Where We Left Off
Before we begin
Part Six built the strongest case for the old arrangement that could be assembled, argued it without hedging for seven chapters, and then scored it. The result was that every argument in it turned out to be about household stability, partnership, male behaviour or women’s economic security — and not one of them reached a single rule this series began with. Several argued against them.
The obvious next step is to do the same job on the other side.
That is what this part does, and I want to be precise about a structural difference between the two, because it matters and Chapter Nine returns to it. Part Six was written as an advocate — I built an argument nobody had made. This part is written as an auditor — the arguments already exist, in enormous quantity, made by people far better placed to make them than I am. What is missing is not advocacy. It is somebody going through them one at a time and saying which ones the evidence supports.
So this part does three things.
It separates the movements. “Feminism” is used as though it named one position. It names at least four, they disagree about fundamental questions, and several of them regard the others as the problem.
It corrects a history. Indian feminism is routinely described, by its opponents and sometimes by its supporters, as a Western import. It is not. A woman opened a school for girls in Pune in 1848, another published a book in 1882 arguing that men were morally worse than women, and a third wrote a feminist utopia in 1905 in which men were kept in seclusion. None of them was waiting for a translation.
And it scores the claims. Supported, contested, or falsified — with the specifics named in each case. Some claims made in feminism’s name are among the best-supported findings in social science. Some are dead. A person who cannot tell you which are which does not have a view about feminism; they have a loyalty.
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.
An empirical claim says something about how the world is, and can be checked. A moral claim says something about what is owed, and cannot be checked by measurement. A strategic claim says what a movement should do, and is judged by whether it works.
Why it matters here: political movements make all three constantly, in the same sentences, and the whole job of this part is separating them. A moral claim is not refuted by data. An empirical claim is not defended by conviction. And Part One, Chapter Two showed what happens when different kinds of claim are welded together and argued as one.
A Word Box appears the first time a hard word does — never later, never only in a glossary.
One figure, before any argument, because it does more to explain the modern earnings gap than every other explanation combined.
Follow a man and a woman with the same education, the same job and the same earnings. Then follow them past the birth of a first child.
His earnings continue on their previous path. Hers drop sharply — and in the country where this was measured most precisely, they never recover. Not for the next decade, not for the next two.
In Denmark, where the records allow the calculation to be done properly, researchers found that by the 2010s something in the region of four fifths of the entire remaining gender earnings gap was attributable to this one event.
Not to unequal pay for the same work. Not to men being preferred at interview. To children — and to the fact that a society decided, without ever voting on it, which parent absorbs them.
A finding that large invites suspicion, so it is worth knowing exactly how it was obtained.
That finding comes from a design worth understanding, because it is the strongest kind available outside a laboratory.
What was done: researchers used national administrative records covering an entire population over decades — every person’s earnings, employment and children, linked. Rather than comparing mothers with non-mothers, which runs straight into the selection problem from Part Six, they compared each person with their own earnings path before the birth, and against the path of a similar person whose first child arrived later.
What that buys: the comparison is a woman against herself. It cannot be explained by the kind of woman she is, because she is the control.
What it cannot show: why the burden lands where it does. Biology, employer behaviour, partner behaviour, policy and preference all remain candidates, and the design says nothing about which. It measures the penalty with great precision and is silent on its cause.
An Argument box appears wherever competent people genuinely disagree, and there is one about this part’s whole undertaking.
Worth settling before eight chapters of doing it.
A movement that makes factual claims in public, and uses them to argue for laws and budgets, has invited exactly this. If the claims are wrong, the policies built on them are built on nothing, and no amount of good intention repairs that.
Movements are not research programmes. They are coalitions doing politics, and they are judged by what they achieve, not by the accuracy of their slogans. Scoring feminism’s claims is like scoring the claims of a trade union or an independence movement: technically possible, and beside the point.
And underneath that argument sits something both sides are taking for granted.
Both sides above assume that a movement’s claims and its achievements are connected — that it won because it was right, or that being wrong should have cost it.
Movements do not work like that. They win through mobilisation, coalition, timing and the exhaustion of opponents. The arguments that mobilise best are chosen for their power to move people, and there is no reason those should be the same arguments that survive scrutiny.
Which produces the situation this part has to navigate. Feminism achieved enormous legal change — property, franchise, divorce, employment, the criminalisation of things that had not been crimes. Several of the empirical claims made along the way have since failed. Both of those are true and neither bears on the other.
The general form: assuming the persuasive argument and the correct argument are the same argument. They rarely are, in any movement, on any subject — and a person who discovers this about a movement they dislike, and not about one they support, has discovered nothing.
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
This part contains a chapter listing claims that failed, and it is specific. Not “some feminists have said silly things,” which is true of every movement and worth nothing. Named claims, with what was asserted, what was found, and by whom. If that chapter reads as an attack, check Chapter Six’s opening: nearly every one of those claims was overturned by researchers working inside the tradition, not by its opponents.
It also contains a chapter on claims that hold, and that chapter is longer. Several of them are among the more robust findings in the social sciences, and a reader who arrived expecting a demolition will find the opposite in Chapter Four.
My own position, and the asymmetry. Part Six gave the traditional case a lawyer. This part gives feminism an examiner. Those are not the same treatment, I chose it deliberately for reasons Chapter Nine sets out, and the reader is entitled to weigh it. I do not think the choice was wrong. I do think it should be visible rather than buried, and it is the kind of thing a writer normally hopes nobody notices.
1Four Movements With One Name
“Feminism says” is the opening of an unreliable sentence, whoever is speaking. There are at least four positions under the word, they disagree about fundamental questions, and several of them regard the others as part of the problem.
1.1 — The four
Start by laying them out, because most people arguing about this have met one and assume it is the whole.
Liberal feminism. The problem is exclusion. Women were kept out of property, law, education, employment and the vote, and the remedy is to remove the barriers and let them in on the same terms as everybody else. This is the strand that produced almost all of the legal change of the last two centuries, and it is what most people mean by the word without knowing there is another kind.
Radical feminism. The problem is not exclusion from a good system but the system itself. On this account male dominance is not an accident of history to be legislated away but the organising structure of society, reaching into the family, sexuality and the household — which is what the phrase about the personal being political was for. The remedy is not entry. It is transformation.
Socialist feminism. The problem is the way women’s unpaid domestic labour underwrites an economy that does not count it. On this account a woman succeeding in a corporation has not solved anything; she has joined the thing that was extracting the labour. The remedy runs through economics rather than through law.
Difference feminism. The problem is that qualities and work associated with women — care, relationship, nurture, the running of households — are systematically undervalued. The remedy is not to make women more like men but to revalue what women already do. This strand takes sex differences seriously and treats them as a resource rather than an embarrassment.
And running across all four, a fifth position that is less a strand than a critique.
Intersectionality: the idea that a person’s situation is produced by several categories acting together rather than by any one of them separately — so a Dalit woman’s position is not simply a woman’s position plus a Dalit’s position, but something the two make jointly.
The term was coined in American legal scholarship in the late 1980s. The observation is much older in India, where the interaction of caste and sex was the first thing the movement in §2.1 ran into.
Why it matters here: it is the basis of the sharpest internal criticism the Indian women’s movement has faced, and it is domestic rather than imported.
The intersectional and Dalit critique. That “women” is not one category. A woman’s situation is produced jointly by sex and by caste, class, religion and region, and a movement led by women at the top of those hierarchies will produce a programme suited to them. In India this critique is specific and it is domestic: that the mainstream women’s movement was substantially upper-caste, and that its priorities reflected that.
1.2 — Where they collide
These are not four emphases within one view. On several questions they give opposite answers, and the disputes are conducted with more heat than any argument with outsiders.
Is selling sex work or exploitation? One strand holds it is an institution of male dominance and should be abolished, with the buyers criminalised. Another holds it is labour, that criminalising it endangers the people doing it, and that the abolitionist position is a middle-class judgement about poor women. Both call themselves feminist. Neither regards the other as a variation on itself.
Is staying at home oppression or a choice? Liberal feminism’s whole case rests on women entering paid work. Difference feminism regards that as accepting a male standard and abandoning the argument that domestic work has value.
Should differences be minimised or celebrated? One strand’s central empirical claim is that psychological sex differences are largely constructed and will diminish. Another’s central claim is that women bring something distinct and that erasing it is the injury. Part Four’s findings are welcome news to one and a problem for the other.
And in India, whether to reform religious personal law. This split the Indian women’s movement genuinely and painfully. Reforming personal law would improve many women’s legal position. It would also hand a weapon to a politics that wants to reform minority practice for reasons that have nothing to do with women. Feminists took both sides, in public, and the argument is not resolved.
