Before We Begin — Where We Left Off
Part Thirteen went looking for the experiments history ran by accident. It set out six tests and four tiers of evidence, and then applied them honestly, which had the effect of demoting almost every case that gets quoted in public.
Two findings came out of it that this part is built on.
The first is the strongest thing in the series. India’s 1993 constitutional amendment reserved a third of village council leaderships for women and rotated them by village serial number — a lottery nobody intended. Villages that got a woman in the chair saw bias fall, saw more women win seats afterwards in elections that were not reserved, and saw the gap between what parents wanted for their sons and for their daughters narrow by about a quarter within seven years. No law changed for those girls. No one lectured their fathers. A woman was in the chair at the village meeting, and their fathers saw her there.
The second came from the failures. Four Indian reforms — the marriage-age act of 1929, dowry prohibition, the ban on sex determination, and protective labour laws — all produced the opposite of what they intended, and all four share one signature. Each prohibited a visible symptom while leaving the underlying demand completely untouched, and the behaviour re-emerged somewhere less visible and less regulated.
Put those two together and you have a diagnostic and its opposite. Change a structure and beliefs follow within a decade. Ban an output and it moves.
This part takes those two findings seriously and asks what follows from them. It is the least descriptive part of the series and the most exposed: everything before this could hide behind “here is what the evidence says”, and a design chapter has to commit.
I should say at the outset what it is not. It is not a manifesto and it does not tell anybody what a society should want. Chapter Eight is entirely about the questions in this subject that are not design problems and cannot be solved by being cleverer about them — and refusing to treat a value conflict as an engineering task is, I think, the most useful thing in the part.
How this document is built
Six kinds of box, each doing one job.
The first explains a hard word the moment it appears.
Intervention: anything deliberately done to change an outcome — a law, a payment, a building, a campaign, a quota, a school.
Why the word matters here: the whole argument of this part is that interventions come in a small number of shapes, that the shape predicts whether it works far better than the intention does, and that the shape most societies reach for first is the one with the worst record.
The second takes a number too big or abstract to picture and gives it a body.
“Seventy-nine per cent of the funds went on media advocacy.” That is a line in a committee report and it slides past.
Turn it round: for every hundred rupees released to states under India’s flagship scheme for the girl child, about seventy-nine bought advertisements about girls and about twenty-one reached anything else. Chapter Two is about why that is not a scandal of corruption. It is a scandal of design.
The third shows the actual evidence and then says what it cannot show.
This part reuses the evidence tiers from Part Thirteen, because a design argument is only as good as the findings underneath it.
Tier one is the randomised Indian reservation study. Tier two is the quasi-random cases — Bangladeshi factory placement, divided Germany. Tier three is sharp before-and-after in one place. Tier four is a comparison somebody noticed.
Where a recommendation in this part rests on tier one or two, I say so. Where it rests on tier three or four, I say that too — and there is more of the second than I would like.
The fourth is for places where informed people genuinely disagree.
A question, then the sides, then a verdict that does not claim more certainty than exists.
Its strongest case, put the way its best advocate would put it, with the evidence it actually has.
The same, with equal care. If a position sounds foolish here, I have failed to understand it, not proved it wrong.
What would settle it: the evidence that would decide it. Sometimes the honest answer is that nothing available would.
The fifth digs out what both sides are assuming without noticing.
Five in this part. The last one is turned on this series, and given what this part concludes about arguments, it is the one I have been walking towards since Part One.
The sixth closes every chapter in the plainest words available.
Six boxes. Word explains. In Real Terms converts. How We Actually Know This shows the evidence. The Argument gives every side its best shot. The Hidden Assumption digs underneath. Remember This closes the chapter.
Simple words, serious content, nothing left out.
Two warnings specific to this part
The first is that design writing is where honest research usually goes to die. Thirteen parts of carefully graded evidence can be spent very quickly on a list of confident recommendations, and the recommendations always sound more certain than the findings they came from. I have tried to hold the line by attaching an evidence tier to each proposal and by putting the trade-offs that cannot be designed away into their own chapter rather than burying them. Where I am extrapolating past what the evidence supports, I say so in the sentence rather than in a footnote.
The second is about who this is addressed to. Part Twelve found that the three sensible answers to a question about online harm were addressed to three completely different people — a parent on a Tuesday, a state over years, a community with no timeline — and that the argument was incoherent because everybody pretended they were advice to the same person. That finding governs this whole part. Chapter Nine is an explicit map of who holds which lever, and several of the most effective things in this document are not available to any reader of it. Saying so is not defeatism. It is the difference between a design and a wish.
1The Design Problem
Before proposing anything you have to say what working would look like, and who would have to do it. Both questions have answers that most of this argument never reaches, because it is conducted between people who agree that the unit under discussion is a rule.
1.1 — What “work” would mean
Thirteen parts have produced a large number of findings and no criterion. So here is the criterion, stated before any proposal, so that it cannot be adjusted afterwards to fit whatever I end up recommending.
An intervention works if it changes an outcome somebody actually cares about, without moving the cost onto someone who did not choose it, and without the behaviour reappearing somewhere worse.
Each clause is doing work and each corresponds to a failure this series has documented.
The first clause rules out interventions that change a number without changing a life — Part Thirteen’s warning about scoreboards, in operational form.
The second rules out the entire category identified in Part Twelve: solutions whose whole cost lands on a girl, which are cheap for everyone else precisely because of that.
The third rules out the four backfires in Part Thirteen, where the behaviour was banned and simply relocated.
Incidence: an idea from public finance meaning who actually ends up bearing a cost, as distinct from who is nominally charged.
A tax on a company may be paid by the company and borne by its customers. A rule aimed at protecting a daughter may be written by a legislature, applied by a family and borne entirely by her.
Why it matters: almost every argument in this series is conducted about the nominal target of a rule. Almost none of it is conducted about the incidence, which is where the interesting question lives.
1.2 — Removing a rule is not the same as replacing it
Here is the second half of the framing, and it is what separates this part from most reform writing.
Every rule described across this series is doing a job for somebody. Not a good job, not a just job, but a job — solving some problem that would still exist if the rule vanished tonight. Part Thirteen’s four backfires are all the same event: a rule was removed or forbidden, the problem it had been solving did not go anywhere, and something worse grew in the gap.
That gives a design instruction that is uncomfortable for reformers and does not comfort traditionalists either. You cannot evaluate the removal of a rule without knowing what it was doing. And you cannot know that from the rule’s stated purpose, because stated purposes are unreliable — Part Twelve found that the rule’s declared beneficiary is very often not the one it is organised around.
Take dowry, which Part Thirteen showed grew after being prohibited.
What was it doing? Among other things it was moving property to a daughter at the one moment her family transferred anything to her at all — in a system where she would inherit little or nothing otherwise. It was also pricing a groom in a marriage market where families compete.
Prohibition removed none of that. Daughters still did not inherit. Grooms were still priced. So the transfer continued, in cash, undocumented, with every check on it gone.
The reform that would have worked was not a ban on the payment. It was making daughters inherit — which is a change to what the rule was for, rather than to what the rule says.
1.3 — Intentions predict nothing
One more piece of the frame, and it is the one that makes this part possible.
Go back through the failures. Every one was designed by people who wanted what this series wants. The 1929 marriage act was passed by reformers. Dowry prohibition was passed by reformers. The ban on sex determination was passed by reformers. Protective labour laws were written by people trying to protect women, in response to genuine hazards.
And the successes were frequently not designed at all. The Indian reservation lottery was an administrative rotation rule. The Bangladeshi factories were built for export earnings by people with no interest in girls’ schooling. Iran’s education expansion was produced by a government that would have been horrified by the demographic result.
That pattern is not a coincidence and it is not an argument against trying. It is evidence that the shape of an intervention predicts its result far better than the intention behind it does — which is good news, because shape is something you can inspect in advance and intention is not.
Before going further it is worth taking seriously the position that the whole enterprise of this part is a mistake.
The record in Part Thirteen is four serious failures against a handful of accidental successes, and the successes were accidents. That is what you would expect if these arrangements are far too complex to steer — evolved solutions to problems nobody has fully articulated, where every deliberate change produces consequences the designer did not foresee. The honest lesson is humility: describe, understand, and stop reaching for the lever. Most of the improvement in women’s lives over two centuries came from economic growth and technology rather than from anybody’s programme.
