Whatever you think of the rules that followed, they were not a response to nothing.
2.2 — Disease
The second danger has also left living memory in rich countries, and it should not have.
Syphilis had no cure until the twentieth century. Untreated it disfigured, then destroyed the nervous system, then killed. It passed to infants in the womb. Gonorrhoea, also incurable, caused sterility and blinded newborns as they were delivered.
There was no test, no treatment and no way to know. A person could not tell whether a partner was infected, and frequently the partner could not tell either.
Put the two dangers together and price a single act.
For a woman, in a world with no contraception, no antibiotics and no test, one sexual encounter carried: a real chance of pregnancy; a real chance that the pregnancy would kill her; a real chance of an incurable disease that would disfigure and then kill her slowly; and a real chance of passing that disease to a child.
Now compare it with the same act today for a woman with access to contraception and a pharmacy. Pregnancy is preventable at a cost of a few rupees a month. The great bacterial diseases are curable in a week. The chance of dying in childbirth in a country with functioning hospitals is a small fraction of one per cent.
These are not the same act. They have the same name and they are separated by a difference in consequence of two or three orders of magnitude.
Anybody who wants to argue that the old rules were irrational has to explain how a rational person prices that first act. And anybody who wants to argue that they still apply has to explain why a rule written for the first act governs the second.
2.3 — No way to know
The third absence is the one Part Two built its argument on, and it is worth restating from the inside.
A man could not establish that a child was his. Not with difficulty — at all. Blood testing from the early twentieth century could exclude a man; it could not confirm one. Certainty arrived in 1985.
Now consider what a man in this position is being asked to do. He is asked to spend twenty years of labour, food, protection and inheritance on a child, on the basis of a belief he cannot check. In a society without a welfare state, that transfer is not a gesture. It is most of what he will produce in his life.
Any arrangement that could not give him reasonable confidence would not attract that investment. And a society in which men do not make that investment is a society in which children are raised by their mothers alone, in an economy where a woman’s independent earning capacity was frequently zero.
That is the problem. It is not a nice problem and the solution was not nice either.
2.4 — No floor
The fourth absence is the one that most changes how the rules read.
There was no state that would feed you. No pension, no unemployment payment, no housing, no support for a woman with a child and no husband. In most of the societies in this series, a woman also could not own land, could not inherit in her own right, and could not earn a wage sufficient to live on.
Which means her survival ran through a household, and losing her place in one was not a hardship. It was a category of destitution with no bottom to it.
Read the rules in Part Three again with that in view. Enforced widowhood is monstrous. It is also, from the inside of a system with no floor, a place in a household — food, shelter, a roof — for a woman who has lost the person her position depended on. The alternative on offer was not freedom. It was nothing at all.
I want to be exact about what I am and am not saying. I am not saying the widow was well treated or that the arrangement was justified. I am saying that a system offering a bad position to somebody whose alternative is starvation is doing something, and that calling it pure cruelty misdescribes the mechanism and therefore misdescribes what would be needed to replace it.
2.5 — No law you could reach
The fifth absence completes the picture. Professional policing is about two centuries old and arrived in most of the world much later than that. For most people in most places, there was no authority that would enforce an agreement, punish an assault, or recover what was taken.
Part Two, Chapter Six set out what people do in that situation: they build reputations for retaliation, because deterrence is the only protection available when there is no law. A family’s standing was its security system, and a woman’s conduct was the most visible and least fakeable signal of that standing.
Again, monstrous in operation and functional in design. Both.
2.6 — The rulebook as engineering
Now assemble it. Take a world with those five absences and ask what instructions you would write.
You would restrict sexual activity outside a committed arrangement, because the cost of it falls catastrophically on women and children and cannot be mitigated.
You would make the arrangement hard to leave, because the alternative to a household is destitution and because male investment requires confidence about duration.
You would place enormous weight on paternity confidence, because that is what buys the investment.
You would tie a family’s public standing to visible conduct, because standing is the security system and it needs a signal.
And you would enforce all of it socially rather than legally, because there is no law to reach.
That is the rulebook. Not most of it — nearly all of it, derived from five absences without any appeal to purity, honour, scripture or the nature of women.
This is the strongest thing in this part and it should be sat with. The rules are not arbitrary and they are not primarily an expression of contempt. They are what a reasonable and even a kind person might write, given those five conditions, trying to protect the people in front of them from dangers that were entirely real.
Five absences define the world these rules were written for. No contraception and no safe childbirth — roughly one death per hundred deliveries, six or eight deliveries a life. No cure for the great sexual diseases, which disfigured, killed slowly and passed to infants.
No way to establish paternity until 1985 — so a man was asked to spend twenty years of labour on a belief he could not check. No floor — no state that would feed a woman without a household, in economies where she could not own, inherit or earn. And no law anybody could reach.
One sexual act, for a woman in that world, carried a real chance of pregnancy, of dying from it, of an incurable disease, and of passing it to a child. The same act today is separated from that one by two or three orders of magnitude of consequence. They share only a name.
Now derive the rulebook from the five absences: restrict sex outside a committed arrangement, make the arrangement hard to leave, weight paternity heavily, tie standing to visible conduct, enforce it socially because there is no law. That is nearly all of it — with no appeal to purity, honour, scripture or the nature of women.
These are not the rules of people expressing contempt. They are what a reasonable and even a kind person might write, given those conditions, trying to protect the people in front of them from dangers that were entirely real.
3Children
This is the strongest empirical card the traditional position holds. Children raised by two parents who stay together do better, on nearly every measure anybody has looked at. The interesting question is not whether that is true but how much of it survives when you ask it properly.
3.1 — The raw picture
The associations are not subtle and they are not disputed.
Compared with children raised by two married parents, children raised in single-parent households are substantially more likely to leave school without qualifications, more likely to have a child as a teenager, more likely to be neither working nor studying in early adulthood, more likely to encounter the criminal justice system, and more likely to report psychological difficulties.
Landmark work in the 1990s put the schooling risk at roughly double. Subsequent studies in many countries have found the same direction, repeatedly, on many outcomes.
That is the raw finding, and anybody arguing about family structure who begins by disputing it is arguing about something that is not in dispute.
3.2 — And then the problem
Nobody is randomly assigned to a family. That single sentence governs this entire literature.
Selection: the problem that arises when people are not randomly sorted into the groups being compared, so the groups differ in ways that have nothing to do with what is being studied.
An illustration that makes it obvious. People who go to hospital die at much higher rates than people who do not. Nobody concludes hospitals are lethal, because the comparison is between the sick and the well, not between two similar groups given different treatment.
Why it matters here: every comparison of family arrangements has this shape. It cannot be removed, only reduced, and the whole quality of a study in this field is a question of how well it was reduced.
Parents who marry and stay married differ from those who do not, before any child is born. They have more money, more education, more stable employment, better health, lower rates of addiction and, on average, different temperaments. Every one of those independently affects children.
So the raw gap contains two things fused together: whatever the family arrangement does, and whatever the people who end up in each arrangement were already like.
Researchers have several ways of separating them, and they converge on a similar answer.
Comparing siblings who experienced different arrangements — one child born before a separation, one after — holds the family constant and looks only at exposure.
Following the same child before and after a change, which asks whether the child’s trajectory bends at the moment the arrangement changes.
Using changes in divorce law that made separation easier in some places and years and not others. This is the strongest design available, because a legislature changing a statute is not responding to any individual family.
What they find: the effects are real and smaller than the raw comparisons. A substantial share of the raw gap is accounted for by what the families were like beforehand — income in particular does a great deal of the work. A meaningful part is not, and survives every adjustment researchers have applied.
What they cannot show: which specific mechanism does it. Money, supervision, stability, conflict, the number of adults in the house and the number of transitions a child experiences are all candidates, and separating them has not been done convincingly.