1.3 — What this does to the sentence
The consequence is that “feminism claims X” is nearly always false, and it is false in a specific and useful way.
An opponent picks the strand with the least defensible position on a given question and attributes it to the whole. A supporter picks the strand with the strongest and does the same. Both sentences are accurate about somebody and misleading about the movement, and neither person is lying.
So this part will not say what feminism claims. It will name claims, say who makes them, and score them individually. Where a claim is contested inside the movement, that will be stated, because a claim that half of feminism rejects is not a claim about feminism.
At least four positions share the name. Liberal — the problem is exclusion, the remedy is entry, and this strand produced nearly all the legal change. Radical — the problem is the system itself, not exclusion from it. Socialist — the problem is uncounted domestic labour, and a woman succeeding in a corporation has joined the thing extracting it. Difference — the problem is that women’s work is undervalued, and the remedy is to revalue it rather than make women more like men.
Across all four, the intersectional and Dalit critique: that “women” is not one category, and that a movement led from the top of a hierarchy produces a programme suited to the top. In India this critique is domestic and specific.
They collide on real questions. Is selling sex work or exploitation. Is staying home oppression or a choice. Should differences be minimised or celebrated — where Part Four’s findings are good news for one strand and a problem for another. And whether to reform religious personal law, which split the Indian movement genuinely.
So “feminism claims X” is nearly always false. An opponent picks the weakest strand and attributes it to all; a supporter picks the strongest and does the same. Both are accurate about somebody. This part will name claims, say who makes them, and score them one at a time.
2It Did Not Come From The West
The accusation that Indian feminism is a foreign import has a date problem. A woman opened a school for girls in Pune six months before the first women’s rights convention was held in America.
2.1 — 1848
On the first day of January 1848, Savitribai Phule and her husband Jyotirao opened a school for girls in a house in Pune. She taught in it. She is generally regarded as the first Indian woman to work as a teacher in the modern sense, and the school was for girls of all castes at a time when that was itself the offence.
What happened to her for doing it is recorded. She was abused in the street on her way to the schoolroom, and reportedly carried a second sari to change into after arriving.
Put that date beside the one usually treated as the beginning.
The Seneca Falls Convention — the meeting in New York generally described as the founding event of the American women’s rights movement — took place in July 1848.
The Pune school opened in January 1848.
Six months earlier. In a country under colonial rule, by a woman from a caste excluded from literacy, teaching girls from castes excluded from literacy, while being abused in the street for it.
Whatever Indian feminism is, it is not a translation of something that had not happened yet.
2.2 — And then it kept happening
The 1848 date would be a curiosity if it were isolated. It is not.
1882. Tarabai Shinde published Stri Purush Tulana — a comparison between women and men — in Marathi, prompted by the case of a widow prosecuted for infanticide. Its argument is that men are morally worse than women and that the entire apparatus of blame is inverted. It is frequently described as the first modern Indian feminist text and it was written for an Indian audience in an Indian language.
1887. Pandita Ramabai, a Sanskrit scholar honoured for her learning by the pandits of Calcutta, published The High-Caste Hindu Woman. She went on to found institutions for widows and for women with nowhere to go.
1905. Rokeya Sakhawat Hossain published Sultana’s Dream in an Indian magazine. It describes a country in which women run everything and men are kept in seclusion, having proved unfit for public life. It was written in English, by a Bengali Muslim woman, as a satire on purdah — inverting the practice she lived under, in the language of the people ruling her country, in 1905.
Then the organisations. A women’s association in 1917. A national council in 1925. An all-India conference in 1927. All of them Indian, all of them predating independence, all of them arguing about Indian law.
This chapter is unusual in the series because its evidence is almost entirely documentary, which makes it stronger than most of what surrounds it.
What the evidence is: published books with dates and printers. Magazine issues. School and institutional founding records. Organisational constitutions and conference proceedings. Government reports with publication dates. Court judgments.
Why that matters: nothing here depends on a survey, a sample, a statistical model or an interpretation of somebody’s motives. The claim is that a text existed in a year, and the text exists.
What it cannot show: reach. A book published in 1882 tells you somebody wrote it and somebody printed it. It does not tell you how many people read it, whether anybody agreed, or whether it changed a single household — and the honest position is that these were minority voices in their own time, as reformers usually are.
2.3 — Different problems, different movement
The most useful thing about this history is not the dates. It is what the arguments were about, because the Indian and European movements were addressing different worlds.
European and American feminists of the nineteenth century were arguing about property rights for married women, the franchise, and access to professions.
Indian feminists of the same period were arguing about child marriage, widow remarriage, the treatment of widows, purdah, temple entry, and — from the beginning — caste. The Pune school was a caste offence before it was a gender offence.
That difference matters, because it means the two movements were not the same movement at different stages of development. They were separate responses to separate problems, and the Indian one was engaging from its first year with a question the European one did not have.
The accusation is constant and it deserves a proper answer rather than an indignant one.
The vocabulary is English. The theoretical frameworks taught in Indian universities are largely Western in origin. The organisational forms — the conference, the association, the petition — were colonial. And the nineteenth-century reformers were operating inside a colonial education system, addressing a colonial state, using arguments that state would recognise. Independence of origin is not established by being early.
The Pune school predates Seneca Falls. Tarabai Shinde wrote in Marathi for Marathi readers. The problems addressed — child marriage, widowhood, purdah, caste — were Indian problems with no European equivalent, and no imported framework generates them. A movement is defined by what it is fighting, and this one was fighting things that did not exist in the countries it is accused of copying.
Indian feminism has always been in conversation with foreign ideas, as every Indian intellectual movement has, including the nationalism that accuses it. Colonial encounter supplied vocabulary and institutional forms. It did not supply the grievance, the analysis, or the people. Part Three, Chapter Nine established that nothing in this series has a single origin, and this is another composite.
What would settle it: nothing, because it is a question about origin and Part One, Chapter One established that origin does not determine legitimacy. That is the genetic fallacy, and Part Three, Chapter Nine caught me committing it deliberately.
Why people care so much: because “foreign” is not an argument about a claim. It is a way of relocating the person making it, which Part Three identified as the manoeuvre assembled in 1891 and hardened in 1927.
There is something further underneath that dispute, shared by both sides of it, and it is the more interesting thing.
Both sides of that argument — the Indian conservative and the Western feminist — share a premise neither states: that there is one feminism, with an origin, which spread.
The conservative uses it to say the thing arrived from elsewhere and is therefore alien. Western accounts use it to describe waves emanating outward from Europe and America, with other countries joining later. Both are describing diffusion from a source.
The evidence does not look like diffusion. It looks like several movements arising independently, in different places, in response to different local problems, at roughly the same time. Pune in January and New York in July did not learn it from each other. Neither had heard of the other.
Part Two found exactly this shape once before. Farming appeared independently in at least half a dozen places among people with no contact — and the conclusion drawn there applies here. When separate populations arrive at the same arrangement without copying, the situation is producing it, not the culture.
The general form: reading independent invention as transmission, because a single origin is easier to argue about. A thing with one source can be rejected as foreign or claimed as an inheritance. A thing that arose in six places at once can only be responded to.
2.4 — And then it was restarted, by a government report
One more piece of this history, because it explains the shape of the movement in India today and is almost unknown outside it.
After independence the women’s question largely went quiet in Indian politics, for reasons Part Three, Chapter Seven set out — the nationalist settlement had answered it in advance by declaring the home already sovereign.
In 1974 a government committee published a report on the status of women in India. It had been expected to document progress.
It documented the opposite. On several central measures — the ratio of women to men in the population, women’s share of paid work, women’s political participation — the report found that things had got worse since independence.
That finding, produced by the state’s own committee, is generally credited with restarting the Indian women’s movement. What followed within a decade was a campaign over a Supreme Court judgment acquitting policemen of raping a girl in custody — a case that produced an open letter from four law professors, nationwide protest, and a change to the criminal law in 1983.
Note what that sequence contains. A state committee producing an inconvenient count. Academics writing a letter. Street protest. Statutory change. It is an entirely Indian sequence, conducted in Indian institutions, about an Indian case, and it is the founding event of the movement as it now exists.
January 1848: Savitribai Phule opens a school for girls in Pune and teaches in it, taking a second sari to change into after being abused in the street. July 1848: the Seneca Falls Convention. Six months later.
Then it kept happening. 1882, Tarabai Shinde publishes in Marathi that men are morally worse than women. 1887, Pandita Ramabai publishes on the high-caste Hindu woman. 1905, Rokeya Hossain publishes a utopia in which men are kept in seclusion and women run the country — a satire on purdah, written in English, in 1905.