The status quo is not the absence of intervention. Part Thirteen established this about West Germany: a tax rule, a school timetable and a legal provision together constituted a vigorous programme to keep married women at home, and it read as the natural state of things precisely because nobody announced it. Declining to intervene means leaving in place arrangements that were built, are maintained, and have an incidence. And the successes were not accidents in the relevant sense — the reservation lottery was designed by somebody, even if the attitude effects were not foreseen. What was accidental was the discovery, not the mechanism.
What would settle it: a systematic accounting of interventions in this area scored by shape and by outcome. Part Thirteen noted that nobody has assembled one, and it remains the single most useful piece of unfunded research in this whole subject.
Why people care so much: because “you cannot design a society” is a genuine intellectual position and also the most convenient possible thing for anybody who benefits from the current arrangement to believe.
Both sides of that exchange are arguing about the same object, and it is worth asking whether it is the right object.
Watch how this argument is conducted, anywhere, by anyone. One side wants a rule gone. The other wants it kept. A third wants it modified. Every participant is treating the rule as the thing under discussion — the object to be attacked, defended or amended.
But a rule is an answer. Somewhere behind it is a question, and the question is the durable thing. Dowry answers “how does property reach a daughter and how are grooms priced”. Seclusion answers “how is a woman kept safe in a place with no policing”. Early marriage answers “how does a family secure a match before the market moves against it”. Son preference answers “who will support us when we are old”.
Notice what follows. If you win the argument about the rule and the question is still there, you have not won anything — you have created a vacancy, and something will fill it. That is precisely what Part Thirteen documented four times over. And it explains something otherwise puzzling: why these arrangements are so resilient in the face of law, education and prosperity. They are not being held in place by stubbornness. They are being held in place by unsolved problems.
The general form: arguing about an answer without identifying the question. It runs through every reform debate. Abolishing a practice without asking what it was doing. Closing an institution without asking what it was absorbing. Banning a substance without asking what people were treating with it.
Why this one is load-bearing for the rest of this part: it converts the whole exercise from a moral argument into an engineering one, and it does so without conceding anything moral. You can believe a rule is monstrous and accept that it is solving a real problem. In fact those two beliefs together are the only position from which anything ever gets fixed, because the first supplies the motive and the second supplies the specification.
With that established, the rest of this part is an attempt to write the specification.
An intervention works if it changes an outcome somebody cares about, without moving the cost onto someone who did not choose it, and without the behaviour reappearing somewhere worse. Each clause corresponds to a documented failure.
Incidence — who actually bears a cost, rather than who is nominally charged — is where the interesting question lives, and almost none of this argument is conducted about it.
Every rule in this series is doing a job. Remove it without replacing the function and something worse grows in the gap. That is what all four of Part Thirteen’s backfires were.
Intentions predict nothing. Every documented failure was designed by reformers; several of the successes were accidents produced by people with entirely different aims. The shape of an intervention predicts its result far better than the intention behind it, which is good news, because shape can be inspected in advance.
And the rule is not the unit. A rule is an answer to a question, and the question is the durable thing — which is why these arrangements survive law, education and prosperity. They are not held in place by stubbornness. They are held in place by unsolved problems.
2Structure Beats Argument
The central finding of this part, and the most expensive one to accept — because it disqualifies almost everything currently being done, almost everything both sides are arguing about, and this document.
2.1 — The finding, stated plainly
Beliefs about women follow arrangements rather than leading them. Part Thirteen established this from three directions and one of them was randomised.
In Indian villages, a rotation rule put a woman in the council chair and, within seven years, bias fell, more women won unreserved seats, and the gap between what parents wanted for sons and daughters narrowed by about a quarter. Nobody argued with those parents. Something was rearranged and their views moved.
In Bangladesh, garment factories opened where the roads were. Girls nearby stayed in school longer than girls in a scheme that paid their families to keep them there. Nobody persuaded anyone; a wage made visible what school was for.
In East Germany, forty years of arrangements produced attitudes that outlived the state, the ideology and the childcare system by another thirty years.
Three cases, three countries, three mechanisms, one direction. Change what people do and what they think follows. The reverse — change what they think and hope the doing follows — has a much worse record, and India has run an unusually clean test of it.
2.2 — Seventy-nine per cent on advertisements
India launched a flagship national programme in 2015 aimed at the falling child sex ratio and at girls’ education. It is one of the most recognised government campaigns in the country.
In December 2021 a parliamentary committee on the empowerment of women tabled its report on the scheme. Of ₹446.72 crore released to states between 2016 and 2019, it found that 78.91 per cent was spent on media advocacy. Across a slightly different window, a government answer in the upper house put media spending at ₹401.04 crore out of ₹683.05 crore — about 58 per cent.
The committee also found that of the funds actually released to states across the scheme’s first six years, only about a quarter had been spent by them at all.
What this evidence is good at: it is the government’s own accounting, examined by a committee of its own legislature, chaired by a member of the governing party. Nobody hostile produced these numbers.
What it cannot establish: that the scheme achieved nothing. India’s sex ratio at birth has improved over the same period, and attributing that improvement between the campaign, enforcement of the older law on sex determination, falling fertility and rising incomes is genuinely hard. The committee’s point — and mine — is not that the outcome failed. It is where the money went.
To see why that matters, put the spending next to what Part Thirteen established about the practice it was aimed at.
Part Thirteen established what actually produces missing girls: small desired family size, plus son preference, plus access to a scan. It also established that the practice was strongest among richer, more educated, urban families — the ones most exposed to national media and most likely to have already seen every advertisement.
So the money was spent telling people that daughters are valuable, and the people doing the selecting were disproportionately the ones who had heard that message most often, in the best Hindi, on the largest screens.
The campaign was aimed at a shortage of belief. The problem was not a shortage of belief.
2.3 — Why persuasion is reached for anyway
It would be easy and wrong to treat this as waste or cynicism. There are four reasons a ministry reaches for advertising, and only one of them is bad.
It is the only lever some actors hold. A national ministry cannot build a village school, hire a doctor or change what a household does at dinner. Those belong to states, districts and families. What a ministry with a budget and no operational reach can do is buy media. Advertising is what an institution does when it has money and no lever — and Chapter Nine is entirely about that mismatch.
It is measurable immediately. Reach, impressions, recall. A campaign can be evaluated within a quarter. A school built this year shows up in a sex ratio in twenty.
It is politically safe. Nobody’s interests are threatened by an advertisement. A change to inheritance law threatens property; a change to school hours threatens employers; a quota threatens incumbents. A message threatens nobody, which is precisely why it changes nothing.
And it is genuinely popular. People like these campaigns. They feel like a country deciding to be better.
Behaviour change communication: the technical name for campaigns that try to shift what people do by changing what they believe or what they think others believe.
It is a real field with real successes — handwashing, oral rehydration, seatbelts, vaccination, smoking. Those successes are not in doubt and they have saved enormous numbers of lives.
Why it matters that the successes look alike: every one of them changed a cheap behaviour that nobody had an interest in defending. No family’s property, lineage or standing depended on not washing their hands. Chapter Two’s argument is about what happens when the behaviour is load-bearing.
That distinction is where the real disagreement sits, and both sides of it have serious evidence.
The people who run norm-change campaigns are not fools and their field has genuine victories. The disagreement is about which category this problem is in.
Social norm change is one of the best-evidenced tools in development. Campaigns have moved handwashing, sanitation, vaccination and smoking at national scale, and several have moved practices with deep social roots — attitudes to open defecation shifted in India within a decade after generations of failure. What matters is design: campaigns work when they target what people believe others think, when they are delivered by trusted local figures rather than by celebrities, and when they are paired with something concrete. The Indian campaign discussed above did the celebrity version without the concrete pairing. That is an argument about execution, not about the category.