3.3 — Boys
One finding in this literature is larger and more consistent than the rest, and it is barely present in public argument.
The effects of family disruption fall harder on boys.
Studies using large administrative datasets and comparing brothers and sisters within the same families find that boys’ behaviour, schooling and later employment are more damaged by family instability and disadvantage than their sisters’ are. The gap between boys and girls in behavioural difficulty is itself wider in disadvantaged and disrupted households.
This is the traditionalist case at its sharpest and it is almost never made. A society that stops forming stable two-parent households is not producing an evenly distributed cost. It is producing a specific, measurable, sex-differentiated cost, concentrated on boys, showing up as school failure and disconnection from work — which Part Fifteen counts and which is the largest social problem nobody is arguing about.
There is a second pattern that changes the shape of this entire argument, and it is about class rather than culture.
In the wealthy world, marriage has not declined evenly. Among the university-educated it remains common and comparatively stable. Among those without a degree it has fallen sharply. The result is that the two-parent household has become, in effect, a class marker.
Put it in two streets. On one, professional couples marry in their thirties, stay married, and raise children with two incomes and two adults’ attention. On the other, partnerships form early, dissolve often, and children pass through several household arrangements before adulthood.
The children who most need the stability are concentrated in the second street.
Which means the decline of the two-parent family is not a story about liberal values reaching everybody. It is a story about a stabilising arrangement being retained by the people who need it least.
3.4 — The comparison nobody makes
Every argument in this area assumes that the alternative to a broken family is the same family, unbroken.
Both camps do it. The traditionalist compares children of divorce with children of intact marriages and concludes the divorce caused the difference. The reformer disputes the size of the gap. Neither asks what the actual alternative was.
Because for the child in question, the counterfactual is not “the same parents, happily married.” It is “the same parents, in the marriage they were actually in” — the one that ended.
And when researchers make that comparison, the answer splits. Where a marriage was high in conflict, children generally do better when it ends. Where it was low in conflict — an unhappy marriage rather than a hostile one — children generally do worse. Estimates put the high-conflict category at something like a third of separations.
So “divorce harms children” is true on average and false for a substantial minority, and there is no way to know from the average which child is in front of you.
The general form: comparing what happened with an alternative that was never available. It is the most common error in policy argument of any kind, and it is why the family-structure evidence supports “fewer bad marriages forming” far better than it supports “fewer marriages ending.”
With that counterfactual on the table, the professional argument about causation can be stated properly.
The central methodological dispute in the field, and both sides have serious people in them.
The association appears in every country studied, across dozens of outcomes, and survives every statistical adjustment researchers have applied. Natural experiments using changes in divorce law — where no individual family chose anything — find worse adult outcomes for the affected cohorts. At some point a relationship that survives this much testing is a relationship.
Most of the raw gap disappears when income and parental education are properly accounted for, and the residue is small enough to be explained by variables nobody has measured. Single parenthood in countries with strong income support produces far smaller gaps than in countries without — which is what you would expect if the operative variable were money rather than structure.
A genuine effect remains after adjustment, it is considerably smaller than the raw numbers, it is larger for boys than girls, it is larger where income support is weaker, and it is driven more by instability and transitions than by the number of parents at any moment.
What would settle it: comparing outcomes across countries with similar family structures and very different levels of income support. This has been done partially and points towards money doing a large share of the work.
Why people care so much: because this is the one place where the traditional position has genuinely strong quantitative evidence, and both camps know it.
Which is why this chapter matters more than its length suggests. If the traditional case has a foundation anywhere, it is here.
The raw associations are large and undisputed. Children in single-parent households leave school without qualifications at roughly double the rate, and show worse outcomes on many other measures across many countries.
But nobody is randomly assigned to a family. Parents who marry and stay married differ in income, education, health and temperament before any child is born. Better designs — sibling comparisons, divorce-law changes — find the effects are real and smaller. Each camp quotes one of those two words.
The sharpest finding is barely discussed: the damage falls harder on boys. Family instability hits boys’ behaviour, schooling and later employment more than their sisters’.
And marriage has become a class marker. It remains common among the university-educated and has collapsed among everyone else — so a stabilising arrangement is being retained by the people who need it least.
The comparison nobody makes: the alternative for a child of divorce is not “the same parents, happily married.” It is the marriage they were actually in. Where conflict was high, children do better when it ends. Where it was merely unhappy, they do worse. Which means this evidence supports “fewer bad marriages forming” far better than it supports “fewer marriages ending.”
4The Vanishing Couple
The promise was freer partnerships. What arrived in much of the developed world was fewer of them. This is countable, it is happening almost everywhere, and it is the part of the traditional case that requires no interpretation at all.
4.1 — Marriage
Marriage rates across the wealthy world have roughly halved since the 1970s. That is a large change in a short period for an institution of that age.
But rates per year are a poor way to see it, because they move with the age structure of a population. The clearer measure is how many people never marry at all.
In the United States in 1980, around six per cent of forty-year-olds had never been married.
By the early 2020s that figure was around twenty-five per cent.
Put it in a room of a hundred people at their fortieth birthdays. Four decades ago, six of them had never married. Now twenty-five have.
That is not a shift in when people marry, which is what a rising average age at marriage would show. Forty is late enough that most of those twenty-five will not. It is a change in whether.
The pattern is not American. Japan and South Korea show it more sharply, with lifetime never-married shares rising steeply in recent cohorts. Southern Europe shows it. It is arriving in urban India.
4.2 — And not replaced
Cohabitation: living as a couple in one household without being married.
It is worth distinguishing three different things that a falling marriage rate could mean, because they have very different implications. People marrying later. People partnering without marrying. Or people not partnering at all.
Why it matters here: the first is not a decline in partnership, the second is a change in its legal form, and only the third is what the traditional case is describing. Chapter Four is an attempt to work out how much of each is happening.
The obvious reply is that marriage was replaced by cohabitation — that people still partner, without the ceremony.
Cohabitation did rise substantially, and for a period it absorbed much of the decline. But two things complicate the reply.
Cohabiting unions are less stable. They dissolve at higher rates than marriages, including where children are present. So a child in a cohabiting household experiences, on average, more transitions — and Chapter Three found that transitions, rather than the number of adults at any given moment, may be the operative variable.
And the growth of cohabitation has not kept pace with the decline in marriage. Across a number of countries, the share of adults in any co-residential partnership at all has fallen. Not merely unmarried — unpartnered.
These figures come from two different kinds of source, and the difference matters.
Marriage and never-married shares come from registries and censuses. A marriage is a legal event that gets recorded, and a census asks a factual question with a definite answer. This is about as reliable as social data gets.
Cohabitation and partnership come from surveys, and are harder. Whether two people living together count as partners is a judgement, definitions differ between countries, and the category has changed meaning over the period being measured — which makes long comparisons shakier than they look.
What follows: the marriage decline is solid. The claim that partnership overall has fallen is real but weaker, and anybody presenting the two with equal confidence is overstating the second.
4.3 — The recession
There is a further measure, and it is the one that most surprised the people who first reported it, because everyone expected the opposite.
Surveys of sexual behaviour in the United States show a rising share of young adults reporting no sexual activity in the previous year. Among men in their late teens and twenties the share reporting none roughly tripled over the decade to the late 2010s. Japanese surveys report high and rising shares of young adults with no sexual experience at all.
This is worth pausing on, because it inverts every prediction anybody made. The expectation, from both camps, was that liberalisation would produce more sex among the young. The measured result in several wealthy countries is less.
The traditionalist reading writes itself: the arrangement that was dismantled was not primarily a restriction on sex. It was an infrastructure for pairing people — arranged introductions, community supervision, clear expectations, a defined sequence with a defined endpoint. Removing the restrictions also removed the machinery, and nothing was built to replace it.