And they were arguing about different things. Europeans argued about property and the franchise. Indians argued about child marriage, widowhood, purdah, temple entry and — from the first year — caste. The Pune school was a caste offence before it was a gender offence.
Both the Indian conservative and the Western account assume one feminism with an origin that spread. The evidence looks like independent arising in several places at once — the same shape Part Two found in farming, where the conclusion was that the situation produces it, not the culture.
And it was restarted by a government report. In 1974 a state committee expected to document progress found that on the sex ratio, on women’s paid work and on political participation, things had got worse since independence. That count, produced by India’s own committee, is what relaunched the movement.
3The Claims, Extracted
Before anything can be scored it has to be separated. A movement’s output is a mixture of findings, values and slogans, and only one of the three is the kind of thing that can be right or wrong.
3.1 — Three kinds of sentence
Part One’s central tool was separating three claims welded into one sentence. The same operation is needed here and it separates different things.
An empirical claim says how the world is. Women do more unpaid work than men. The pay gap is caused by discrimination. Sex differences in personality are socially produced. Each can be checked, and each is either true, false, or currently unresolved.
A moral claim says what is owed. Women are entitled to equal standing. A person’s opportunities should not depend on their sex. No measurement bears on these. They cannot be confirmed by a study or refuted by one, and Part One, Chapter Four established why.
A strategic claim says what should be done. Quotas work. Legal change should come before cultural change. These are judged by results, and they can fail without anything moral or factual being wrong.
Most public sentences about feminism contain at least two of these fused, and the fusion is where the arguments go wrong. Somebody presents a moral claim, an opponent attacks an empirical one, and both leave satisfied.
3.2 — The standard this part applies
Only empirical claims are scored here. That is not a demotion of the others; it is the only category that can be scored at all.
Three verdicts are available.
Supported. The claim has been tested by people trying to break it, using designs strong enough to have found it false, and it survived. Repeatedly, across countries or methods.
Contested. Competent researchers disagree, the evidence is mixed, or the finding has replication problems. This verdict is not a polite way of saying false. Several claims here are contested and may well be true.
Failed. The claim was made, it was specific enough to be checked, it was checked, and the evidence went the other way.
One rule governs all three, and it is the one this series has applied to every camp. A claim is scored on the evidence, not on whether the person making it was sympathetic. Part Six scored the traditional case that way and this part scores this one identically.
3.3 — Why this is harder for a movement than for a theory
A scientific theory has an agreed statement. A movement does not, and this creates a real problem that is worth acknowledging rather than exploiting.
Any claim attributed to feminism can be disowned. Serious feminists never said that. Sometimes true — Chapter One showed how many positions the word covers. Sometimes it is a way of never being wrong about anything.
The discipline this part adopts: a claim is scored if it was made in public by people with standing in the movement, used to argue for a policy, and repeated widely enough to shape what non-specialists believe. That excludes anything found in one obscure paper. It includes things that appeared in campaigns, in training materials, in legislative testimony and in ordinary conversation.
By that standard, everything in Chapters Four to Six qualifies, and I have tried to note where a claim is disputed inside the movement rather than pretending it is universal.
Three kinds of sentence get fused. An empirical claim says how the world is and can be checked. A moral claim says what is owed and no measurement touches it. A strategic claim says what should be done and is judged by results.
Most public argument fuses at least two, so somebody presents a moral claim, an opponent attacks an empirical one, and both leave satisfied.
Only empirical claims are scored here, with three verdicts. Supported — tested by people trying to break it and it survived. Contested — competent researchers disagree, which is not a polite way of saying false. Failed — specific enough to check, checked, and the evidence went the other way.
A claim is scored if it was made publicly by people with standing, used to argue for a policy, and repeated widely enough to shape what non-specialists believe. And it is scored on the evidence, not on whether the person making it was sympathetic — which is exactly how Part Six scored the other side.
4What Holds
This is the longest of the three scoring chapters, which will surprise a reader who came for the demolition. Several claims made in feminism’s name are among the better-supported findings in the social sciences, and one of them was settled by a randomised experiment across Indian villages.
4.1 — The historical claims
Start with the ones nobody disputes, because they are the foundation and they are frequently skipped past as though everyone already agrees.
Women were systematically excluded from property ownership, from inheritance in their own right, from higher education, from the professions, from the franchise, and from legal personhood in specific respects — in most of the world, for most of recorded history, by law rather than by custom alone.
This is not an interpretation. It is a description of statutes, and the statutes are published. Part Three found one such provision still in force in India.
Verdict: supported, and not contested by any serious person.
And a second historical claim, which is contested by some: that removing those barriers produced very large changes. Within a century of the relevant legal reforms, women went from a small minority to a majority of university graduates in most of the developed world, and from near-absence to substantial presence in medicine, law and public administration. Whatever else is disputed, the barriers were load-bearing.
4.2 — Unpaid work
The claim: that women perform a large majority of unpaid domestic and care work, that this work is economically substantial, and that it is excluded from the measures by which countries judge themselves.
India ran a national time use survey, asking people to account for their day.
Women reported spending roughly five hours a day on unpaid domestic work. Men reported roughly an hour and a half.
Put that in a week. The difference is somewhere in the region of twenty-four hours — a full extra day, every week, performed by one half of the population, for no wage, and counted in no national account.
Now put it in a working life. Over forty years, that difference is roughly equivalent to twelve additional years of unpaid full-time work.
This is not disputed by anybody. It is measured by governments, using their own surveys, and published. The dispute is entirely about what should follow from it.
Verdict: supported. Time use surveys in dozens of countries find the same direction, with India at the more extreme end. The exclusion from national accounts is a definitional fact, not an allegation.
4.3 — The child penalty
This is the most important entry in the chapter, because it is a feminist finding that overturned a feminist claim.
The child penalty: the permanent drop in a person’s earnings following the birth of a first child, measured against the path their own earnings were on beforehand.
The word "penalty" is not moral language here. It is the technical name for a measured quantity — the vertical distance between the earnings path a person was on and the one they ended up on.
Why it matters here: it is measured for both parents, and in every country studied it is close to zero for one of them.
For decades the pay gap was described in public campaigns as women being paid less for the same work. Chapter Six scores that version and it does not survive.
What replaced it is stronger. Using national administrative records covering whole populations, researchers tracked men and women through the birth of a first child. Before the birth, their earnings paths are indistinguishable. After it, hers falls sharply and does not return to the previous path — not in five years, not in ten, not in twenty.
In Denmark, where the records allow the cleanest calculation, the share of the total gender earnings gap attributable to this single event rose over three decades to something in the region of four fifths. The same pattern, at varying magnitudes, has been found across many countries.
Verdict: supported, strongly. The design compares a woman with her own earlier earnings path, which removes the selection problem that governs Part Six. This is among the best-identified findings in labour economics.
Everybody arguing about the pay gap assumes there are two possible explanations: discrimination or preference.
One camp says employers pay women less. The other says women choose differently — fewer hours, different fields, more time with children. The entire public argument is conducted between those two options.
The child penalty is neither, and that is why it took so long to find.
No employer is deciding to pay mothers less. And no woman is freely choosing a twenty per cent permanent earnings cut in the way a person chooses a career. What is happening is structural: work is organised around uninterrupted availability, care is organised around one person absorbing the interruption, and the two are incompatible. Neither institution intended this. Both would deny responsibility, correctly.
The general form: a false dichotomy that hides the mechanism. When the only options on offer are somebody’s bias and somebody’s choice, any cause that is nobody’s bias and nobody’s choice becomes invisible — and structural causes are usually of exactly that kind.
Notice what follows practically. If it is discrimination, the remedy is enforcement. If it is preference, there is no remedy. If it is structural, the remedy is neither — it is changing how work and care are arranged, which is expensive, boring, and requires nobody to be blamed.
4.4 — Undercounted violence
The claim: that violence against women within households and relationships was massively under-recorded, under-prosecuted, and in some respects not legally recognised as violence at all.
The evidence is documentary. For long periods there was no offence corresponding to violence by a husband against a wife in the ordinary sense; the marital rape exemption in Part Three is one surviving instance. Reporting rates measured against survey prevalence show large gaps in every country where both have been measured. Conviction rates for reported sexual offences are low nearly everywhere.
Verdict: supported. The claim that these were undercounted is established by comparing survey prevalence with recorded crime — two independent measures of the same thing, differing by a large factor.
And there is a specifically Indian instance of this claim producing change, worth recording because it is recent and because Part Three left it hanging.