Line up the victories. Handwashing, seatbelts, sanitation, vaccination: in every case the behaviour was cheap, the beneficiary was the person changing, and nobody lost anything when it changed. Now line up the norms in this series. Son preference is load-bearing for old-age security and property. Dowry is load-bearing for a marriage market. Seclusion is load-bearing for a family’s standing. Each has a constituency that loses when it goes. A message cannot outbid an interest, and no amount of design makes it able to.
What would settle it: a campaign of comparable scale and quality run against a load-bearing norm, with a proper comparison group. Several sanitation programmes come close and their results are contested. Nothing of that quality has been run on son preference or on dowry.
Why people care so much: because campaigns are the one intervention that a national government, an international agency and a well-meaning individual can all fund, announce and feel good about. Withdrawing that leaves most of the people currently working on this with nothing to do that is within their reach.
Underneath that whole exchange is a premise nobody in it states, and it is the one this chapter has been circling.
Look at what everybody in this argument is actually doing. Traditionalists write, preach and post, trying to convince people that the old arrangements were right. Reformers write, campaign and post, trying to convince people they were wrong. Governments buy advertisements. Both sides believe the other side’s beliefs are the problem and that changing them is the work.
The evidence says beliefs are downstream. They followed a rotation rule in West Bengal. They followed a factory in Bangladesh. They followed a childcare system in East Germany for thirty years after it closed. In none of those cases did anybody argue anyone into anything.
Which means the entire visible surface of this argument — the essays, the campaigns, the speeches, the ferocious online exchanges — is being conducted on the layer that responds least, by people who have concluded that the other side simply needs to hear a better case.
The general form: intervening where the argument is rather than where the causation is. It appears everywhere the visible conflict and the operative mechanism sit on different layers. Debating a national mood rather than the housing market that produced it. Arguing about a workforce’s attitude to safety rather than the shift pattern that produces the accidents.
And this box has an obvious application that I am going to state now rather than hide until the end. This document is an argument. Fourteen parts of it. If the finding in this chapter is right, then the form I have chosen is the least effective one available, and I have spent a great deal of care on the layer the evidence says moves last. Chapter Nine is where I take that seriously rather than acknowledging it and carrying on.
None of which means the work is worthless. It means the layer has to be chosen deliberately rather than by default, and the next chapter is about what the choices actually are.
Beliefs follow arrangements. A rotation rule in West Bengal, a factory in Bangladesh, a childcare system in East Germany — three countries, three mechanisms, one direction, and nobody argued anyone into anything.
India’s flagship scheme for the girl child spent 78.91 per cent of the funds released to states over three years on media advocacy, by its own parliament’s accounting — while the practice it targeted was strongest among the families most exposed to that media.
Ministries reach for advertising for good reasons, not corrupt ones: it is often the only lever they hold, it is measurable within a quarter, it threatens nobody’s interests, and people like it. The third reason is why it does not work.
Persuasion has real victories — handwashing, seatbelts, sanitation — and they share a feature: the behaviour was cheap and nobody lost anything when it changed. Every norm in this series is load-bearing for property, lineage, security or standing.
Which means both sides of this argument, and this document, are working on the layer that responds least.
3The Four Shapes
Every intervention in this series is one of four shapes, and the shape predicts the outcome. Two of them have almost no successes. One of them has almost no failures. Nearly all the money goes to the first two.
3.1 — The taxonomy
Strip the intentions away and sort by mechanism, and everything in fourteen parts falls into four kinds.
Shape one: ban the output. Identify the visible behaviour and forbid it. Dowry prohibition. The ban on sex determination. Raising the marriage age. Forbidding night work. Village councils banning girls’ phones.
Shape two: change the price. Leave the decision where it is and alter what it costs. Cash for a third child. Conditional payments to keep girls in school. Tax relief for mothers. Subsidised transport for working women.
Shape three: change who is in the room. Put a person where people can see them, repeatedly, doing a thing they were believed not to do. Reserved council seats. Women in a workplace, a courtroom, a police station.
Shape four: supply the substitute. Work out what the rule was doing and provide that function another way. A pension instead of a son. Inheritance instead of dowry. Childcare instead of a wife at home. Policing and transport instead of seclusion.
And the shape from Chapter Two, which is really a category error rather than a fifth option: tell people to stop.
3.2 — The scoring
| Intervention | Shape | Result, and evidence tier |
|---|---|---|
| Reserved council seats, India 1993 | Who is in the room | Bias fell, aspirations narrowed, more women elected after. Tier one — randomised. |
| Garment factories, Bangladesh | Substitute, by accident | Later marriage, later childbirth, longer schooling — beating a cash scheme. Tier two. |
| East German childcare | Substitute | Near-universal employment; attitudes outlived the institutions by thirty years. Tier two. |
| Saudi driving and travel rules | Ban removed | Participation roughly doubled in under a decade. Tier three — bundled with inducements. |
| Conditional payments for girls’ schooling | Price | Real but smaller than a nearby factory. Tier two. |
| Pronatalist cash, several countries | Price | A few hundredths to two tenths of a child for enormous sums. Tier three. |
| Dowry prohibition, 1961 | Ban the output | Practice spread and grew; payment moved into undocumented cash. Tier three. |
| Sex determination ban, 1994 | Ban the output | Ratio kept falling for years; enforcement rare. Tier three. |
| Marriage age act, 1929 | Ban the output | Documented surge of child marriages before commencement. Tier three. |
| Protective labour laws | Ban the output | Removed the women rather than the danger. Tier three. |
| National campaign for the girl child | Tell people to stop | 79% of released funds on media; effect on the target practice unestablished. Tier four. |
The pattern is not subtle. Every entry in the “ban the output” row failed or backfired. The one unambiguous tier-one success is “who is in the room”. The other successes are substitutes, one of which was built by accident by an export industry.
The strongest test of shape three outside village politics is Norway, which from 2003 required large listed companies to have at least 40 per cent female board representation, with full compliance by 2008.
What it achieved: the target, comprehensively. Board representation went from under a tenth to the mandated level within five years, which settles the question of whether quotas fill seats.
What it did not achieve: much movement in the layers beneath. Studies looking for spillover into senior management and the wider pipeline have generally found modest effects, and a small number of women ended up holding several directorships each.
Why this matters for the taxonomy rather than against it: a board is invisible. Nobody in Norway watches a directors’ meeting. If the active ingredient is repeated public observation, as the Indian case suggests, then a board quota supplies the seat without supplying the mechanism — and the modest result is what the theory predicts rather than a refutation of it.
3.3 — Why the shapes behave this way
There is a single explanation and it comes from Chapter One.
Shapes one and five leave the underlying demand completely untouched. A ban tells people they may not do the thing; the reason they were doing it is unchanged, so the behaviour relocates. Exhortation does not even do that much.
Shape two touches the margin. A payment changes the calculation without changing who is doing the calculating or what their alternatives are — which is why its effects are real, small, and stop when the money stops.
Shapes three and four change the structure itself. A woman in the chair alters what people have observed. A pension alters why you needed a son. Neither requires anyone to agree with anything, and neither has to be maintained by persuasion once it is in place.
Take a family deciding whether to educate a daughter, and run the four shapes at them.
Ban: it is now illegal to keep her home. They keep her home; nobody comes.
Price: they are paid to send her. She goes while the payment lasts, and it buys attendance rather than schooling — she is in the room and nobody expects anything of her there.
Who is in the room: the woman running the village council is educated and everyone can see what that got her. The father revises what school is for.
Substitute: a factory forty minutes away hires literate young women at a wage the family can see arriving in a neighbour’s house every month.
The last two do not need the father to agree with anything. They change what agreeing would even mean.
One of those two shapes has a name that everybody argues about, and the arguing is mostly about the wrong thing.
Quota or reservation: a rule setting aside a share of positions for a defined group.
Why it belongs in shape three rather than shape one: a quota is usually attacked as a restriction and defended as fairness, and both descriptions miss what the evidence says it does. In the Indian case the measurable effect was not on the women who held the seats. It was on everybody who watched them hold the seats — their neighbours’ bias, their neighbours’ daughters’ school attendance, their neighbours’ expectations. The seat was the delivery mechanism. Exposure was the intervention.
Which raises the question that decides how far any of this can be taken.
This matters because it is the shape with the best evidence, and the temptation is to apply it everywhere immediately.