On that reading, what looks like liberation for the confident and attractive is abandonment for everybody else, and the measured decline in partnership is what abandonment looks like in a survey.
The numbers are agreed. What they mean is not.
Partnership is among the strongest correlates of reported wellbeing, of physical health and of longevity that has ever been measured. A society producing fewer partnerships is producing more unhappiness, more illness and more early death, and doing so at scale. The people worst affected are not the ones writing about it — they are the least advantaged, for whom the old machinery did the most work.
What declined was a system in which enormous numbers of people were compelled into partnerships they did not want and could not leave. Fewer marriages also means fewer forced marriages, fewer women trapped with violent men, and a sharp fall in intimate partner violence across the same period. Counting the missing marriages without counting the marriages that no longer have to happen is a one-sided ledger.
The people freed by the change and the people stranded by it are largely different populations. Women in bad marriages gained enormously. Men and women who would have been matched by the old machinery and are not matched by anything now have lost something with no compensation. Averaging them produces a number describing nobody.
What would settle it: measures of wellbeing among the never-partnered, tracked over time and separated by whether the state was chosen. That distinction is rarely made and it is the whole question.
Why people care so much: because both sides are describing real people and neither is describing all of them.
4.4 — What is countable and what is not
A discipline before leaving this chapter, because the traditional case is often overstated here by people who have the good arguments available and use the weak ones.
Countable and solid: marriage rates, never-married shares, partnership rates, the timing of first births, the share reporting no recent sexual activity. These are survey and registry data and they show what they show.
Countable and contested: loneliness. Measures disagree, definitions differ, and the claim that loneliness has risen sharply among the young is weaker than it is usually presented. Some datasets show it; others show it flat.
Not countable at all: whether people are less happy in some general sense, whether communities have less warmth in them, whether something has been lost that does not appear in a survey. These may be true. They are the mood, not the prediction, and Part One’s rule applies: a claim that cannot say what would count as it being wrong is not a claim.
Marriage rates across the wealthy world have roughly halved since the 1970s. The clearer measure is never-married shares: in the United States, around six per cent of forty-year-olds had never married in 1980, against around twenty-five per cent in the early 2020s. That is a change in whether, not in when.
Cohabitation absorbed part of it, but cohabiting unions are less stable, producing more transitions — which Chapter Three suggested is the operative variable. And in several countries the share in any co-residential partnership has fallen.
The measure that inverted everybody’s predictions: the share of young adults reporting no sexual activity in the past year has risen, roughly tripling among young American men over a decade. Both camps expected the opposite.
The traditionalist reading: what was dismantled was not mainly a restriction on sex but an infrastructure for pairing people — introductions, supervision, expectations, a sequence with an endpoint. Removing the restrictions removed the machinery, and nothing replaced it.
The change freed people with options and stranded people without them, and those are different populations. Which makes this, like the class gradient in Chapter Three, a distributional question wearing a moral costume.
5Cheap Sex
The most unpleasant argument in this part, and it is not stupid. If sex becomes available without commitment, the argument goes, then commitment becomes scarce — and the people who wanted commitment are the ones who pay.
5.1 — The argument
State it in its own terms, without flinching, because a version that flinches is a version that has not been understood.
Sexual economics: a framework analysing sexual behaviour as an exchange, in which sex is treated as a resource that women supply and men seek, with a "price" — paid in commitment, attention or resources — set by women’s collective behaviour rather than by any individual.
Why it matters here: this framework generates the argument in this chapter. It is a model rather than a finding, it does real analytical work, and §5.3 examines what it assumes before any evidence arrives.
Historically, the argument runs, sex outside a committed arrangement was expensive for a woman — Chapter Two priced it — and therefore scarce. A man who wanted it generally had to offer something substantial in exchange: marriage, support, permanence, the whole apparatus.
Contraception removed most of the cost. Antibiotics removed the rest. Modern arrangements for meeting people expanded the field of possible partners from a village to a city to a country.
The consequence, on this account, is that sex became obtainable without commitment. And once it is obtainable without commitment, a man who does not want commitment has no reason to offer it — which means the men most willing to commit are competing against an option that did not previously exist.
So who loses? Not the men who wanted sex without commitment; they gained. Not the small number of men most in demand; they gained enormously. The losers are women who wanted a committed partnership and now find that the thing they are offering no longer purchases what it used to, and men who are not in demand, who under the old arrangement would have been matched by machinery that no longer exists.
That is the argument. It is uncomfortable, it is put more crudely than this by most of the people who hold it, and it deserves a hearing.
5.2 — The arithmetic underneath
There is a version of this that requires no claim about sex at all, and it is stronger.
Women now substantially outnumber men in higher education across most of the developed world — in the United States earning close to sixty per cent of first degrees, with similar patterns across Europe. Part Four noted this reversal; here is what it does.
Suppose, as surveys consistently find, that women on average prefer partners of equal or greater education and earnings, and men are more indifferent to this. That preference is not required to be innate, universal or admirable. It only has to exist.
Take a hundred graduates from a city’s universities this year. Roughly sixty are women and forty are men.
If every one of those forty men partners with a graduate woman, twenty women are left over with nobody in the category they were looking in. Not because anybody was rejected. Because of subtraction.
And the same arithmetic running the other way strands men without degrees, who are competing for partners in a pool where the women’s expectations were set by a different distribution.
Now notice the property that makes this argument hard to answer. It requires no bad behaviour by anybody. Not one person in the example is selfish, exploitative or unreasonable. Every individual acts on a preference nobody would call outrageous, and the aggregate produces two groups of people who cannot find what they are looking for.
This is the strongest form of the case in this chapter, and it is not about sex. It is about counting.
5.3 — What a market metaphor does
The whole argument runs on a metaphor: that there is a market, in which sex is a good, women are the suppliers, men are the buyers, and there is a price.
Everybody in the dispute accepts the metaphor. Its advocates build on it. Its critics attack the conclusions while leaving the frame in place.
Look at what the frame does before any evidence arrives. It makes sex something women have and men want — so a woman who wants it is anomalous within the model rather than ordinary. It makes female collective behaviour the variable that sets the price, so the responsibility for the outcome sits with women. And it makes male demand a constant, the way demand for wheat is a constant, which places it outside the analysis entirely.
That last move is Part One, Chapter Eight in economic dress: female behaviour is the variable and male behaviour is the weather. The metaphor performs it automatically, before anybody has argued for anything.
The general form: a metaphor doing analytical work while presenting as description. The test is what happens if you swap the roles — if commitment is the scarce good and men are the suppliers, every conclusion reverses using the same data. Nothing in the evidence tells you which way round to run it. Somebody chose.
5.4 — And a rival explanation with better evidence
Chapter Four established the decline. This is a claim about its cause, and there is a serious competitor.
The timing fits: contraception, then the sexual revolution, then falling marriage. The mechanism is coherent and requires no exotic assumptions. And the arithmetic in \u00a75.2 operates whether or not anybody accepts the moral framing \u2014 an educational reversal of that size mechanically produces unmatched people at both ends.
Marriage did not decline evenly. It collapsed among men whose earnings and employment collapsed, and held up among men whose did not. Where regions lost manufacturing employment for reasons entirely external to them \u2014 a genuine natural experiment \u2014 young men\u2019s earnings fell, and marriage rates fell with them while single parenthood rose. Nobody in those places changed their views about sex. The factories closed.
The two accounts disagree about culture and agree about counting. Whether the driver is the price of sex or the price of labour, a large imbalance in who obtains credentials and earnings produces unmatched people. That part requires no theory of anything.
What would settle it: more shocks of the kind described. Every time an economy removes male employment from a region for external reasons, it runs the experiment again, and so far the results point one way.