After a case in Delhi in late 2012 that produced sustained national protest, a committee headed by a former Chief Justice was constituted to recommend reform of the criminal law. It reported in under a month. Legislation the following year broadened the legal definition of rape, created offences for stalking, voyeurism and acid attacks, and increased penalties.
The committee also recommended removing the marital rape exception described in Part Three. Parliament did not. Which is a precise illustration of Chapter Three’s categories: a movement made an empirical case about undercounted violence, it was accepted, a strategic recommendation followed, and the part touching the household was the part that did not pass.
What is not settled by that evidence is prevalence itself. Estimates vary widely depending on definitions and instruments, and Chapter Five takes up a specific case where the variation is an order of magnitude and the figure is quoted without it.
4.5 — The experiment India ran on itself
The claim: that women in positions of political power produce different decisions, and that seeing women in power changes what girls expect of themselves.
These sound like the kind of claims that can never be settled, because you cannot randomly assign leadership. Except that India did.
Random assignment means deciding who gets a treatment by a process unconnected to anything about them — a lottery, a rule based on an arbitrary number.
It is the single most powerful tool in empirical research, and the reason is worth stating plainly. If assignment is random, the two groups cannot differ systematically beforehand, so any difference afterwards must come from the treatment. The selection problem that governs Part Six — that people who end up in an arrangement differ before it — cannot arise.
Why it matters here: almost nothing in this series has it. This does.
What follows is therefore not a correlation and not an inference. It is the result of an experiment, run at national scale, by a government that was not trying to run one.
A constitutional amendment in the early 1990s reserved a third of village council leadership positions for women. Crucially for research purposes, the villages selected were chosen by a rule that amounted to random assignment — which means India accidentally ran one of the largest randomised experiments in the history of political science.
What was found on policy: councils led by women invested differently, and the difference tracked what women in those regions had said they wanted. In one state that meant more spending on drinking water; in another, more on roads. Not a general “female” priority — the local priority that local women had stated.
What was found on aspirations: after two rounds of reservation, the gap between parents’ aspirations for sons and for daughters narrowed substantially, adolescent girls’ own aspirations rose, and the gender gap in adolescent educational attainment was closed. Girls also spent less time on household work.
Why this carries so much weight: the villages were assigned randomly. There is no selection story. The difference cannot be that better villages got women leaders, because assignment was not by merit, ambition or wealth.
What it cannot show: whether the effects persist once reservation ends, whether they transfer to national politics, or whether the mechanism is the leader’s decisions or simply the fact of being seen. And it is one country.
Verdict: supported. Both claims — that representation changes decisions, and that visible female authority changes what girls expect — have randomised evidence behind them at national scale. Very few claims in this entire series are supported this well.
It is worth noting who this inconveniences. It is evidence for quotas, from India, produced by economists, using a policy that Indian conservatives largely opposed. And it is evidence against the argument in Part Five that representation efforts achieve nothing — because here is a representation effort, randomised, that measurably achieved something.
Supported and undisputed: women were excluded from property, inheritance, education, the professions and the franchise by statute. And removing those barriers produced very large changes, which means the barriers were load-bearing.
Supported: Indian women report roughly five hours a day of unpaid domestic work against about an hour and a half for men — a full extra day every week, counted in no national account. Governments measure this themselves.
Supported strongly: the child penalty. Earnings paths are indistinguishable before a first birth; hers falls sharply after and never returns. In Denmark, around four fifths of the remaining gap traces to this one event. The design compares a woman with her own earlier self.
And it is neither discrimination nor preference — no employer decides to pay mothers less, and no woman chooses a permanent twenty per cent cut. It is structural: work organised around uninterrupted availability, care organised around one person absorbing the interruption.
Supported by randomised evidence at national scale: India reserved a third of village council leaderships for women, assigned randomly. Women leaders spent differently, matching what local women had said they wanted. And after two rounds, the gender gap in adolescent educational attainment was closed. Very little in this entire series is supported that well.
5What Is Contested
Three claims that are genuinely open. One of them has an industry built on top of it that is considerably more confident than the evidence underneath.
5.1 — Stereotype threat
The claim: that reminding people of a negative stereotype about their group, immediately before a test, depresses their performance on it. Applied to women and mathematics, the finding was that simply mentioning sex before an exam produced a measurably worse result.
It is an elegant idea. It became one of the most cited findings in social psychology, it was taught in teacher training, and it shaped a generation of interventions.
The problem is replication. Attempts to reproduce the effect, particularly in large samples of schoolchildren rather than selected university students, have often failed. Analyses pooling the published studies have found signs that results showing the effect were more likely to be published than results showing none — which inflates the apparent size of anything.
Verdict: contested, with recent weight against the strong version. Something may be there in specific circumstances. The claim that it explains a substantial part of measured performance gaps is not supported by the current evidence, and it was asserted with far more confidence than that.
5.2 — Implicit bias
This one matters more than the others, because a large industry rests on it.
Implicit bias is the idea that people hold associations they are not aware of and would deny, which shape their behaviour towards others.
The main instrument for measuring it asks people to sort words and images into categories as fast as they can, and treats the difference in reaction time between pairings as evidence of an association.
Why it matters here: the existence of automatic associations is not seriously disputed. Three further claims are, and they are the ones the industry rests on: that the test measures an individual’s bias reliably, that the measure predicts how a person will actually behave, and that training reduces it.
Those three have been examined and they do not fare equally.
An unusual dispute, in that several of the sharpest critics are people who think discrimination is a serious problem.
Automatic associations are real and measurable, they differ systematically between groups of respondents, and it would be strange if mental content that reliably shows up on a timed task had no effect on behaviour. Discrimination in the modern world is rarely announced, so a measure of what people will not say is exactly what is needed.
Analyses pooling hundreds of studies find that scores on the test correlate weakly with actual behaviour, weakly enough to be useless for predicting any individual. The same person retested gets noticeably different scores. And a large pooled analysis found that interventions which successfully shifted the measure produced no corresponding change in behaviour — which is the finding that matters, because changing behaviour was the entire point.
Automatic associations exist. Whether this instrument measures an individual’s are doubtful. Whether it predicts what they will do is weakly supported at best. Whether training changes conduct is not supported. Those are four separate claims and public discussion runs them together.
What would settle it: studies measuring actual behaviour rather than the test, before and after, over time. Some exist and they are not encouraging.
Why people care so much: because it is a rare thing that is simultaneously a research question, a very large budget, and a public ritual. Part One, Chapter Six applies: when a countable question has an industry attached, the counting gets harder rather than easier.
5.3 — Hiring discrimination
The claim: that women face bias in hiring and evaluation, demonstrable by sending identical applications under different names.
An audit or correspondence study sends identical applications, differing only in a name that signals sex or ethnicity, and records how each is treated.
It is the cleanest test of discrimination available, because the applications are genuinely identical — there is nothing else that could produce a difference in response.
Why it matters here: this is one of the few methods in the whole field with the power to settle a question, and when applied at different levels of seniority it has produced answers pointing in opposite directions.
This method is powerful and it has produced results in both directions, which is the honest finding.
One influential study sent identical applications for a junior laboratory position to science faculty and found the male-named applicant rated more competent and offered a higher salary. Another, examining hiring for permanent academic posts at national scale, found the opposite — a substantial preference for the female-named candidate across most fields.
These are not contradictory in the way they are usually presented. They examined different levels, in different decades, with different designs. Taken together they suggest something more interesting than either: that the direction and size of bias varies by field, by seniority and by period, and that confident general statements in either direction are not supported.
Verdict: contested. Discrimination in hiring is real, it has been demonstrated, and the claim that it uniformly disadvantages women in all fields at all levels today is not established. Anybody quoting one of those two studies without the other is selecting.
5.4 — How common is it?
One further contested area, and it is a measurement problem rather than a dispute about reality.
Estimates of how many women experience sexual assault vary between surveys by something close to an order of magnitude. This is not because one set of researchers is dishonest. It is because the surveys ask different questions, over different periods, with different definitions of what counts, and small changes in wording produce very large changes in the number.
The honest position is that the true figure is not known within a narrow band, that the low estimates are almost certainly too low because of under-reporting, and that the high estimates depend on definitions many respondents would not themselves apply to their experience.
Verdict: contested, and usually quoted without the range. A figure that varies tenfold by method is a figure that should always arrive with its method attached, and it almost never does — by either side, since opponents quote the low estimate with equal confidence.
Stereotype threat — that mentioning a stereotype before a test depresses performance — became one of the most cited findings in social psychology and has serious replication problems, particularly in large samples of children. Contested, with recent weight against the strong version.