What produced the effect was repeated public observation of a woman doing something believed to be beyond women. That is not specific to councils. It is the same mechanism people invoke about the first woman in any profession, and it predicts that the effect should appear wherever visibility and repetition are present. The Indian evidence is tier one, it comes from one of the least promising settings anybody could have chosen, and it worked within seven years. Demanding a separate randomised trial for every institution before acting is a standard nobody applies to interventions they already like.
Corporate board quotas are the best-studied comparison and their results are modest: they reliably increase women on boards and have done much less to change the pipeline beneath, with a small number of women taking multiple seats. That is exactly what you would expect if the effective ingredient were visibility rather than the seat — and almost nobody watches a board. A village leader is seen by every household every week; a director is seen by an annual report. Applying the panchayat finding to boardrooms is applying a mechanism to a setting that lacks the mechanism.
What would settle it: a randomised or quasi-random quota in a high-visibility, non-political institution, with attitudes measured in the surrounding population. Health and education systems in several countries allocate posts in ways that come close to this and have never been analysed for it.
Why people care so much: because quotas are the most politically contested intervention in this entire subject, and both the case for them and the case against them are usually made about fairness to the individuals appointed — which the evidence suggests is not where the effect is.
Shapes three and four both depend on knowing what a rule was doing in the first place. That is the next chapter.
Four shapes: ban the output, change the price, change who is in the room, supply the substitute. Plus telling people to stop, which is a category error rather than a fifth option.
Scored across fourteen parts, every “ban the output” case failed or backfired. Price changes are real, small and stop when the money stops. The one tier-one success is who is in the room, and the strongest accidental success was a substitute built by an export industry.
The explanation is one sentence: bans and exhortation leave the demand untouched, price changes touch the margin, and the other two change the structure so that nobody has to be persuaded of anything.
A quota is not usefully described as a restriction or as fairness. In the Indian case the effect was not on the women who held the seats but on everybody who watched them hold the seats.
Which makes the active ingredient visibility, not the position — so the shape predicts a large effect in village councils, schools, clinics, buses and shopfronts, and a small one anywhere the public never looks.
4Find the Job the Rule Is Doing
Chapter One said a rule is an answer to a question. This chapter names the questions. Four of them have obvious substitutes that have been partly tried, and one turns out not to be doing the job it claims to be doing at all.
4.1 — Dowry: property that arrives at a wedding
Ask what dowry is for and the usual answers are about greed or about tradition. Neither is a function. Two things are.
It moves property to a daughter at the one moment anything is transferred to her, in a system where she historically inherited little. And it prices a groom in a market where families compete for the better ones.
Prohibition touched neither, which is why Part Thirteen found the practice growing after 1961. The substitute is not a stricter ban. It is making daughters inherit — which India did in 2005, giving daughters equal rights in ancestral property.
The 2005 amendment has been studied, partly because several Indian states had already made similar changes earlier, which lets researchers compare states that reformed at different times.
What the work broadly finds: daughters in reformed settings became more likely actually to inherit land, and there is evidence of gains in their education. That is the shape the theory predicts.
What complicates it seriously: at least one careful study found that in some settings the reform was associated with worse survival outcomes for girls — on the reasoning that a daughter who will now take a share of the land is more expensive to have, in exactly the families that were already making that calculation. This finding is contested and the literature has not settled.
What that means for this chapter: even a correctly shaped reform, aimed squarely at the function rather than at the symptom, can produce a backfire — because supplying the substitute changes the price of a daughter as well as her position. Nothing here is safe, and anybody who tells you otherwise is selling something.
4.2 — Seclusion: safety in the absence of policing
Restricting where a woman may go is the oldest rule in this series. Two jobs sit under it. It manages a genuine physical risk in places with no reliable policing, no lighting and no safe transport. And it protects the family’s standing, by controlling what can be said about her.
The first of those has an obvious substitute and it is not an argument. It is infrastructure.
Research on women students in Delhi found something precise enough to build policy on. Faced with a choice between a better college and a safer route, many chose the safer route — accepting a substantially lower-quality institution to avoid a journey they perceived as dangerous.
Put that in a life. A young woman’s career ceiling is being set, before she sits a single examination, by the lighting and crowding on a particular stretch of road between her house and a campus.
That is not a norm about women. It is a street. And a street can be changed by a municipality without anybody’s beliefs moving at all.
This is shape four from the last chapter in its purest form. Lighting, buses, staffed stations and enforcement against harassment do not require a single family to revise its view of women. They remove the reason the restriction was rational.
4.3 — Early marriage and son preference
Early marriage is doing three jobs. It secures a match before the market moves against a family. It shortens the window in which a daughter’s reputation could be damaged — which, after Part Twelve, is a window that no longer closes. And it stops the cost of keeping her.
The substitute that has demonstrably worked is not a higher legal age. It is a visible return to keeping her: the Bangladeshi factories delayed marriages more effectively than a scheme that paid families directly, because a wage changes what the extra years are for.
Son preference is doing the most concrete job of all, and Part Nine named it: children are the retirement system. In India that is not a metaphor — a 2007 statute obliges children to maintain elderly parents, and roughly one person in five over sixty has any pension at all.
So the substitute is a pension, and the evidence for the mechanism runs the other way round in a way that should make everyone pause. Part Nine found that introducing public pensions is followed by measurable falls in fertility, because the reason for having many children weakens. The same lever that reduces the need for a son reduces the number of children — which is a benefit and a cost at once, and Chapter Eight is where that stops being a design problem.
Load-bearing: borrowed from building. A load-bearing wall is holding something up; remove it and the structure above comes down. A decorative wall can go without consequence.
Why the distinction organises this whole chapter: the rules in this series are not equally load-bearing. Some are holding up a real function — property transfer, old-age security, physical safety. Others are holding up nothing but themselves.
And you cannot tell which is which by looking at how fiercely they are defended. Fierceness tracks how central a rule feels, not what it is carrying.
4.4 — The rule that is not doing its stated job
Now the interesting case, because it is the one where the functional analysis produces a surprise.
Rules about women’s sexual conduct — the subject of Parts Ten and Twelve — have a standard functional explanation, offered by anthropologists and by traditionalists alike. They exist to secure paternity certainty. A man investing in children needs to know they are his, and in the absence of any way to establish that, the rules do the work.
That is a genuine function and for most of human history it was unsolvable. Then, in the 1990s, it was solved. Paternity testing is cheap, reliable and universally available.
And nothing happened. The rules did not weaken. No society revised its expectations of daughters because a laboratory could now settle the question in a week. In India, in Europe, anywhere — the norm continued exactly as before.
Which tells you something the functional analysis could not tell you any other way. If paternity certainty were the job, the invention that supplied it would have retired the rule. It did not, so the job is something else — and Parts Ten and Twelve identified what: a marriage market that prices a claim about her, and a household asset that her conduct can damage.
This is worth holding on to as a method. Supplying a substitute is a test as well as a remedy. If you provide the stated function and the rule does not move, the stated function was not the function.
Everything above assumes that solving the underlying problem makes the rule loosen. That assumption deserves examination, because there is evidence on both sides of it.
These arrangements survive because somebody needs them, not because people are stubborn. Remove the need and the defenders drift away, because defending a rule is costly and nobody pays a cost for nothing. This is why prosperity, pensions and policing have coincided with the loosening of exactly the rules whose functions they replaced, in country after country. It also explains the timing: the loosening does not arrive with the argument, it arrives a decade or two after the substitute does.
Part Thirteen measured this directly: East German attitudes persisted thirty years after the institutions that produced them ceased to exist. Norms are sticky in both directions. And they acquire new functions — a rule that began as paternity management becomes a marker of respectability, of caste, of piety, of belonging, and each of those is a fresh reason to keep it after the original one has gone. The paternity testing case is the proof: the function was supplied and the rule did not notice.
What would settle it: tracking a specific rule in a specific place across the decades after its function was substituted — pension coverage against son preference is the cleanest available pairing, and the data to attempt it exists in several countries.
Why people care so much: because the lag is exactly long enough that every reform can be declared a failure by its opponents and a success by its supporters, using the same evidence, for about twenty years.