Why people care so much: because one account locates the problem in women\u2019s choices and the other in the labour market. Those imply completely different responses, and only one of them requires anybody to be told how to behave.
The chapter after this takes the argument that survives best in the whole part, and it is not this one.
The argument: contraception and antibiotics removed the cost of sex, so sex became obtainable without commitment, so men who did not want commitment stopped needing to offer it. The losers are women who wanted partnership and men who were not in demand.
The stronger version requires no claim about sex at all. Women now earn close to sixty per cent of first degrees. If women on average prefer partners of equal or greater education, then in a hundred graduates \u2014 sixty women, forty men \u2014 twenty women are left over. Not through rejection. Through subtraction. It requires nobody to behave badly.
But the market metaphor is doing work before any evidence arrives. It makes sex something women have and men want, makes female collective behaviour the price-setting variable, and places male demand outside the analysis as a constant. That is Part One’s finding in economic dress: female behaviour is the variable and male behaviour is the weather.
And there is a rival with better evidence. Marriage collapsed where male employment collapsed. Where regions lost manufacturing for entirely external reasons, young men’s earnings fell and marriage fell with them. Nobody there changed their views about sex. The factories closed.
6What Marriage Does To Men
Married men earn more, drink less, commit fewer crimes and live longer. The obvious reply is that better men get married. The obvious reply is only partly right, and what remains is the most under-used argument in the traditional case.
6.1 — The pattern
On almost every measure anybody has recorded, married men do better than unmarried men.
The marriage premium: the consistent finding that married men earn more than otherwise similar unmarried men. Estimates vary by country and method and commonly fall somewhere in the range of ten to twenty per cent.
The word "premium" is doing something worth noticing — it implies the marriage caused the earnings. §6.2 is about whether that is true, and part of it is not.
Why it matters here: it is the headline number in this chapter and the one most often quoted without its caveat.
They earn more — the gap is large enough to have its own name in economics. They work more hours and are less likely to be out of the labour force. They drink less and take fewer physical risks. They report better health and they live substantially longer.
Two features of this pattern are worth noticing immediately.
It is much larger for men than for women. Marriage is associated with better outcomes for women too, but the size of the male advantage is in a different range. Whatever is happening, it is happening more to men.
And it applies to crime. This is the finding that matters most, because crime is where the social cost sits.
6.2 — Selection, and what survives it
The obvious objection is the right one to raise. Men who are employable, stable, sober and law-abiding are more likely to be chosen as husbands. The correlation could be entirely a description of who gets married rather than of what marriage does.
The way round it is to follow the same man over many years and ask whether he changes when his circumstances do.
The strongest evidence comes from long-running studies that tracked men from adolescence into late middle age, including men with substantial criminal records. Rather than comparing married men with unmarried men, these compare each man with himself \u2014 the same person, offending at one rate before a marriage and another rate during it.
What they find: offending falls within individuals during periods of marriage, by a substantial margin, and rises again if the marriage ends. Because the comparison is a man against his own earlier self, it cannot be explained by the kind of man he is.
There is a physiological finding pointing the same way. A long-term study following men in the Philippines measured testosterone before and after partnership and fatherhood. Men who became partnered fathers showed markedly larger declines than men who did not \u2014 and the men who went on to become fathers had higher testosterone beforehand, which rules out the simplest selection story.
What it cannot show: what specifically does it. Routine, supervision, a person who notices, obligations that make risk expensive, or a change in how a man sees himself are all candidates. And these are studies of particular cohorts in particular countries; the mechanism may not travel.
6.3 — The society-level version
Now the argument that follows, and it is the one with the sharpest edge.
Violent crime is overwhelmingly committed by young men, and disproportionately by young men who are not partnered. Part Four’s arithmetic explains the shape: a moderate average difference in aggression becomes a nine-in-ten outcome at the extreme.
If marriage reduces offending within individuals, then a society that produces fewer marriages is producing more offending, mechanically, without anybody’s character having changed.
And there is a natural experiment for the extreme version. Where sex ratios have been distorted — by war, by migration, or by the sex-selective abortion Part Twelve counts — societies end up with large numbers of men who will not find partners. Research on China’s skewed sex ratio has attributed a substantial share of the rise in crime over a period to precisely this. The finding is contested, and the correlation is there.
India has the same problem, for the same reason, on a similar scale. A country with several million more young men than young women has a structural feature that no amount of policing addresses.
The dispute that decides how much weight this chapter can carry.
Within-individual designs remove the selection objection by construction. The same man offends less while married and more after it ends. The hormonal evidence points the same way and rules out the simple version of selection, because the men who became fathers started higher. Something about the arrangement acts on the person.
Even within-individual studies cannot rule out that a man was already changing \u2014 maturing, finding work, tiring of the life \u2014 and that marriage is a symptom of that change rather than its cause. People marry at the point when their lives are improving, so improvement precedes the marriage and gets credited to it.
Selection is real and cannot be fully removed. A causal component survives the best available designs. Nobody has established the split, and the honest range is wide.
What would settle it: a policy that increases marriage rates without changing anything else, which is close to impossible to arrange. Studies of marriage-promotion programmes exist and their results are weak, which is evidence, though weak programmes are also just weak programmes.
Why people care so much: because if marriage changes men, then family policy is crime policy, and that is a conclusion with very large consequences that almost nobody on either side wants to own.
6.4 — Where this argument actually points
One observation before leaving the chapter, and it is the reason this argument is under-used by the people it would most help.
Every finding here is about men. Men’s earnings, men’s offending, men’s drinking, men’s hormones, men’s mortality. The claim is that a certain arrangement acts on male behaviour and improves it.
Which means the problem being described is a male problem, and the intervention being proposed is an intervention on men.
Traditionalists reaching for this argument almost always deploy it as a reason to restrict women — to make marriage more available, more compulsory, harder to leave. But nothing in the evidence supports that direction. The evidence says that being in a stable partnership changes a man. It says nothing about how the partnership should be obtained, and it certainly does not say that a woman should be required to supply one.
Chapter Nine returns to this, because it turns out to be the pattern in nearly every argument in this part.
Married men earn more, work more, drink less, offend less and live longer. The advantage is much larger for men than for women.
The obvious objection is selection — better men get chosen. The way round it is to follow the same man over decades. Studies doing this find offending falls within individuals during marriage and rises again when it ends. A long-term study measuring testosterone found men who became partnered fathers dropped sharply — and had started higher, which rules out the simple selection story.
At society level: if marriage reduces offending within individuals, a society producing fewer marriages produces more offending with nobody’s character having changed. Research on China’s skewed sex ratio attributes a substantial share of a rise in crime to surplus unpartnered men. India has the same problem for the same reason.
Selection is not fully removable and the split is unknown. But even a modest causal component, multiplied across a population, is a very large number of offences.
And notice where the evidence points. Every finding here is about men — men’s earnings, offending, drinking, mortality. The problem described is a male problem and the intervention is an intervention on men. Traditionalists deploy this as a reason to restrict women, and nothing in it supports that direction.
7The Insurance That Was Cancelled
This is the strongest argument in the part. It is made most often by women and dismissed most often by men, and it does not require a single claim about biology, morality or tradition. It is about who carries the risk of a change.
7.1 — The argument
Suppose everything the reformer says is true. The rules are unjust, they were built for a world that no longer exists, and the costs fall on women.
A woman is now asked to stop complying. What is she being offered in exchange?
Under the old arrangement she had a guarantee. It was a bad guarantee — constrained, humiliating in places, administered by people with power over her. It was also a guarantee. If she followed the rules, a household was obliged to house and feed her, her children had a claim on a lineage, and she had a defined position that could not simply be withdrawn.
The replacement is not a guarantee. It is an opportunity. She may work, own property, leave a marriage, marry whom she chooses. Every one of those is a chance rather than a promise, and every one requires a functioning labour market, enforceable law, safe transport, childcare, and a family that will not cut her off.