Implicit bias involves four separate claims that get run together. That automatic associations exist — not disputed. That the standard test measures an individual’s reliably — doubtful. That it predicts behaviour — weakly supported at best. That training changes conduct — not supported, and a large pooled analysis found that interventions which shifted the measure produced no change in behaviour.
Hiring discrimination has been demonstrated in both directions. One study found a male-named applicant preferred for a junior lab post; another, at national scale for permanent academic posts, found a substantial preference the other way. Different levels, decades and designs — which suggests direction and size vary by field and period rather than that one study is wrong.
And prevalence estimates for sexual assault vary by close to a factor of ten depending on wording and definition. That is a figure which should always arrive with its method attached, and it almost never does — from either side.
6What Failed
Four claims that were made publicly, used to argue for policy, specific enough to be checked, and checked. They did not survive. And the people who overturned them were mostly not the movement’s opponents.
6.1 — “The same work”
The claim, as it appeared on posters, in speeches and in legislative campaigns for decades: that women are paid a fraction of what men are paid for the same work.
The raw figure was real. Comparing all working women with all working men, a large gap exists in every country.
The framing was not. Once you compare people doing the same job, with the same hours, the same seniority and the same continuity of employment, the gap shrinks to a small fraction of the headline number. What produces the headline figure is that men and women are distributed differently across occupations, hours and career continuity — and, as Chapter Four established, overwhelmingly that the interruption of a first child is absorbed by one parent.
Take the headline gap and pull it apart, in the order the pieces come off.
A large slice is hours — men work more paid hours on average, and part-time work pays less per hour.
A large slice is occupation — men and women are distributed differently across fields, and the fields pay differently. Part Four found the largest measured psychological sex difference is precisely an occupational preference.
A large slice is continuity — time out of employment costs earnings permanently, and it is overwhelmingly women who take it.
What is left after all of that is small, and it is not zero.
Now notice what has happened to the claim. “Women are paid less for the same work” is false as stated. “Women earn substantially less over a lifetime, and the mechanism is that society allocates the cost of children to one parent” is true, larger, harder to fix, and was not what the posters said.
The campaign chose the version that was easier to explain and could be shown false. The version that was true was more damning.
That decomposition is not a rhetorical move and it is worth knowing how it is done, along with what it quietly hides.
Pulling a pay gap apart is done with a standard technique that has been applied to this question in dozens of countries for fifty years.
What it does: take the raw difference in average earnings, then ask how much of it is accounted for by measured differences — hours worked, occupation, industry, years of experience, breaks in employment. Whatever is left over after all of those is the part that measured factors cannot explain.
What it shows: the explained portion is large. The unexplained residue is much smaller than the headline figure, and it shrinks further as the measurement of experience and continuity improves.
What it cannot show: that the residue is discrimination, or that the explained part is innocent. If women are steered into lower-paying fields, that steering is a real phenomenon which this method files under "occupation" and thereby removes from view. The technique measures what is accounted for, not what is justified — and both camps read it as though it measured the second.
Verdict: failed as stated. The underlying inequality is real and Chapter Four scored it as supported. The mechanism was misdescribed for decades, and the misdescription handed opponents a permanent and legitimate objection.
6.2 — That the differences are constructed
The claim: that psychological differences between men and women are produced by socialisation, and will diminish as societies become more equal.
This was not a fringe position. It was the working assumption of a great deal of writing and teaching for several decades, and it generated a specific, checkable prediction.
Parts Four and Five checked it. Several differences are real and substantial — the Things–People interest difference is the largest measured psychological sex difference there is. And the prediction runs the wrong way: in richer and more gender-equal countries the differences are generally larger, not smaller.
Part Five spent sixty pages establishing how contested that finding is. What is not contested is the narrower point that kills this claim: the differences did not disappear as barriers came down. Whatever explains the pattern, the prediction failed.
Verdict: failed. The strong version — all differences constructed, all will vanish — is dead. Weaker versions survive: that many differences are smaller than assumed, that some are constructed, and that the causes are unresolved. Those are much more modest claims than the one that was made.
6.3 — Patriarchy as a general explanation
The claim: that the distribution of advantage and disadvantage in society is organised by male dominance, and that this explains the pattern of outcomes generally.
As an account of specific domains it does real work. Property law, franchise, personal law, the composition of governments and boards, the marital rape exemption — all of these fit, and Chapter Four scored several as supported.
As a general theory of who does well and who does badly, it has an obvious problem: the bottom of society is overwhelmingly male.
Prison populations. Homelessness. Deaths at work. Deaths by suicide. Deaths of despair. Failure to complete school. Disconnection from employment and from family. Part Four found that a moderate average difference in aggression becomes a nine-in-ten outcome at the extreme, and that the same arithmetic putting more men at the top puts more men at the bottom. Part Fifteen counts all of it.
A theory saying society is organised for the benefit of men has to explain why nearly everybody at the very bottom of it is one. The available answers — that these men are casualties of a system that still benefits men overall, or that the metric is wrong — are not absurd, and they are additions to the theory rather than predictions from it.
Verdict: failed as a general explanation; supported in specific domains. That distinction is doing a lot of work and it is the honest position. A theory that explains every outcome forbids none, and Part One noted that a claim which cannot be wrong is not a claim.
6.4 — That removing barriers produces proportional representation
The claim: that the absence of women from particular fields is caused by barriers, and that removing the barriers will produce something close to proportional presence.
Part Five is entirely about this. The finding that retires it is narrow and survives every criticism made of the wider literature: there is no simple positive relationship between a country’s gender equality and women’s share of technical fields, and in the raw counts it runs the wrong way.
Chapter Four of this part complicates it usefully, and I am not going to smooth that over. The Indian randomised evidence shows a representation policy that measurably changed both decisions and girls’ aspirations. So “representation efforts never work” is also false.
Verdict: failed as stated, in a specific way. Barriers were real and removing them produced enormous change — Chapter Four scored that as supported. What failed is the further assumption that removal would continue producing change all the way to proportionality. That assumption underlies a great deal of policy and it is not supported.
6.5 — And one that was simply not true
A short entry, included because it illustrates something about how claims travel.
A widely repeated assertion held that reports of domestic violence spike dramatically on the day of a major American sporting event. It was announced at a press conference, reported everywhere, and repeated for years.
Journalists went to look for the underlying data. It was not there. The researchers whose work had been cited said their findings did not show it.
Verdict: failed, and it was never supported at any point.
Two things are worth taking from it. The claim was corrected — by reporters, within a couple of years, in public. And it is still repeated, decades later, which tells you that correction and circulation operate on different timescales. Part One, Chapter One measured that asymmetry in a different context and it applies identically here.
Everybody assumes that a movement’s failed claims are exposed by its opponents.
Both camps need this. The critic tells a story in which brave outsiders punctured an ideology. The defender tells a story in which hostile actors attacked good work. Both are describing an assault from outside.
Look at who actually did the work in this chapter.
The child penalty finding, which overturned the “same work” framing, came from economists whose research programme is gender inequality. The replication failures in stereotype threat came from psychologists, several of whom had built careers on studying disadvantage. The most damaging analyses of implicit bias measures include work by researchers who think discrimination is a serious problem and wanted a better instrument. The false claim in §6.5 was corrected by journalists within two years.
This was self-correction, and it is what a functioning field looks like. A movement that produced no falsified claims would not be one that was always right; it would be one that had never said anything checkable.
The general form: crediting a field’s self-correction to its critics. It flatters the critics, insults the correctors, and makes the whole process look like a scandal instead of the thing that is supposed to happen.
Which is worth holding through Chapter Nine, where the failures in this chapter get weighed against the record in Chapter Four.
“The same work” failed as stated. Pull the headline gap apart and hours, occupation and continuity take most of it, with the child penalty underneath. The true version — that a lifetime gap exists and society allocates the cost of children to one parent — is larger, harder to fix, and more damning. The campaign chose the version that was easier to explain and could be shown false.
That all differences are constructed and would vanish, failed. They did not disappear as barriers came down, and the largest measured psychological sex difference is an occupational preference. Weaker versions survive; the strong one is dead.
Patriarchy as a general explanation failed; it holds in specific domains. A theory saying society is organised for men’s benefit has to explain why nearly everybody at the very bottom — prison, homelessness, workplace death, suicide — is one.
That removing barriers produces proportional representation failed, in a specific way: barriers were real and removing them produced enormous change, and the assumption that removal continues all the way to proportionality is not supported.
And nearly all of this was done by people inside the tradition — economists studying gender inequality, psychologists studying disadvantage, researchers who wanted a better instrument. That is self-correction, and it is what a functioning field looks like. A movement producing no falsified claims would not be one that was always right. It would be one that had never said anything checkable.