All of which assumes somebody can act at all. The next chapter is about who can actually reach the place where these decisions are made.
Dowry moves property to a daughter in a system where she inherits little, and prices a groom. The substitute is inheritance, not a stricter ban — and even that produced a documented complication, because a daughter who will inherit is more expensive.
Seclusion manages real physical risk where there is no policing. Research on Delhi students found women accepting substantially worse colleges to avoid routes they perceived as dangerous. That is not a norm about women. It is a street, and a municipality can change it.
Early marriage is answered better by a visible return to keeping her than by a higher legal age. Son preference is answered by a pension — which also reduces fertility, so the lever cuts both ways.
Not all rules are load-bearing, and you cannot tell which are by how fiercely they are defended. Fierceness tracks how central a rule feels, not what it carries.
And the test case: paternity certainty was the stated function of the rules about women’s sexual conduct. It was solved in the 1990s, cheaply and universally, and nothing moved — which means the stated function was never the function.
5Who Can Reach Inside a Household
Part Eleven found that a rule lands where it can be enforced. The same is true of a reform, and almost everything in this series happens in a room the state cannot enter. Four things get in anyway, and India already operates the largest of them.
5.1 — The reach problem
Part Eleven asked why the rules in this series fall on daughters rather than sons and found an answer that was not about biology or belief. They fall where they can be enforced. A household can reach a daughter living in it. It cannot reach a son who has moved to another city, and it has never seriously tried.
Turn that finding round and it becomes a design constraint, and a brutal one.
A reform also lands where it can be enforced. And nearly everything documented across fourteen parts happens somewhere no state can go: a conversation at dinner, a decision about a scan, a phone handed over on demand, a marriage arranged between two families, a daughter told what she may study.
Passing a law about any of that is passing a law about a room nobody official will ever be in. That is why Part Thirteen’s four backfires all involved prohibitions on things that occur inside families, and why the successes involved schools, factories and council chambers.
5.2 — What a state actually controls
Be precise about it, because the list is shorter than the rhetoric on both sides suggests.
A state controls the institutions it runs — schools, clinics, buses, courts, police stations. It controls money it transfers and to whose name. It controls records: who is registered as born, married, employed, owning land. And it controls what it licenses and who it hires.
That is genuinely a lot. What it does not control is who decides anything in a household, what is said to a daughter, or what a family concludes about her.
Household bargaining power: the idea that a household is not one decision-maker with one set of wants, but several people with different wants, and that what the household does depends on their relative positions.
What determines a person’s position: what they contribute, what they could get by leaving, and what they control directly.
Why this is the central concept of the chapter: it explains how something outside a household can change what happens inside it without anybody entering. You do not need to reach the room. You need to change what one person in it could do if she left it.
5.3 — Four things that get in
A wage. An income arriving in her name is the single most studied instrument here. It changes what she contributes, and — more importantly — what she could survive on if she left. Both raise her position in every decision the household makes, including decisions that have nothing to do with money. Nobody has to enter the house for this to happen. Note that a wage does this and a payment to the household does not, or does much less, which is why the Bangladeshi factories beat a scheme that paid families.
A record. A registered marriage is checkable; an unregistered one is invisible, and an invisible marriage cannot be underage because nothing about it exists. This is why marriage-age laws without registration produce nothing — they forbid an event that leaves no trace. Registration is unglamorous, cheap, and the precondition for every other rule about marriage being anything but a sentence.
A school roll. Attendance is observable from outside, which makes it one of the very few things about a girl’s life that a state can actually see. And the school itself is a place where she is outside the household’s supervision for six hours a day, five days a week, next to other girls.
Other women, in her house, for another reason. This is the one that is under-used, and India already operates it at a scale no other country comes close to.
Since 2005 India has run a village health worker programme staffed by roughly a million women, each recruited from and resident in the village she serves. They are trained on maternal and child health, they are paid on activity rather than salary, and their job requires them to enter other people’s houses and talk to the women in them.
What is established: the programme has been credited with substantial contributions to institutional delivery, immunisation coverage and maternal health outcomes, and it received international recognition in 2022. Its health effects are the reason it exists and the reason it is documented.
What has never been studied, as far as I can find: its effect on anything in this series. Nobody has asked whether villages with a more active health worker show the exposure effects that Part Thirteen’s reserved seats produced — changed attitudes, changed aspirations for daughters, changed school attendance.
Why that is a striking gap. This is a million women, in a million villages, with a legitimate reason to be inside houses where a council leader never goes, doing a job that requires literacy and produces an income. On the mechanism identified in Chapter Three — visible, repeated observation of a woman doing a thing — it is the largest such intervention ever built anywhere, and it was built for something else.
Set that beside the one case in this series where the mechanism was actually measured.
A village council leader is seen at meetings. A health worker is seen in your kitchen, every few weeks, for years, weighing your baby.
The reservation study found that watching a woman run a council for two terms narrowed the gap between what parents wanted for their sons and their daughters by about a quarter.
Nobody has ever measured what watching a woman do a skilled job in your own house, repeatedly, over a decade, does to the same number. India has been running that experiment on a million villages for twenty years and has never looked.
Before recommending any of that, there is an objection worth taking seriously, and it is not a small one.
This chapter has been treating household reach as a design goal. That deserves a challenge, and there is a serious one.
Part Twelve established that a capability does not know what it is for. Build the ability to see and act inside families to protect daughters, and you have built the ability to see and act inside families. The same visits, records and registers serve a benign state and a hostile one, and the family is the one institution that has historically been able to resist both. Every authoritarian project of the last century began by getting between parents and children, usually for reasons that sounded exactly like the ones in this chapter.
Nearly every harm documented across fourteen parts happens inside a family: the marriage arranged, the scan taken, the phone confiscated, the schooling stopped, the beating. A remedy that stops at the front door is not a remedy. And the premise is false in any case — states already reach into households routinely and uncontroversially for vaccination, for school attendance, for taxation and for child protection. The question was settled long ago; what is being debated is only whether daughters are a good enough reason.
What would settle it: nothing evidential; this is partly a question about what a state should be. But the design implication is testable: indirect instruments should outperform supervisory ones, and where both have been tried, they do.
Why people care so much: because “the family is private” is simultaneously a genuine principle with a long and honourable history, and the sentence that has ended every attempt to do anything about what happens to women inside families.
So much for who can act. The next chapter costs the largest thing anyone has proposed that nobody has built.
Part Eleven found that rules land where they can be enforced. So do reforms — and almost everything in this series happens in a room the state cannot enter.
A state controls its institutions, its transfers, its records and its hiring. It does not control who decides anything in a household, or what is said to a daughter.
Four things get in without anybody crossing the threshold: a wage in her name, which changes what she could survive on if she left; a record, because an unregistered marriage cannot be underage since nothing about it exists; a school roll, which is one of the few things about her life the state can see; and other women, in her house, for another reason.
India runs the largest version of the fourth anywhere: about a million village health workers, resident women doing skilled work inside houses a council leader never enters. Nobody has ever measured what that does to anything in this series.
The distinction that resolves the objection: reaching in is not the same as entering. A wage, a register and a school roll change what happens inside a family by changing a person’s position — not by supervising her.
6What the Humane Road Would Cost
Part Nine found that the only road to a higher birth rate that costs women nothing has never actually been built. This chapter specifies it, prices it against what countries already spend, and identifies what is really blocking it — which is not the money.
6.1 — The specification
Part Nine set out three roads to raising a birth rate. Make children cheaper. Make the alternatives worse. Remove the exit. Only the first costs women nothing, and Chapter Ten of that part listed the serious version of it among the things nobody has tried.
Here is what “serious” would mean, assembled from what the research says actually predicts the gap between the children people say they want and the children they have.
One: childcare that matches a working day. Not a few hours, not from age three. Full-day, from infancy, universal, at a price a median household can pay.
Two: housing that people in their twenties can obtain. Delayed household formation is one of the strongest correlates of delayed childbearing, and in the countries with the lowest fertility it is also where the cost pressure is most extreme.
Three: employment protection that survives childbearing in practice. Not a statutory right on paper but a labour market where two years out does not permanently reset a woman’s earnings — which requires something about how employers are compensated for it, not merely what they are forbidden to do.