So the offer is: give up a bad certainty for a good chance.
That is a real trade with a real downside, and the person being asked to make it is not being irrational when she hesitates. She is pricing an option correctly.
7.2 — Why she had nothing to fall back on
The argument above assumes the woman has no outside option. It is worth explaining how she came not to have one, because the mechanism is an argument in its own right and it is the most respectable economic case ever made for the traditional household.
Specialisation: two people producing more together by each doing one thing well than by both doing everything adequately.
It is the oldest idea in economics. A village where one person farms and another makes tools produces more than two people each doing half of both, because skill accumulates with practice and switching between tasks wastes effort.
Why it matters here: applied to a household, it is the strongest economic argument for the traditional division of labour — and it contains, built into it, the reason that division became a trap.
The argument, made carefully in economics in the twentieth century, runs like this. A household is a small productive unit. It has to generate income and it has to produce a home, meals, childcare and everything else that keeps people alive and functioning.
If both adults split their time across both tasks, neither becomes very good at either, and both keep interrupting one task to do the other. If instead one specialises in earning and the other in the household, each accumulates skill, neither wastes effort switching, and the total output is higher.
Which one specialises in which is then a question of comparative advantage — and in a world where employers paid women less, where childbirth interrupted employment repeatedly, and where infants required a nursing mother, the answer fell out without anybody needing to hold a view about women.
Notice what this argument does not require. No claim that women are better at domestic work. No claim that men are better at earning. It works even if both are false, provided only that the labour market treats them differently — and it did.
So the traditional household was, on this account, not an expression of contempt or a theory about female nature. It was a firm, organised the way firms are organised, extracting the gains that specialisation produces.
7.3 — And why that is exactly the problem
Now the part that makes this the best argument in the chapter rather than merely a clever one.
Specialisation has a cost, and the cost is concentrated entirely on the specialising partner.
The person who specialises in earning acquires skills, contacts, a record and a wage — all of which are portable. If the household dissolves, he takes them with him.
The person who specialises in the household acquires skills that are real, demanding and valuable, and that no employer will pay for. Twenty years of running a home produces no salary history, no references, no promotion record and no market wage. If the household dissolves, she takes with her nothing that can be sold.
Which means the arrangement is only rational for her if it is permanent. Specialisation is a bet on the durability of the partnership, and the whole return on her side of the bet arrives only if it lasts.
Put two people at the start of a marriage and run them forward twenty years.
He has two decades of employment. A wage that rose. A profession. People who would hire him. If the marriage ends, every one of those walks out of the door attached to him.
She has two decades of work that was equally real — and there is no institution anywhere that will convert it into an income. If the marriage ends, she is a person with no recent employment at forty-five.
The arrangement did not make her dependent by accident. Dependence is what specialisation produces, and it was the price of the extra output the household enjoyed for twenty years — output he also consumed.
Which is why the old system had to make marriage hard to leave. Not from cruelty. Because an arrangement that requires one partner to make herself unemployable is not survivable for her unless the exit is closed for him.
And that yields the sharpest version of this chapter’s argument. Divorce became easy in the second half of the twentieth century. The specialisation arrangement did not immediately stop — large numbers of women were already twenty years into a bet whose terms had just been rewritten without their agreement.
A generation of women had made an irreversible investment on the assumption of permanence, and the permanence was removed retrospectively. Whatever else the traditional position gets wrong, it is right about that, and it happened to real people who are still alive.
7.4 — What the numbers say about the trade
Follow a household through a separation, using the pattern found across many countries.
After a divorce, a woman’s household income typically falls sharply \u2014 in a number of studies by something in the region of a third or more, and considerably further where children remain with her. A man’s household income falls less, and in some studies rises.
Put that in a house. The same woman, the same qualifications, the same city, one year later, running the same household on substantially less money with the same number of children in it.
Now add the Indian conditions. Maintenance exists in law and is slow and difficult to enforce. Inheritance rights exist in law and are frequently signed away under family pressure. Paid work is available to a minority of women. Returning to her own family is a request that may be refused, and Part Three explains why a family might refuse it.
The right to leave is not the ability to leave. A legal right that a person cannot exercise without destitution is a document rather than an option, and everybody in the situation knows the difference.
7.5 — Why women make this argument
One of the persistent puzzles for reformers is that large numbers of women support the arrangement, enforce it on other women, and object to its removal. The usual explanations are that they have been conditioned, or that they are complicit.
There is a simpler explanation and it does not require anybody to be deceived.
A woman deciding whether to support the old arrangement is not choosing between the system and freedom. She is choosing between the system and whatever exists where she lives. If what exists is an unenforced maintenance law, no childcare, unsafe transport and a labour market that will not employ her, then the arrangement is the better of the two options actually on the table.
Part Three found something related and worth restating here: women in communities that adopted stricter norms gained real things — standing, better marriages for their children, security for the household. That was not false consciousness. It was a purchase, made with somebody’s freedom, by people who were paying for a benefit they received.
Any argument that treats women who defend these rules as merely brainwashed has not looked at what they are being offered instead.
The whole chapter so far \u2014 and every version of this argument, on both sides \u2014 assumes that the old arrangement was insurance.
Notice what insurance actually is. You pay a premium. When the bad thing happens, you are paid, regardless of your conduct, by an institution with no interest in whether you deserve it, and you can enforce the claim against them.
Now check the old arrangement against that description.
The payout was conditional on continuous compliance, judged by the people making the payment. There was no enforcement against them. And the same household that provided the security also administered the punishment for failing to earn it \u2014 which means the protector and the threat were the same body.
That is not insurance. There is an older word for an arrangement in which somebody guarantees your safety on condition of your obedience, and in which they are also the danger you are being protected from.
The general form: a comforting name applied to a structure whose actual shape is different. And the correction matters practically, because the reform implied by “restore the insurance” is completely different from the reform implied by “build actual insurance” \u2014 which is a welfare state, and which nobody has to obey.
7.6 — The Indian version
Everything above applies with more force in India than in the countries where the argument is usually made, and it is worth setting out specifically.
Widows. India has one of the largest widowed populations in the world, and their circumstances are documented and poor. Part Three explained the historical machinery; what persists is a group of women whose position depended on a person who has died, in a country with no general old-age support.
The elderly. Part Two, Chapter Nine identified this as the problem the audit could not clear, and it belongs here too. India has no meaningful universal pension. For most families, old age means being supported by a son. That single fact makes a son an asset and a daughter a transfer to another household — and it means the insurance argument is not historical. It is a live description of how most Indian retirement is financed.
Enforcement. Maintenance after separation exists in law and moves slowly through overloaded courts. Inheritance rights exist in law and are, by consistent report, frequently surrendered under family pressure by daughters who do not wish to be the person who sued their brother. A right that costs a relationship to exercise is priced differently from one that does not.
The financial consequences of separation are among the better-measured things in this literature, because they can be traced through tax and survey records rather than asked about.
What the data shows: across many countries, women’s household income falls substantially after a separation and men’s falls much less. The gap is wider where children remain with the mother, which is most cases. The pattern holds in countries with strong welfare systems, and is far larger in those without.
What it can establish: that the financial risk of a marriage ending is distributed very unevenly, and that this is a feature of the arrangement rather than of any individual’s conduct.
What it cannot establish: the Indian magnitude. Indian data on post-separation income is thin, because separation is under-reported, informal separation is common and not recorded, and household surveys are not designed to follow a household that has divided. The direction is not in doubt. The size is not known.
Put those together and the Indian version of Chapter Seven’s argument is stronger than the version made in richer countries. The floor there is thin. The floor here has holes you can see through.