7The Predictions
A claim about the world is one thing. A forecast is better, because it can be checked against what arrived. Here are five that were made in terms specific enough to be wrong — and one result nobody forecast at all.
7.1 — Five forecasts
Part One set the standard: a prediction that cannot say what would count as it being wrong is not a prediction, it is a mood. By that test, the movement made several genuine predictions, which is more than most political movements manage.
One: legal equality would produce equal participation. Remove the bars and women will enter in proportion.
Two: women entering paid work would produce shared domestic work. If she earns, the household tasks redistribute.
Three: the earnings gap would close.
Four: educational parity would produce earnings parity. The gap was attributed substantially to women’s lower qualifications, so equalising qualifications should equalise pay.
Five: women in positions of power would change what institutions do.
7.2 — What arrived
On the first: partly, then a stall. Entry into education, medicine, law and public administration was enormous and rapid. Entry into technical and manual fields was much smaller and has been flat or reversing for decades. Part Five is about why, and it does not resolve it. Half right.
On the second: partly. Men’s share of domestic and care work rose substantially through the later twentieth century. It did not reach parity, and the convergence slowed considerably. The phrase for what happened — a woman finishing paid work and beginning a second unpaid one — entered the language because it described something real. Half right, and the half that failed is the one Chapter Four scored as the mechanism behind the earnings gap.
On the third: partly, then a stall. The gap narrowed sharply for about two decades and then largely stopped narrowing in many countries. Half right.
On the fourth: wrong. Women overtook men in educational attainment across most of the developed world — Part Four found the reversal is now large and rarely discussed. The earnings gap did not close correspondingly. This is the cleanest failed prediction in the list, and it failed because the mechanism had been misidentified: the gap was never primarily about qualifications, it was about the allocation of children. Wrong.
On the fifth: right, with the strongest evidence in this part behind it. Chapter Four’s randomised Indian evidence establishes that women in council leadership changed spending, and that visible female authority closed the gap in girls’ educational attainment. Right.
7.3 — And one nobody forecast
Across roughly thirty-five years — a period covering the greatest expansion of women’s legal rights, education, earnings and employment in history — researchers tracked how women in the United States rated their own happiness.
It fell. Absolutely, and relative to men, whose ratings held steadier.
The same pattern was found in most of the developed countries where comparable data existed.
Nobody predicted this. Not the movement, which expected the opposite, and not its opponents, who mostly expected social collapse rather than a quiet decline in self-reported wellbeing among the beneficiaries.
And nobody has explained it. Proposed accounts include rising expectations against which reality is measured, the second shift in \u00a77.2, comparison across gender lines rather than within it — which Part Five identified as a real mechanism — and the possibility that self-reported happiness scales do not mean the same thing across decades. None is established.
Part Eight examines it properly. It is included here because a movement’s forecasting record should include the thing it did not see coming, and this is the largest.
I want to be careful about what this finding does and does not license, because it is quoted constantly by people who have not read past the headline.
It does not establish that the changes made women worse off. Reported happiness is one measure among many, and the same period saw enormous gains in life expectancy, education, earnings, physical safety within marriage and control over reproduction — every one of which most people would take over a point on a happiness scale.
What it establishes is narrower and still important: that a movement’s beneficiaries reporting lower wellbeing than before is a real result which nobody predicted and nobody can account for. An honest movement has to hold that rather than explain it away, and an honest opponent has to notice it does not say what they want it to.
Five predictions specific enough to be wrong. Legal equality producing equal participation — half right, enormous in the professions, flat or reversing in technical fields. Paid work producing shared domestic work — half right, men’s share rose substantially and convergence stalled short of parity. The earnings gap closing — half right, narrowed sharply for two decades then stopped.
Educational parity producing earnings parity — wrong. Women overtook men in education and the gap did not close, because the mechanism had been misidentified: it was never about qualifications, it was about who absorbs children.
Women in power changing institutions — right, with the randomised Indian evidence from Chapter Four behind it.
And one nobody forecast. Across thirty-five years of the greatest expansion of women’s rights in history, women’s self-reported happiness fell — absolutely and relative to men — in most developed countries measured. Neither camp predicted it and nobody has explained it.
It does not establish that the changes made women worse off; the same period brought enormous gains in safety, health, earnings and control over reproduction. It establishes that a real result exists which nobody saw coming and nobody can account for — and that both camps have to hold it rather than use it.
8The Fork
Part Four established that given a real difference in what people want, equal opportunity and equal outcomes are in conflict. A movement built on one of those has spent forty years being measured by the other, and largely without noticing.
8.1 — Two goals that came apart
The argument that won the legal battles was an argument about freedom.
A woman should be able to own what she earns, to enter a profession, to vote, to leave a marriage, to choose a husband, to control whether she becomes a mother. Every one of those is a claim about a person directing her own life, and every one is a moral claim of the kind Chapter Three said no measurement can touch. They were argued that way and they won that way.
The way institutions now measure whether that argument succeeded is outcomes. Percentage of board seats. Share of graduates in each field. Representation in a legislature. Pipeline metrics. Proportions.
Those are different things and Part Four, Chapter Nine established that they are not merely different — given a real difference in preferences, they are in conflict. Perfect freedom does not produce proportional outcomes. Producing proportional outcomes requires acting against what some people choose.
8.2 — Why the drift happened
It is worth understanding how a movement founded on one goal came to be measured by another, because nobody decided it.
Freedom is not countable. There is no statistic for how free a woman is to direct her life, and an institution cannot put it in an annual report or a policy target.
Outcomes are countable. A percentage of board seats is a number that can be reported, compared, targeted and improved.
So when the argument moved from courtrooms and parliaments into organisations, it acquired the metric that organisations can use. Not because anybody preferred it. Because it was the one that fits on a form.
The live dispute inside the movement, usually conducted without either side stating which they are defending.
The moral case was always about a person’s right to direct her own life, and that case is complete when the barriers are gone. Pursuing proportional outcomes past that point means overriding what women actually choose, which contradicts the founding argument. A movement that started by insisting women can decide for themselves cannot end by deciding for them.
Freedom on paper is not freedom. Preferences form inside a society that spent centuries teaching girls what to want, so a choice made under those conditions is not a clean signal. And outcomes are the only way to detect barriers that nobody admits to — if the field is genuinely open, the numbers should move, and when they do not, something is still operating.
These are not two routes to one destination. They are two destinations, and past a certain point pursuing one means giving up the other. Both are defensible. Neither is the feminist position, because Chapter One showed there is no such thing.
What would settle it: nothing empirical. This is a choice between values and it has to be made rather than discovered.
Why people care so much: because the outcomes goal is the one that generates budgets, targets and jobs, and the freedom goal is the one that generates moral authority. Most institutions want both and say the second while measuring the first.
But there is a prior question that neither side of that argument raises, and answering it changes what the fork is a fork in.
The entire dispute above assumes that equality is the goal — that the question is how much of it and of what kind.
Chapter One showed that is one faction’s framing. The difference strand never wanted equality in the sense of sameness; it wanted the revaluation of work and qualities associated with women, which is a different objective and is not achieved by putting women into men’s roles. The socialist strand did not want women’s equal access to a labour market it regarded as the problem.
So the fork in this chapter — freedom or outcomes — is a fork inside liberal feminism, presented as the central question of feminism as a whole. Both camps in the public argument accept that framing, which means both are arguing on the territory of one strand while believing they are arguing about the movement.
Notice what disappears when they do. The claim that raising children and running a household is valuable work that a society has chosen not to pay for is not a claim about equality at all. It is not answered by getting more women into engineering, and it is not answered by leaving them alone either. It simply drops out of a debate organised around proportions.
The general form: adopting one faction’s framing as the description of the whole dispute. The faction that supplies the vocabulary wins before anybody speaks.
8.3 — What the drift cost
Three things, and they are the reason this chapter matters more than it looks.
It made the movement vulnerable to Part Five. A programme measured by proportions is embarrassed by a finding that proportions do not respond to freedom. A programme measured by freedom would not be.
It put the movement into conflict with some women’s stated preferences, and gave it only one available reply — that the preferences were installed. That reply is not absurd and Part Five, Chapter Eight showed why it cannot be established either.
And it obscured the finding in Chapter Four. The child penalty is the largest thing in this part. It is not a representation problem and no proportion measures it. A movement organised around counting seats will systematically under-address a mechanism that operates through the structure of work and care — which is where nearly all of the remaining inequality actually is.
The argument that won the legal battles was about freedom — to own, to earn, to vote, to leave, to choose. Every one a moral claim of the kind no measurement touches.