Four: men doing half the domestic work. Part Nine’s Chapter Four found the countries with the lowest fertility are those most equal in public life and least equal at home. This is the component that cannot be purchased.
Five: funded fertility treatment. The one component demonstrably present in the only rich democracy that never fell below replacement.
6.2 — What countries already spend
Family policy spending as a share of national income: the standard way of comparing what countries put into cash benefits, tax relief and services for families with children.
Rough landmarks from Part Nine: Hungary has run at around five per cent, among the highest anywhere. France has run around three and a half to four per cent for decades. The wealthy-country average sits closer to two.
Why the comparison matters here: the ceiling on what a determined country has been willing to spend is about five per cent of everything it produces, and we know what that bought, because Hungary spent it and ended roughly where it started.
The more important question is not how much was spent but what it was spent on, and the answer is consistent across every country that has tried.
The crucial fact about the money already spent is not how much. It is what shape it was.
Hungary’s five per cent went overwhelmingly into cash and loans — zero-interest borrowing forgiven as children arrive, tax exemptions, housing subsidies tied to family size. South Korea’s two hundred billion dollars went into grants, leave entitlements and childcare in a country where the binding constraint turned out to be that people were not marrying at all. The Nordic countries built the services but did so alongside some of the most severe housing pressure and job insecurity for young people in Europe.
So the honest statement is not that the humane road failed. It is that no country has run the five components together, sustained, for a generation. Every case is a partial version, and the partial versions differ from each other in which parts they left out.
What this cannot tell you: whether the full package would work. Nobody knows, and anyone claiming to is extrapolating from packages that were missing two or three of the five parts.
It helps to put the sums in a form that makes the scale visible.
Put the ceiling in Indian terms. Five per cent of national income is an enormous number — comparable to what India spends on defence and education combined, roughly speaking.
Now put it against what the same money buys elsewhere in this series. Part Nine calculated that pronatalist cash transfers, at their most effective, buy somewhere between a few hundredths and two tenths of a child per woman, at a cost per additional birth running into hundreds of thousands of pounds.
So the sums involved are not small and they are not unprecedented. They are being spent. They are being spent on the shape with the worst record.
Which leaves the question Part Nine could not answer and this chapter cannot close either.
This is the question Part Nine left open and this chapter cannot close. Both positions are held by serious demographers.
Nobody has run all five together for thirty years without a recession or a change of government. Each part has support on its own: childcare availability predicts return to work, housing costs predict delayed household formation, employment penalties predict the second birth, domestic sharing predicts the third, and treatment funding is present in the one rich country that never fell. Judging the road by Hungary’s cash or Korea’s grants is judging a treatment by trials that omitted most of the medicine. France’s decades near the top of Western Europe show what even a partial version buys.
The Nordic countries built the best services in existence and their fertility fell by half a child anyway. Part Four of Part Nine established that the gender-equity theory’s central prediction failed exactly where the mechanism was strongest. And in the lowest-fertility countries the decline is now largely a partnership decline — people are not pairing up, and every component in that list is aimed at parents, arriving after the point where the number is decided. You cannot subsidise your way to a relationship.
Which reframes the question, and this is the useful part. If the package is judged solely on the birth rate, it is a gamble with poor odds. But every one of the five components is a benefit to people who are already alive, regardless of what happens to fertility. Childcare helps the parents who have it. Housing helps the people who live in it. Job protection helps the women who keep their earnings. None of that becomes worthless if the birth rate does not move.
Why people care so much: because attaching these things to a demographic justification made them contingent on an uncertain payoff — and gave every opponent a way to argue against childcare by pointing at Finland’s birth rate.
And there is a premise underneath the whole affordability discussion that nobody on either side has examined.
Everybody discusses the humane road as an affordability question. Advocates say it would be expensive but worth it. Opponents say it is unaffordable. Governments say the fiscal space is not there. All of them are treating money as the binding constraint.
It is not, and the evidence is that the money has already been spent. Korea put over two hundred billion dollars into this. Hungary sustained five per cent of national income for fifteen years. Those are not countries that could not afford the humane road. They are countries that spent comparable sums on a different shape.
So look at what actually separates the two. Cash transfers, tax breaks and baby bonuses have one property in common: nobody has to give anything up. The money comes from general revenue, arrives at households, and is politically costless because it takes nothing from anyone who can object.
Now look at the five components. Childcare at that scale requires a workforce and a building programme. Housing for the under-thirties requires supply, which means somebody’s asset values. Employment protection that actually works requires employers to absorb something. Men doing half the domestic work requires men to do half the domestic work. Four of the five require an identifiable group to hand something over.
That is the difference, and it is not financial. Cash is chosen because it is distributionally invisible, not because it is cheap — Korea’s version was extraordinarily expensive.
The general form: naming a distributional obstacle as a financial one. It runs through every policy area where the affordable option is the one that asks nothing of anybody in particular. Transfers preferred to reform. Subsidies preferred to supply. Compensation preferred to change.
What follows is uncomfortable for the reformer side specifically. If money is not the obstacle, then “we cannot afford it” was never the argument to answer, and fifteen years of costing exercises were addressed to the wrong objection.
That is what can be built. The next chapter is about the things that are already cheap and are still not being done.
The humane road has five components: full-day childcare from infancy, housing people in their twenties can get, employment protection that survives childbearing in practice, men doing half the domestic work, and funded fertility treatment.
No country has run all five together for a generation. Hungary’s five per cent of national income went into cash and loans; Korea’s two hundred billion dollars went into a country where people had stopped marrying; the Nordics built the services alongside severe housing pressure.
Even its strongest advocates now expect a few tenths of a child, not a return to replacement — which reframes the question, because every component helps people who are already alive whether or not the birth rate moves.
Attaching childcare and housing to a demographic justification made them contingent on an uncertain payoff, and handed opponents a way to argue against childcare by pointing at Finland.
And the money is not the obstacle: it has already been spent, on cash, because cash is the only shape that requires nobody to hand anything over. Four of the five components need an identifiable group to give something up. That is a distributional obstacle wearing a financial costume.
7The Levers Nobody Pulls
Across fourteen parts the same kind of thing kept turning up: cheap, dull, high-yield, and undone. They share one property, and it is not cost.
7.1 — The list
Every part of this series ended with a chapter on what is not known, and a pattern accumulated across them that I did not notice until the thirteenth. Here it is collected.
| The lever | Where it came from | What is missing |
|---|---|---|
| Cervical screening at scale | Part Ten | About two in a hundred Indian women aged 30–49 have ever been screened, for a cancer killing on the order of seventy-five thousand a year that is detectable a decade early. |
| Two survey questions | Parts Twelve, Thirteen | Whether a woman has private use of a phone, and whether her messages or images have been shared without permission. Neither is asked anywhere at national scale. |
| Marriage registration | Chapter Five | The precondition for any marriage-age rule to be more than a sentence. An unregistered marriage cannot be underage, because nothing about it exists. |
| Measuring a law already passed | Part Twelve | India’s two-hour takedown obligation came into force in February 2026. No published figures on orders issued, categories used, or outcomes. |
| Analysing the health worker network | Chapter Five | A million village women doing skilled work inside houses for twenty years, never examined for the exposure effects the reservation study found. |
| Street lighting and transport | Chapter Four | Route safety is setting women’s educational ceilings before they sit an exam. Municipal, not national; cheap relative to a college. |
| Three analyses of existing data | Part Thirteen | The Iranian provincial comparison, the East German migration question, and a systematic accounting of reforms by shape. No new collection required for any of them. |
The screening figure is worth sourcing precisely, because it is the most alarming number in the table and it would be easy to overstate.
It comes from India’s own National Family Health Survey, which asks women aged 30 to 49 directly whether they have ever been screened for cervical cancer. Around two in a hundred said yes. That is a nationally representative household survey, not an administrative count, so it is not distorted by which facilities keep records.
What it establishes: the coverage gap is real and enormous, and it is not an artefact of reporting.
What it does not establish: that screening everybody is achievable at current capacity, or what the achievable coverage would be. A survey can tell you how few have been reached. It cannot tell you what it would take to reach them, and Chapter Seven’s argument box is about exactly that gap.