7.7 — What follows
The argument in this chapter is the strongest in the part and it points somewhere specific, so it is worth being exact.
It does not establish that the rules are good. It establishes that the transition is dangerous, and that the danger falls on the person being asked to make it.
Which is the asylum lesson from Part One, Chapter Three, arriving for the third time in this series. Removing a system that provided something, without building the replacement, does not produce liberation. It produces people who have lost something and gained a document.
And it tells you what the replacement has to be. Not a change in attitudes. Enforceable maintenance. Enforceable inheritance. Employment a woman can actually reach. Childcare. Transport she can use after dark. A pension, so that Part Two’s unsolved problem stops making sons into assets and daughters into costs.
Every one of those is expensive, boring, and does not require anybody to be persuaded of anything.
Grant that the rules are unjust and built for a vanished world. A woman is now asked to stop complying. What is she offered in exchange?
The old arrangement was a bad guarantee \u2014 constrained, humiliating, administered by people with power over her \u2014 and it was a guarantee. The replacement is an opportunity: a chance requiring a functioning labour market, enforceable law, safe transport, childcare and a family that will not cut her off. The offer is a bad certainty for a good chance, and hesitating is correct pricing rather than false consciousness.
And the dependence was not accidental. A household that specialises produces more — that is the oldest idea in economics, and it needs no claim about female nature, only a labour market that paid women less. But specialisation is portable for the earner and worthless outside the home for the other partner. It is only rational for her if the marriage is permanent, which is why the old system closed the exit. Then divorce became easy, and a generation of women found the terms of a twenty-year bet rewritten retrospectively.
The numbers support her. Women’s household income falls sharply after separation \u2014 often by a third or more \u2014 and in India maintenance is slow, inheritance is signed away under pressure, and paid work reaches a minority. The right to leave is not the ability to leave.
But it was never insurance. Insurance pays out when the bad thing happens, whatever your conduct, from an institution you can enforce against. This paid out only on continuous obedience, judged by the payer, with no appeal \u2014 and the household providing the security also administered the punishment. The protector and the threat were the same body.
Which changes what follows. “Restore the insurance” and “build actual insurance” are different programmes. The second is a welfare state — enforceable maintenance, reachable employment, childcare, transport, a pension — and nobody has to obey it.
8The Argument From Uncertainty
Part One examined the fence in the field and found it missing a word. Now that six parts have identified what the fence actually does, the argument is considerably stronger than it was — and its best version is genuinely hard to answer.
8.1 — The fence, after the investigation
Part One, Chapter Three set out the rule: do not remove what you cannot explain. And it found the flaw — the argument tells you to ask why the fence is there when the honest question is who built it and who was on the wrong side.
That objection was correct and it has now been answered, at length, by the intervening parts.
We know what these rules did. They secured inheritance. They maintained group boundaries. They managed disease and pregnancy in a world with no defence against either. They bought male investment in children. They provided a floor for women in economies that offered none. They domesticated men, apparently causally. They produced the household arrangement in which, on the best available evidence, children do somewhat better.
The traditionalist can now say something they could not say at the start of this series: we did the investigation you demanded, and the fence turns out to be load-bearing in at least six places.
That is a substantially stronger position than “leave it alone because it is old.”
8.2 — Asymmetric risk
The second component is about the shape of the bet rather than its content.
Irreversibility: the property of a change that cannot be undone by reversing the decision that caused it.
Repealing a law is reversible — you can pass it again. A generation that did not have children is not, because the decision and its consequence are separated by biology and time.
Why it matters here: the whole argument in this chapter turns on the claim that the two possible errors differ in this property, and it is worth checking that claim carefully rather than accepting it because it sounds cautious.
Suppose you are uncertain whether a change is beneficial. The rational response depends not only on the probabilities but on what happens if you are wrong in each direction.
Keeping a rule that is no longer needed costs something real: constrained lives, wasted talent, unnecessary suffering. That cost is continuous, visible, and — this is the point — reversible. A rule can be removed next year.
Removing a rule that was doing something may cost something that cannot be put back. A generation that did not form families does not form them retrospectively. A body of tacit knowledge about how to run a household, once a generation has not learned it, is not recoverable from a book. And the demographic version, which Part Nine takes up, may be the clearest case: a population that falls below replacement for long enough cannot decide to reverse it, because the people who would have had the children were not born.
So the two errors are not symmetrical. One is expensive and undoable. The other may be cheap and permanent.
Here is the version of that argument with the least rhetoric in it.
Every claim in this series has been tested against evidence from societies. There is no laboratory, no control group and no second attempt. There is one run of the experiment and everybody is inside it.
A medical trial that goes wrong is stopped, and the next trial is designed better. A society that changes its family arrangements and finds out in forty years that it went badly has no equivalent move. The forty years happened. The people are the age they are. The children who were not born are not available to be born later.
This is not an argument that change is wrong. It is an argument about what standard of evidence a person should want before making an irreversible move in a system they cannot test \u2014 and the honest answer is: a higher standard than we have for almost anything in this series.
8.3 — And who has to prove what
There is a component of this argument that is not about evidence at all, and it decides more arguments than any evidence does.
Whoever carries the burden of proof loses ties. So the question of where the burden sits is worth more than most of the facts in this series, and it cannot be settled by facts.
The traditionalist position: the arrangement exists, it has lasted, people live inside it, and the person proposing to change it must show that the change is safe. This follows directly from §8.1 and §8.2, and it is how we treat almost everything else. Nobody demands that an existing bridge prove itself before being allowed to remain standing.
The reformist position: the arrangement restricts a person’s life without her consent, and restriction is the thing requiring justification. A rule that has constrained people for four thousand years has been unjustified for four thousand years, and the passage of time is not a defence.
Both are reasonable and there is no neutral ground between them. Part One, Chapter Nine said this and declined to hide it, and it is worth repeating here because this part is written from the other side.
What can be said is that the two positions are not symmetrical in one respect. The traditionalist default protects an arrangement; the reformist default protects a person. Which of those you think should be protected first is a value, it is stated rather than derived, and everybody arguing about this has one whether or not they have noticed.
8.4 — Where the argument breaks
I said this part would drop arguments that fail. This one nearly fails, and it is worth showing exactly where, because what survives is the useful part.
The problem is that it proves too much.
Run the identical reasoning in 1828. Widow-burning has been practised for centuries. Nobody fully understands its function. The consequences of abolishing it are unknown and possibly severe. Society is a complex system, the change is irreversible, and there is only one run of the experiment.
Every premise holds. The conclusion is that sati should have been left alone.
The same applies to foot-binding, to slavery, to the abolition of duelling, to every reform in Part Two, Chapter Three. An argument that would have blocked all of them is not a decision procedure. It is a general-purpose brake, and a brake that stops everything is not a brake, it is a wall.
8.5 — And what survives, which is not nothing
Strip out the part that proves too much and something specific remains, and it is the strongest thing the traditional position holds.
The argument is not do not change. It is: build the replacement before you remove the thing, and check that it works.
That version does not block sati’s abolition, because nothing was needed to replace it. It does not block widow remarriage. It does not block letting women work or own property or choose a husband.
What it does block is the specific move this series has now identified three times. Part One found it in the emptied asylums, where an institution was closed and the promised community care was never built. Part Two found it in India’s dowry ban, where a transaction was prohibited and the market that produced it was untouched. Part Seven of this part found it in a woman being offered a document instead of a guarantee.
Each is the same failure. The rule was removed and the function it was performing was not reassigned to anything.
That is a real argument, it is empirically supported, and it is the one thing in this part that survives Chapter Nine intact.
Part One found Chesterton’s fence missing a word. Six parts later the traditionalist can answer it: we did the investigation, and the fence is load-bearing in at least six places — inheritance, boundaries, disease, male investment, women’s floor, and male behaviour. That is much stronger than “leave it alone because it is old.”