The way institutions measure success is outcomes — board seats, graduate shares, pipeline metrics. Part Four established these are not merely different but in conflict: perfect freedom does not produce proportional outcomes, and producing them requires acting against what some people choose.
Nobody decided the drift. Freedom is not countable and outcomes are, so when the argument moved into organisations it acquired the metric that fits on a form.
And the fork is inside liberal feminism, presented as the central question of the whole. The difference strand wanted revaluation of women’s work, not sameness. The socialist strand did not want equal access to a labour market it thought was the problem. Both drop out of a debate organised around proportions.
The drift cost three things. It made the movement vulnerable to Part Five, which embarrasses a proportions programme and would not touch a freedom one. It put the movement against some women’s stated preferences with only one available reply. And it obscured the child penalty — the largest thing in this part, which no proportion measures and which is where nearly all the remaining inequality actually is.
9Scoring It
Eight chapters of claims, sorted. Now what the sorting actually establishes — which is less than either camp will want, and includes an admission about how these two parts were built.
9.1 — The ledger
Set the three categories side by side.
Supported: that women were excluded by statute from property, inheritance, education, the professions and the franchise. That removing those barriers produced enormous change. That women perform several times more unpaid work, measured by governments. That the child penalty is the dominant mechanism of the modern earnings gap, established by comparing women with their own earlier earnings. That violence within households was massively undercounted. And that female political representation changes both spending and girls’ aspirations — established by a randomised experiment across Indian villages.
Contested: stereotype threat, with recent weight against the strong version. Implicit bias measures, where the existence of associations is not in doubt and the predictive claim is weak. Hiring discrimination, which has been demonstrated in both directions at different levels and periods. Prevalence figures for sexual assault, which vary by close to a factor of ten with method.
Failed: the pay gap as described on the posters. That psychological differences are constructed and would vanish. Patriarchy as a general explanation of who ends up where. That removing barriers produces proportional representation. And at least one widely circulated statistic that was never supported at all.
9.2 — What the failures establish
Specific things were wrong. That is the whole of it, and it needs saying because two larger conclusions get drawn and neither follows.
The failures do not touch the moral claims. Chapter Three separated them for this reason. That a woman is entitled to own what she earns, to leave a marriage, to vote and to choose a husband is not a finding and cannot be falsified by one. Every legal change the movement won rested on that kind of argument. You could delete Chapters Four to Seven entirely and the case for a married woman’s right to her own property would be exactly where it was.
And the failures were mostly produced from inside. Chapter Six’s hidden assumption box set this out and it bears repeating in the scoring: the child penalty finding came from economists studying gender inequality, the replication failures came from psychologists studying disadvantage, and the sharpest critiques of implicit bias measures include work by people who wanted a better instrument for detecting discrimination.
A movement whose claims get corrected by researchers working inside its own tradition is behaving like a field. The alternative — no falsified claims at all — would not indicate a movement that was always right.
9.3 — The result that mirrors Part Six
Now the thing that took two parts to see, and it is the most interesting finding in either.
Part Six built the strongest case for the old rules and found that every argument in it was about household stability, partnership, male behaviour or women’s economic security — and none of them reached the rules. The best traditionalist arguments pointed away from the thing they were supposed to defend.
This part has produced the same shape.
The strongest finding here is the child penalty. It accounts for most of the remaining earnings gap, it is established by the best design available, and it is a mechanism operating through the structure of work and care.
And no representation metric measures it. Not board seats. Not graduate shares. Not pipeline percentages. A woman’s earnings collapsing permanently after a first birth does not appear in any of the numbers the movement’s institutions report against, because it is not a proportion.
So the same thing has happened twice. In Part Six, the best arguments for the traditional position argued against the rules it defends. In Part Seven, the best evidence for the feminist position points away from the metric its institutions use.
Each side’s strongest evidence is inconvenient to its own programme. That is a strange result and I did not expect it, and it is the most useful thing these two parts produced.
9.4 — And how I built these two parts
A reader who has just finished both parts has encountered one written by an advocate and one written by an examiner, and this part is the examiner.
Part Six argued a position at full strength for seven chapters with no undercutting. This part sorted claims into piles. Those are not the same treatment, and I made the choice deliberately.
My reason: the traditional case had no lawyer available. Nobody had assembled it from the best evidence, so somebody had to. Feminism has thousands of advocates, in print, better placed than me. What it did not have was somebody going through the claims one at a time with a scorecard.
Why that reason is not sufficient. It is an argument about what was missing from the literature. It is not an argument about what a reader experiences. Somebody reading these in sequence meets a warm part and a cold one, and will attribute the difference to the subject matter rather than to my construction — because that is what tone does, and it does it below the level at which arguments operate.
The general form: a defensible reason for a choice that still produces an effect the reason does not justify. Every writer has this available. The honest move is not to claim the reason cancels the effect. It is to say what I did and let you discount it.
9.5 — What survives
Set the failures against the record and the honest summary is short.
A movement that arose independently in several places at once, including in India six months before it arose in America. That won a set of legal changes resting on moral arguments no evidence can touch. That made a series of empirical claims in support, several of which were wrong, most of which were corrected by its own researchers. That has one finding — the allocation of children — larger and better established than anything its opponents have produced, and which its own institutions do not measure. And that has a genuine unresolved fork at its centre between two goals it has never chosen between in public.
That is a mixed record. It is also a considerably better record than most political movements can show, because most political movements do not make claims specific enough to be scored at all.
Supported: exclusion by statute; the enormous effect of removing barriers; several times more unpaid work, measured by governments; the child penalty; massively undercounted household violence; and female political representation changing spending and girls’ attainment, by randomised evidence in India.
Contested: stereotype threat; implicit bias measures; hiring discrimination, demonstrated in both directions; and prevalence figures varying tenfold with method.
Failed: the pay gap as described; that differences are constructed and would vanish; patriarchy as a general explanation; that removing barriers produces proportionality.
The failures touch none of the moral claims — the case for a married woman’s right to her own property is exactly where it was — and most of them were produced by researchers inside the tradition. That is a field correcting itself.
And the same shape appeared twice. In Part Six, the best arguments for the old rules argued against the rules. Here, the best evidence for the feminist position — the child penalty — points away from the metric its institutions use, because a woman’s earnings collapsing after a first birth is not a proportion and no representation target measures it.
Each side’s strongest evidence is inconvenient to its own programme. I did not expect that, and it is the most useful thing these two parts produced.
10An Honest List Of What We Do Not Know
Two lists, no hedging in either. What is genuinely unknown about the claims scored here, and what is solid enough that anybody arguing about feminism has to concede it.
10.1 — Genuinely unknown
The unknowns in this part cluster around causes rather than measurements, which is the usual pattern and is worse here than most.
Why the child penalty falls on mothers. The size is measured precisely. The cause is not. Biology, employer expectations, partner behaviour, policy and preference all remain live, and countries with very different parental leave arrangements still show a penalty — which weakens the pure-policy explanation without establishing anything else.
Why women’s reported happiness declined. The finding is real and replicated across countries. Every proposed explanation is a hypothesis. Part Eight examines them and does not resolve it.
Whether the contested claims in Chapter Five are true. Contested means contested. Stereotype threat may exist in specific conditions. Implicit associations may affect behaviour in ways the current instruments cannot detect. Saying a claim is unsupported is not saying it is false.
What the true prevalence of sexual violence is. Not known within a narrow band, for reasons of definition and method rather than dishonesty. Both camps quote the estimate that suits them.
Whether the Indian panchayat findings generalise. They are strong and they are about village councils in one country over a specific period. Whether the mechanism transfers to national politics, to corporate boards, or to other countries is untested.
And what women want, which this series has now failed to establish four separate times. Part Five, Chapter Eight explained why: there is no condition free of social influence in which a preference could be read cleanly. This is not a gap that better research fills.
10.2 — Solid
That Indian feminism has its own history, beginning no later than 1848. A school in Pune six months before Seneca Falls, a Marathi text in 1882, a feminist utopia in 1905. Dates, publications and institutions, all documented.
That the movements arose independently rather than spreading from one source. The problems addressed were different — caste, child marriage, widowhood, purdah — and no imported framework generates them.
That women were excluded by statute. Not by custom alone. The statutes are published and Part Three found one still in force.
That women perform several times more unpaid work. Measured by national governments in their own surveys.
That the child penalty is real and dominant. Established with the strongest design available, replicated across countries.
That the “same work” framing was false as stated. This is solid and it is a failure. Both halves belong on this list.
That female political representation changed outcomes in India, by randomised assignment. Very little in this series is supported this well.
That women’s reported happiness declined across the period of greatest gains. The measurement is solid. The interpretation is not, and the finding does not say what its usual quoters want.