7.2 — What they have in common
Not cost. Several of them cost essentially nothing.
What they share is that none of them produces a moment. There is no ribbon to cut on a survey question. No minister stands beside a marriage register. An analysis of data collected twenty years ago cannot be announced, and a screening programme can be announced exactly once and must then be run, unglamorously, for two decades, by people whose names nobody will know.
Compare that with the shape Chapter Two examined. A campaign is announceable indefinitely. It has a launch, a face, a hashtag, a set of measurable outputs available within a quarter, and it threatens nobody. Every property that makes an intervention politically attractive is present in the thing that works least.
Take the screening number and do the arithmetic that nobody does.
Two in a hundred women in the relevant age group have ever been screened. The cancer in question kills roughly two hundred Indian women a day. It is caused by a virus, it is preventable by a vaccine India now manufactures, and it is visible as cell changes years before it becomes dangerous.
Part Ten established the other half: a woman’s risk depends on her husband’s history as much as her own, so married women with one lifetime partner die of it. No rule about her conduct reaches them. A test does.
This is the largest physical harm attached to sex in Indian women’s lives, it is the one nobody warns anybody about, and the intervention is a routine outpatient procedure.
7.3 — Why “just do it” is too easy
I want to take the obvious objection seriously, because a list like the one above is the kind of thing that reads well and dissolves on contact with anyone who has run anything.
State capacity: what a government can actually deliver, as distinct from what it can legislate or announce.
It is made of unglamorous things: trained staff, working equipment, functioning records, supervision, and the ability to keep doing something in year eight that was started in year one.
Why it matters to this chapter: a lever being cheap in money is not the same as it being cheap in capacity. Screening every woman in a country of 1.4 billion requires laboratories, technicians, referral pathways and follow-up. The test is routine. The system is not.
Which sets up the objection that a list like this deserves, and it is a serious one.
Anyone who has worked inside a health or statistical system will have an objection to the table above. It is a good objection and it is only partly right.
None of these is unknown. Every item has been proposed, often repeatedly, by people better informed than any writer. They do not happen because of capacity, not because of inattention. India’s screening gap is not a gap in awareness — it is a shortage of trained personnel, laboratories and referral pathways, and adding a question to a survey is trivial while acting on the answer is not. Lists like this treat delivery as a detail and thereby misdescribe the entire difficulty, which is why outsiders’ recommendations are so often received as insulting by the people doing the work.
The same state delivers vaccines to hundreds of millions of people, runs the largest election on Earth, and counts births and deaths in eight thousand sample areas twice over. The capacity plainly exists; what varies is what it is pointed at. And several items on that list do not need any capacity at all. Adding a question to a survey already in the field, publishing the counts for a law already in force, and analysing censuses collected decades ago require no clinics, no staff and no buildings. For those, the capacity argument is not an explanation. It is a change of subject.
What would settle it: for the free items, nothing — they are either done or not, and they are not. For the others, costed delivery plans, which have been produced in several states and not funded.
Why people care so much: because “we lack capacity” and “we lack will” are very hard to distinguish from outside, and every institution has an interest in the first description.
7.4 — The measurement point, made once
One item on that list deserves separating out, because it recurs in every part of this series and it is the cheapest thing in the whole document.
Part Ten found that India’s crime statistics cannot see online abuse of women, because there is no category for it. Part Twelve found no Indian survey asks whether a woman controls her own phone. Part Thirteen found three questions that could be answered from data already collected. Every one of those parts ended by noting that the state in question can count — it counts fraud to the rupee, it surveys hundreds of thousands of households, it has run the same demographic instrument for fifty years.
A statistical system measures what someone decided was a fact worth having. Across fourteen parts, the things that keep not being measured are the same things, and they concern the same people.
I am not going to claim that is deliberate, because I have no evidence that it is and the more likely explanation is that nobody in the room ever proposed it. But the effect is identical either way: a subject on which the evidence is permanently thin, in which every argument is therefore conducted by assertion, and in which the party with the loudest assertion wins by default.
Seven levers recur across this series and none of them is expensive: screening, two survey questions, marriage registration, measuring a law already in force, analysing a million-woman health network, street lighting and transport, and three analyses of data already collected.
What they share is not cost. It is that none of them produces a moment. No ribbon, no launch, no face, no quarterly metric — and a screening programme can be announced once and must then be run for twenty years by people nobody will name.
Every property that makes an intervention politically attractive is present in the shape that works least.
The capacity objection is real for some of these and is a change of subject for three of them — adding a question to a survey already in the field, publishing counts for a law already in force, and analysing decades-old censuses need no clinics, no staff and no buildings.
And across fourteen parts the things that keep not being measured are the same things, about the same people — which produces a subject where every argument is settled by assertion, and the loudest assertion wins by default.
8What Cannot Be Designed Away
Seven chapters of design. This one is about the questions that are not design questions — where no amount of cleverness produces an option that gives everybody what they want, and where calling it a problem is how a society avoids making a choice.
8.1 — Four things that will not dissolve
The temptation in a chapter like this series’ last few is to keep going until everything has a remedy. Four things resist, and they resist for different reasons.
Fertility against women’s options. Part Nine set out three roads and found that only one costs women nothing. Chapter Six of this part found that even its strongest advocates now expect a few tenths of a child from it. If a society genuinely requires replacement fertility, and the humane road delivers two tenths against a gap of half a child, then the remaining distance has to come from somewhere — and the only somewhere anybody has ever found is a reduction in what women can do instead. That is not a gap in the research. Fifty years and enormous sums have been spent looking for the third road.
The care hours. Part Nine’s Chapter Seven established that you cannot store labour: the care an eighty-year-old needs in 2055 must be delivered by somebody alive and working in 2055. Ageing raises the hours needed. Low fertility reduces the people available. Rising female employment — which every government including every worried one is actively pursuing — removes those same people during working hours. All three push the same way at once, and no financing arrangement changes the number of hours. It only changes who pays and who does them.
Safety against the capability that provides it. Part Twelve found that a two-hour takedown obligation and a two-hour removal capability are the same machine with the target class left as a parameter. You cannot have the protection without building the instrument. This is not a technical limitation to be engineered around; it is what the thing is.
A daughter’s interest against her family’s standing. Part Twelve’s Chapter Eight showed these coincide often enough that most families never have to notice. At the moment they diverge — and they do diverge — somebody loses, and no design makes both win.
A problem has a solution that has not been found yet. More thought, more money or more evidence gets you to it.
A trade-off has no such solution. The options available all involve giving something up, and choosing between them is a judgement about what matters, not a calculation about what works.
Why the distinction is the point of this chapter: the two require completely different responses. A problem calls for research. A trade-off calls for a decision — and treating the second as the first produces the appearance of progress indefinitely, because there is always another study to commission.
The most concrete of the four is worth working through in full, because it is the one where the arithmetic does the arguing.
Take the care hours, which is the most concrete of the four.
India’s old-age support ratio falls from roughly ten working-age people per person over sixty-five to about four by 2050. The people who will be sixty-five then are alive now and can be counted, so this is close to certain.
Somebody has to bathe them, feed them, lift them and sit with them. Those hours exist whether or not anyone budgets for them.
There are exactly three places they can come from: paid workers, who must be trained and paid; unpaid family members, who in every society on Earth are overwhelmingly women; or nobody, in which case the care does not happen.
Every plan currently in existence, in every country, quietly assumes the second — and none of them writes it down, because writing it down would require saying whose decade it is.
8.2 — Grading the four
Not every declared trade-off is real. Plenty of things called iron laws dissolved: that women could not work and raise children, that mass education would bankrupt states, that a large population could not be fed. Announcing a trade-off is a way of closing a search, and searches have a habit of succeeding after they are closed.
So the honest thing is to grade them by how hard anybody has actually looked.
Fertility against options has been searched hard, for fifty years, by wealthy governments with strong incentives and enormous budgets. The search has produced small effects. This one looks real, though Chapter Six’s caveat stands: the full package was never assembled.
The care hours are arithmetic. Automation may change how many hours a given amount of care takes, and that is worth pursuing — but it does not change the structure of the problem, only its size. This one is real.