Then the shape of the bet. Keeping an unnecessary rule is expensive and reversible. Removing a rule that was doing something may be cheap and permanent — a generation that did not form families does not form them retrospectively, and children who were not born cannot be born later.
And there is one run of the experiment and everybody is inside it. No control group, no second attempt, no stopping the trial.
But the argument nearly fails, because it proves too much. Run it in 1828 about widow-burning and every premise holds and the conclusion is that sati should have been left alone. A brake that stops everything is a wall.
What survives is narrow and it is the best thing in this part: not “do not change” but “build the replacement before removing the thing.” That blocks nothing worth doing — and it blocks the failure this series has now found three times, where a rule was removed and the function it was performing was reassigned to nothing.
9Scoring It
Seven chapters of the best case that can be built. Now it gets tested against the four conditions set out at the start — and against the thing this whole series is actually about.
9.1 — The four conditions
Chapter One set the standard, taken from Part One. For a rule to be justified, four things must hold. The problem must still exist. The harm from removing the rule must exceed the harm from keeping it, counting both columns. No cheaper instrument must be available. And the burden must belong where it is placed.
Take the arguments one at a time.
The five absences (Chapter Two). As history, unanswerable — the rules were a response to real dangers and a reasonable person might have written them. As present justification, it fails the first condition comprehensively. Contraception exists. Antibiotics exist. Paternity became verifiable in 1985. Law is reachable for most people in most places. Four of the five dangers are gone. The fifth is not, and that is Chapter Seven.
Children (Chapter Three). Passes the first condition: children exist and stability affects them. Struggles on the second, because the effect is real and smaller than claimed. And fails the third badly — the damage appears driven by instability and income rather than by marriage as such, and income support is a cheaper instrument that restricts nobody. On the fourth it does not even engage: nothing in the evidence says the burden of household stability belongs to the mother rather than the father.
Partnership decline (Chapter Four). Passes the first condition cleanly; it is happening and it is countable. Fails the second as usually presented, because the credit column — collapsing intimate partner violence, fewer forced marriages, women able to leave — is real and is almost never counted. On the third and fourth, the argument identifies a missing matching infrastructure, and nothing about a matching infrastructure requires restricting anybody.
Cheap sex (Chapter Five). The arithmetic passes; the moral framing does not. Fails the third condition, because the economic account has a natural experiment behind it and this one has a fitted narrative. And fails the fourth by construction — the market metaphor assigns the burden to women before any evidence is heard.
Marriage and men (Chapter Six). Passes the first condition and has real causal evidence. Fails the fourth completely and obviously. Every finding in it concerns men. The problem described is a male problem and the intervention indicated is on men.
The insurance argument (Chapter Seven). This is the one that scores best. Passes the first condition in India, where the floor genuinely has not been built. Passes the second, for the transition. And then fails the third — because a cheaper instrument exists, is well understood, and does not require anybody’s obedience. It is a welfare state.
Uncertainty (Chapter Eight). Survives only in the narrow form: build the replacement before removing the thing. In that form it passes everything, because it is not a defence of any rule.
9.2 — The gap the case cannot cross
Now the finding that matters, and it took seven chapters of arguing the other side to make it visible.
Every argument in this part is about one of four things: household stability, partnership formation, male behaviour, or women’s economic security.
Not one of them is about the rules this series began with.
Nothing in Chapter Three requires that a woman marry inside her caste. Stability is not produced by endogamy; endogamy narrows the pool and produces matches nobody chose.
Nothing in Chapter Four requires that a widow may not remarry. A widow remarrying is a partnership forming — the exact thing Chapter Four says there is too little of. The rule and the argument point in opposite directions.
Nothing in Chapter Five requires a dowry, which by Part Two’s account is a bid in a marriage market and by Part Three’s account is the reason a family may not want a daughter at all.
Nothing in Chapter Six requires anything of women whatsoever.
Nothing in Chapter Seven requires seclusion, veiling, or a restriction on movement — every one of which reduces a woman’s ability to earn, and therefore makes the insecurity in that chapter worse rather than better.
And nothing in any of them, at any point, licenses publishing a rumour about a woman who stood in front of a police vehicle.
That is the result. The strongest case for the old rules that can be assembled from the best available evidence is a case for stable families, functioning partnership, intervention on male behaviour, and a welfare state. It is not a case for controlling women, and in several places it argues against it.
I did not set out to reach that. I set out to build the best traditionalist argument I could, and this is where it went.
9.3 — What I have actually done here
The whole exercise assumes that steelmanning is a neutral act — that building somebody’s best case is a courtesy, costing nothing, from which truth benefits.
It is not neutral, in two directions, and both are against me.
The first is against the traditionalist. Chapter One admitted that this case is a modern reconstruction built from evidence nobody in a traditional society had. So Chapter Nine has just scored an argument that nobody actually makes. The real position is not the one I built. It is closer to this is how it is done, this is what your mother did, and you will bring shame on us. I have defeated an opponent I invented, which is a rigged game running in a subtler direction than the usual one \u2014 and a person whose real reasons are untouched by all of this will be exactly as convinced tomorrow.
The second is against me. Chapters Two through Eight are now written down, at full strength, with no undercutting, by design. They are quotable. Somebody will quote them without Chapter Nine, and the quotation will be accurate.
I decided that a reader who has never encountered the good version of an argument cannot be said to have rejected it. I still think that. But it is a decision with a cost, the cost falls on people who are not me, and pretending otherwise would be the one dishonest thing in a part built to be honest.
The general form: improving an argument and then treating the improved version as the one that was on trial.
I am leaving the part as it stands. But if you carry one thing from it into an argument, carry the scorecard rather than a chapter.
Scored against the four conditions: the five absences are unanswerable as history and fail as present justification, because four of the five dangers are gone. Children passes on the problem and fails on the instrument — the damage tracks instability and income, and income support restricts nobody. Partnership decline is real and its credit column is never counted. Cheap sex loses to an economic account with a natural experiment behind it. Marriage and men has real causal evidence and concerns men entirely.
The insurance argument scores best and then fails on the third condition, because a cheaper instrument exists that nobody has to obey: a welfare state. And uncertainty survives only as “build the replacement before removing the thing,” which is not a defence of any rule.
And here is the result that took seven chapters of arguing the other side to see. Every argument in this part is about household stability, partnership formation, male behaviour or women’s economic security. Not one is about the rules this series began with — and several argue against them. A widow remarrying is a partnership forming, which is the exact thing Chapter Four says there is too little of.
The strongest case for the old rules that can be built is a case for stable families, functioning partnership, intervention on men, and a welfare state. It is not a case for controlling women.
And steelmanning was not neutral. I defeated an argument nobody makes, while writing seven quotable chapters that somebody will use without this one.
10An Honest List Of What We Do Not Know
Two lists, no hedging in either. What is genuinely unknown about the case for the old rules, and what is solid enough that any honest opponent has to concede it.
10.1 — Genuinely unknown
This part rests on outcome research, and outcome research about families has a specific and unfixable problem at its root.
How much of the family-structure effect is causal. The honest range is wide. Selection cannot be fully removed because nobody is randomly assigned to a marriage. The best designs agree the effect is real and smaller than raw comparisons; none of them can put a confident number on it.
What the mechanism is. Money, supervision, the number of adults, the number of transitions, conflict levels and parental attention are all candidates, and separating them has not been done convincingly. This matters enormously for policy: if it is money, the instrument is cash; if it is instability, the instrument is anything that reduces churn; if it is two adults specifically, the instrument is something else again.
Whether marriage changes men or selects them, and in what proportion. Within-individual designs find a real effect and cannot rule out that a man was already changing. The split is unknown and the honest range is wide.