And that equal freedom and equal outcomes are different goals that conflict. Established in Part Four, not disputed by anybody who has looked at it, and almost never stated by institutions pursuing one while claiming the other.
10.3 — What a movement is for
One closing observation, following the pattern of the previous parts.
This part scored a political movement as though it were a research programme. Chapter One’s hidden assumption box flagged that as a category error and I did it anyway, for a reason worth stating at the end rather than the beginning.
Movements are not in the business of being right. They are in the business of changing things, and the arguments that change things are selected for their power to move people rather than for their accuracy. That is not a criticism of feminism. It is a description of politics, and it applies identically to the nationalism in Part Three, to the reform campaigns that abolished sati, and to whatever movement the reader happens to belong to.
Which produces the honest closing position. The score in this part tells you about the claims. It does not tell you whether to support the movement, because movements are not judged on their claims and never have been.
What the score is good for is narrower and still worth the eight chapters: knowing which sentence you are allowed to say in an argument. A person who cannot tell you which claims failed does not have a view about feminism. They have a side.
Genuinely unknown: why the child penalty falls on mothers — the size is measured precisely and the cause is not. Why women’s reported happiness declined. Whether the contested claims are true, since contested is not false. The true prevalence of sexual violence. Whether the Indian panchayat findings generalise. And what women want, which this series has now failed to establish four separate times for a reason no research fixes.
Solid: Indian feminism’s own history from 1848 and its independent arising. Exclusion by statute. Several times more unpaid work. The child penalty. That the “same work” framing was false as stated. That female representation changed outcomes in India by randomised assignment. That women’s reported happiness declined across the period of greatest gains. And that equal freedom and equal outcomes conflict.
And a closing admission. This part scored a political movement as though it were a research programme, which is a category error I flagged in Chapter One and committed anyway. Movements are not in the business of being right; they are in the business of changing things, and the arguments that change things are selected for their power to move people. That applies to feminism, to the nationalism in Part Three, and to whatever movement you belong to.
So the score tells you about the claims and not about whether to support the movement. What it is good for is knowing which sentence you are entitled to say — and a person who cannot tell you which claims failed does not have a view. They have a side.
The Claims, Scored
Chapter Nine as a single sheet. Every claim examined in this part, with its verdict and the reason. This is the page to carry into an argument.
| Claim | Verdict | Why |
|---|---|---|
| Women were excluded by statute from property, education, professions, franchise | Supported | The statutes are published. Not disputed by anybody serious. |
| Removing the barriers produced enormous change | Supported | From a small minority to a majority of graduates within a century. |
| Women perform far more unpaid work | Supported | National time use surveys. In India, roughly five hours a day against an hour and a half. |
| The earnings gap is driven by the arrival of children | Supported | Whole-population records, comparing each woman with her own earlier path. Around four fifths of the remaining gap in the cleanest study. |
| Household and sexual violence were massively undercounted | Supported | Survey prevalence against recorded crime — two independent measures differing by a large factor. |
| Female political representation changes outcomes | Supported | Randomised assignment of village council leadership in India. Spending changed; the gap in girls’ educational attainment closed. |
| Stereotype threat explains performance gaps | Contested | Replication failures, particularly in large child samples, and signs of publication bias. |
| Implicit bias tests predict discriminatory behaviour | Contested | Weak correlation with behaviour; shifting the measure does not shift conduct. The existence of associations is not in doubt. |
| Women face hiring discrimination | Contested | Demonstrated in both directions at different levels and periods. General claims either way are unsupported. |
| Prevalence of sexual assault | Contested | Estimates vary by close to a factor of ten with definition and method. |
| Women are paid less for the same work | Failed as stated | Hours, occupation and continuity take most of the headline gap. The true mechanism is larger and more damning. |
| Psychological differences are constructed and will vanish | Failed | They did not vanish as barriers fell, and the largest measured psychological sex difference is an occupational preference. |
| Patriarchy explains the distribution of outcomes generally | Failed as general theory | The bottom of society — prison, homelessness, workplace death, suicide — is overwhelmingly male. Holds in specific domains. |
| Removing barriers produces proportional representation | Failed | No simple positive relationship between national gender equality and women’s share of technical fields; in raw counts it runs the wrong way. |
| Educational parity would produce earnings parity | Failed prediction | Women overtook men in education and the gap did not close, because the mechanism was misidentified. |
Read the verdict column and note the proportions. Six supported, four contested, five failed. That is a mixed record and a better one than most political movements can produce, because most do not make claims specific enough to score.
And note the two entries doing the most work. The child penalty is the strongest supported claim and no representation metric measures it. “The same work” is the clearest failure and the true version underneath it is more damning than the slogan was.
Sources & further reading — Part 7
An Indian Timeline
Because the claim that this is a foreign import is answered better by dates than by argument.
| When | What happened |
|---|---|
| Jan 1848 | Savitribai and Jyotirao Phule open a school for girls in Pune. She teaches in it, and is abused in the street on the way. |
| Jul 1848 | The Seneca Falls Convention, New York. Six months later. |
| 1882 | Tarabai Shinde publishes Stri Purush Tulana in Marathi, arguing that men are morally worse than women. Frequently called the first modern Indian feminist text. |
| 1887 | Pandita Ramabai, a Sanskrit scholar honoured by the pandits of Calcutta, publishes The High-Caste Hindu Woman. |
| 1905 | Rokeya Sakhawat Hossain publishes Sultana’s Dream — a utopia in which women govern and men are kept in seclusion. Written in English, by a Bengali Muslim woman, as a satire on purdah. |
| 1917–1927 | Indian women’s organisations form: a women’s association, a national council, an all-India conference. All predating independence, all arguing about Indian law. |
| 1951 | Ambedkar resigns as Law Minister, with the stalling of the Hindu Code Bill among his reasons. It passes in pieces over the following five years. |
| 1974 | A government committee reports on the status of women, expecting to document progress. It finds that on the sex ratio, women’s paid work and political participation, things have got worse since independence. The report is credited with restarting the movement. |
| 1979–1983 | A Supreme Court acquittal in a custodial rape case produces an open letter from four law professors, nationwide protest, and a change to the criminal law. |
| 1993 | A constitutional amendment reserves a third of village council leaderships for women, with villages effectively randomly assigned — accidentally creating one of the largest randomised experiments in political science. |
| 2000s | Analysis of that experiment finds women leaders spend differently, and that after two rounds the gender gap in adolescent educational attainment closes. |
| 2012–13 | After a case in Delhi produces sustained national protest, a committee headed by a former Chief Justice reports in under a month. Legislation the following year broadens the definition of rape and creates offences for stalking, voyeurism and acid attacks. The committee’s recommendation to remove the marital rape exception is not adopted. |
| 2023 | India replaces the colonial penal code. The marital rape exception is carried across, as Part Three recorded. |
| ↓ | Part Eight starts here — and counts what sixty years of the equality project actually produced, measured rather than asserted. |
Glossary
Every hard word used in this part, in plain English. Each was explained where it first appeared.
| Term | Plain meaning |
|---|---|
| Child penalty | The permanent fall in a woman’s earnings following the birth of a first child, measured against her own previous earnings path. The dominant mechanism of the modern earnings gap, and something no representation metric captures. |
| Difference feminism | The strand holding that qualities and work associated with women are undervalued, and that the remedy is to revalue them rather than to make women more like men. |
| Empirical claim | A statement about how the world is, which can be checked. Distinguished from a moral claim, which says what is owed, and a strategic claim, which says what should be done. |
| Implicit bias | Associations a person is unaware of and would deny. Their existence is not disputed; whether the standard test measures an individual’s, predicts behaviour, or responds to training are three further claims, and they fare differently. |
| Intersectional critique | That “women” is not one category, because sex interacts with caste, class, religion and region. In India this appears as the Dalit feminist argument that the mainstream movement was substantially upper-caste. |
| Liberal feminism | The strand holding that the problem is exclusion and the remedy is equal access. Produced nearly all of the legal change of the last two centuries. |
| Radical feminism | The strand holding that male dominance is the organising structure of society rather than a removable defect, reaching into family and sexuality. Source of the phrase about the personal being political. |
| Second shift | The unpaid domestic work a person begins after finishing paid work. Named because men’s share of domestic labour rose substantially and stopped short of parity. |
| Socialist feminism | The strand holding that women’s uncounted domestic labour underwrites the economy, and that a woman succeeding within that economy has not solved the problem. |
| Stereotype threat | The proposed effect by which reminding people of a negative stereotype before a test depresses their performance. Heavily cited, and with serious replication problems. |