Why partnership is collapsing. Economic precarity has the best evidence and a natural experiment behind it. Sexual liberalisation, housing costs, delayed independence, screens and changed expectations are all live. Nobody has apportioned them.
Whether loneliness has risen. Measures disagree. This is the weakest claim in the part and it is asserted with the most confidence.
What women in these arrangements actually wanted. The same hole as in Parts Two and Three. Chapter Seven argues that women defending the arrangement are pricing an option correctly rather than being deceived. That is an inference from their circumstances, not a report of their reasons, and I do not have their reasons.
10.2 — Solid
That the rules were a response to real dangers. The five absences in Chapter Two are matters of historical record. Childbirth killed at roughly one death per hundred deliveries. The great sexual diseases were incurable. Paternity was unverifiable until 1985. There was no floor and no reachable law. Any account of these rules as pure malice is false.
That children in stable two-parent households do better on average, and that the effect is smaller than the raw numbers. Both halves. Each camp quotes one.
That the effect falls harder on boys. Consistent across large administrative datasets and comparisons within families.
That marriage has become a class marker, retained by the university-educated and collapsing among everybody else \u2014 so the stabilising arrangement is held by the people who need it least.
That partnership is declining across the developed world. Never-married shares at forty rising from around six per cent to around twenty-five in the United States over four decades, with sharper versions in East Asia. And the share of young adults reporting no recent sexual activity has risen, which inverted everybody’s predictions.
That married men offend less, and that a causal component survives. Within-individual designs, plus the hormonal evidence pointing the same way.
That women lose more than men financially when a marriage ends, in every country where it has been measured, and that in India the legal remedies are largely unenforced.
That “build the replacement before removing the thing” is supported by three separate cases in this series — the emptied asylums, the dowry ban, and a woman offered a document instead of a guarantee.
And that none of the above reaches a single rule in Parts Two and Three. This is the finding of Chapter Nine and it is not a matter of interpretation. The arguments are about stability, partnership, male behaviour and economic security. The rules are about caste, widows, dowry, clothing and movement. They do not touch.
10.3 — The studies nobody wants to run
One closing observation, following the pattern of the previous parts.
Notice which questions in \u00a710.1 could be answered and have not been.
Whether the family-structure effect is money or structure could be settled by comparing outcomes across countries with similar family patterns and very different income support. Partial versions exist and point towards money. Nobody has done it properly.
Whether marriage-promotion works has been tested, weakly, and the results were weak. Whether cash support to unstable households produces the same benefits as stability has been tested in places and is rarely cited in this argument.
Each of these would resolve something. Each is expensive, unglamorous, and would produce an answer that disappoints somebody powerful.
Part One, Chapter Six applies for the last time in this part. When a countable question stays uncounted for decades, the absence is a finding. The family argument is more useful to everybody in its current state \u2014 where the traditionalist can quote the raw gap and the reformer can quote the adjusted one, indefinitely, without either being refuted.
Genuinely unknown: how much of the family-structure effect is causal, and what the mechanism is — money, instability, adults, conflict, attention. Whether marriage changes men or selects them, and in what proportion. Why partnership is collapsing. Whether loneliness has risen, which is the weakest claim asserted with the most confidence. And what women defending these arrangements actually want, as opposed to what their circumstances imply.
Solid: the rules answered real dangers, and any account of them as pure malice is false. Children in stable two-parent households do better, and the effect is smaller than the raw numbers, and it falls harder on boys. Marriage has become a class marker. Partnership is declining across the developed world and never-married shares at forty rose from about six to about twenty-five per cent in four decades. Married men offend less and a causal component survives. Women lose more financially when a marriage ends. And “build the replacement first” is supported by three separate cases in this series.
And that none of it reaches a single rule in Parts Two and Three. The arguments are about stability, partnership, male behaviour and economic security. The rules are about caste, widows, dowry, clothing and movement. They do not touch.
Last, notice which questions could be settled and have not been. The family argument is more useful to everybody unresolved — the traditionalist quotes the raw gap, the reformer quotes the adjusted one, and neither can be refuted.
The Scorecard
Seven arguments, four conditions. This is Chapter Nine as a single sheet, so that anybody quoting a chapter from this part can be shown what it scored.
| Argument | Problem still exists? | Cheaper instrument? | Burden in the right place? |
|---|---|---|---|
| The five absences Chapter Two | No — four of five are gone. Contraception, antibiotics, paternity testing, reachable law. | Not applicable. | Unanswerable as history. Fails as present justification. |
| Children Chapter Three | Yes. Stability affects children and the effect is real. | Yes — income support and anything reducing household churn. | No. Nothing says the burden is the mother’s rather than the father’s. |
| Partnership decline Chapter Four | Yes. Countable and happening nearly everywhere. | Untested — the argument identifies a missing matching infrastructure. | No. A matching infrastructure restricts nobody. |
| Cheap sex Chapter Five | The arithmetic, yes. The moral framing, no. | Yes — the economic account has a natural experiment behind it. | No. The market metaphor assigns it to women before evidence is heard. |
| Marriage and men Chapter Six | Yes, with real causal evidence. | Unknown. | No — and obviously not. Every finding concerns men. |
| Insurance Chapter Seven | Yes in India. The floor has not been built. | Yes — a welfare state, which nobody has to obey. | It is an argument about the transition, not about the rules. |
| Uncertainty Chapter Eight | Survives only as “build the replacement first.” | In that form it is not a defence of any rule. | Passes, because it demands nothing of anybody. |
Read the right-hand column downwards. That is the finding of this part.
Every argument is about household stability, partnership formation, male behaviour or women’s economic security. Not one is about caste endogamy, enforced widowhood, dowry, child marriage, seclusion, or a rumour published about a woman at a protest. Several argue against them: a widow remarrying is a partnership forming, which Chapter Four says there is too little of.
Sources & further reading — Part 6
Glossary
Every hard word used in this part, in plain English. Each was explained where it first appeared.
| Term | Plain meaning |
|---|---|
| Asymmetric risk | A situation where being wrong in one direction costs far more than being wrong in the other, so the right level of caution differs by direction regardless of the probabilities. |
| Cohabitation | Living as a couple without marriage. It absorbed part of the decline in marriage and dissolves at higher rates, so children in cohabiting households experience more transitions. |
| Counterfactual | What would have happened otherwise. For a child of divorce it is not “the same parents, happily married” but the marriage they were actually in — which is why the evidence supports fewer bad marriages forming better than fewer marriages ending. |
| Irreversibility | The property of a change that cannot be undone by reversing the decision that caused it. Repealing a law is reversible; a generation that did not have children is not. The whole argument in Chapter Eight turns on whether the two possible errors differ in this respect. |
| Marriage premium | The consistent finding that married men earn more, work more and live longer than unmarried men. Partly selection, and a causal component survives within-individual designs. |
| Natural experiment | A situation where something outside anybody’s control divided people into groups — a change in divorce law, a factory closing for reasons external to a town. The strongest evidence available where real experiments are impossible. |
| Selection | The problem that people who end up in an arrangement differ from those who do not, before the arrangement exists. Governs the entire family-structure literature. |
| Sexual economics | The framework treating sex as a good women supply and men demand, with a collectively set price. Generates the cheap-sex argument — and assigns the burden to women through its metaphor rather than through evidence. |
| Specialisation | Two people producing more together by each doing one thing well than by both doing everything adequately. The strongest economic case for the traditional household — and the reason it became a trap, because the earner’s skills are portable and the other partner’s are not. |
| Steelman | The strongest possible version of a position you disagree with, with the weak arguments removed. The opposite of a strawman. Not a neutral act, as Chapter Nine argues. |
| Within-individual design | Comparing a person with their own earlier self rather than with other people. Removes selection by construction, which is why it carries the weight in Chapter Six. |