Safety against capability is real as a technical matter and is not a trade-off about outcomes. What is undecided is who holds the switch, under what review, with what record. That is a political question with real answers, and Part Twelve’s point was that nobody is having it because one side is defending the machine and the other is objecting to it.
The daughter against the family’s standing is real at the moment of divergence and is much less common than the rhetoric suggests, because most of the divergence is manufactured by the penalty structure. Reduce the penalty and the interests coincide more often. So this one is partly designable and partly not.
There is a real dispute about whether a chapter like this helps or does damage.
A choice that is not made explicitly is still made — by default, invisibly, and usually in favour of whoever benefits from inaction. Every plan that assumes unpaid female labour will materialise is making a decision about women’s time without ever putting it on a page. Naming a trade-off is the only way to make the decision visible enough to argue about, and the alternative is not a better outcome, it is the same outcome with nobody accountable for it.
The history of this subject is a history of iron laws that turned out to be arrangements. People were certain that women could not do skilled work and raise children; that female education would destroy families; that low female employment reflected preference. Every one was declared a fixed constraint and every one dissolved. Declaring a trade-off tells everybody to stop looking, and the record says the looking is what produced the improvements.
What would settle it: for fertility, the full package from Chapter Six, run for thirty years. Nothing else will do it, and nobody is running it.
Why people care so much: because whoever gets to declare a trade-off real gets to say that somebody must lose, and then to nominate who.
Beneath that dispute is the assumption this chapter exists to name.
Every participant in this argument, including this document, has been proceeding as though enough analysis would eventually produce a design that gets everything: a society with women’s full options and replacing generations and care for its old and protection without surveillance. Fourteen parts have been structured around the search for it.
Some of that is not available. Not because the research is incomplete — because the things being asked for are in tension, and no arrangement of institutions makes two incompatible outcomes both occur.
Now notice what treating a trade-off as a problem actually does, because it is not neutral. It generates the appearance of movement without anybody losing anything. There is always another pilot, another study, another commission. Everyone can be seen to act, indefinitely, and no identifiable group ever has to hand anything over.
Which is exactly the function Chapter Six identified in cash transfers. Cash is chosen over childcare because it is distributionally invisible. Research is chosen over decision for the same reason. They are the same move at different points in the process, and they are both extremely expensive ways of not choosing.
The general form: treating a value conflict as an unsolved technical question. It appears wherever a society faces a choice it does not want to make. Commissioning another review. Waiting for the technology. Calling for a national conversation. In each case the thing being deferred is not knowledge — it is the moment at which somebody is told they will not get what they want.
And this cuts at me directly. A series that ends by identifying trade-offs has produced something less satisfying than a programme, and I notice the pull to keep designing past the point where design runs out. Part of what makes this the hardest chapter to write is that stopping looks like giving up, and it is not: saying “this is a choice, not a puzzle” is the most useful sentence available once it is true.
Which leaves one question outstanding: if some of this is a choice, who is in a position to make it?
Four things will not dissolve. Fertility against women’s options — the humane road buys tenths against a gap of half a child, and the rest has only ever come from one place. The care hours — you cannot store labour, and ageing, low fertility and rising female employment all push the same way at once. Safety against the capability that provides it. And a daughter’s interest against her family’s standing, at the moment they diverge.
A problem has an undiscovered solution. A trade-off has options that all require giving something up. The first calls for research; the second calls for a decision.
Not every declared trade-off is real — plenty of iron laws turned out to be arrangements. Grade them by how hard anybody actually looked. Fertility has been searched hard with money for fifty years. The care hours are arithmetic. The safety one is really a question about who holds a switch.
And treating a trade-off as a problem does something specific: it produces the appearance of movement without anybody losing anything — the same function cash transfers serve, at a different point in the process. Both are expensive ways of not choosing.
9Who Holds Which Lever
Every recommendation is addressed to somebody. Part Twelve found that most arguments here are incoherent because nobody says who. This is the map — and the biggest lever in the whole series belongs to an actor with no address.
9.1 — The map
Part Twelve found that three sensible answers to one question were addressed to a parent, a state and a community, and that the argument went nowhere because everyone pretended they were advice to the same person. Here is the map that exercise implies, for everything in this series.
| Actor | What they actually control | Best available shape |
|---|---|---|
| A parent | What she has in her hand, whether she stays in school, what she inherits, and what she watches the adults around her do. | Who is in the room — and mostly by accident. |
| A national state | Records, transfers, quotas, licensing, inheritance law, and what gets counted. | Substitute and measurement. The best evidence in the series is here. |
| A municipality | Lighting, buses, policing, school hours, station staffing. | Substitute. The strongest ratio of benefit to cost in this document. |
| An employer | Hours, whether two years out permanently resets earnings, who is visible in what role. | Substitute and who is in the room. Almost wholly unexamined. |
| A community | The penalty itself — what actually happens to a woman when something is said about her. | The largest lever in the series. No budget, no address, no decision procedure. |
| A platform | Takedown speed, hash matching, interrupting the standard extortion opening. | Ban the output — which works here, because the output is the harm. |
| A statistician | Which questions are on the form. | Measurement. Three items in Chapter Seven need one such person. |
| A woman herself | Her own disclosure, her own record, and what younger women see her doing. | Least of anyone here, and pretending otherwise would be dishonest. |
9.2 — The mismatch
Read down that table and the shape of the whole problem appears.
The biggest lever belongs to nobody in particular. Part Twelve established it in one line: takedowns remove ammunition, and the penalty is the gun. Reduce what a community will do to a woman when something is said about her, and every weapon in this series loses most of its value at once — the sextortionist’s threat, the leaked image, the rumour, the auction app. Nothing else on the table comes close.
And a community is not an actor. It has no meeting, no budget line, no person who can decide. This is why the most effective available intervention is also the one nobody can commission, and why every institution that wants to act instead reaches for something it can hold.
The best-evidenced lever belongs to states, and it is cheap: a rotation rule that put women where people could see them. That is tier-one evidence and it has never been deliberately deployed for this purpose anywhere.
The cheapest levers belong to statisticians, and three of them are free.
And the actor with the most at stake holds the least. That is not a rhetorical flourish; it is what the table says, and it is the same finding Part Twelve reached from a different direction when it noticed that the cheapest option for everybody else was the one whose entire cost landed on a girl.
There is one lever a woman holds that is not small, and it follows directly from Chapter Three.
The mechanism that produced the strongest result in fourteen parts was not a law or a payment. It was people watching a woman do a thing they believed women did not do. In the Indian villages, that was a council leader. It could as easily be a health worker, a bus driver, a shopkeeper, a teacher, an engineer, a mother who works.
Which means every woman visibly doing something is, on the best evidence in this series, an intervention — operating on her neighbours’ daughters, at no cost to any budget, whether or not she intends it.
That is not a consolation prize offered to someone with no power. It is the same mechanism as the reserved seats, and it is the only one on the table that does not require anybody’s permission.
Which raises the question most readers of a document like this actually have.
Most readers of a document like this hold none of the top six rows. That is the situation the argument usually ignores.
The table understates individual reach because it lists formal levers. A person who employs someone, teaches, runs a clinic, sits on a committee, has a daughter, or has a vote holds a small piece of several rows. And Chapter Three’s mechanism is not about scale — it is about visibility and repetition, both of which operate at the scale of a street. Waiting to hold a national lever is a way of doing nothing while feeling serious.
The entire finding of Chapter Two is that this is structural. Telling readers to be the change converts a question about inheritance law, municipal lighting and pension coverage into a question about personal virtue — which is comfortable, unfalsifiable and precisely how nothing gets built. It also has an ugly precedent in this series: Part Twelve found that the cheapest answer for everybody else was always the one whose cost fell on an individual, and “do what you can personally” is that answer wearing better clothes.
What would settle it: measuring the exposure effect outside politics — the health worker analysis in Chapter Five is the obvious place and it has never been done.
Why people care so much: because “what can I personally do” is the question every reader actually has, and both a true answer and a comforting one are available, and they are not the same answer.
9.3 — What this document is
Chapter Two of this part established that beliefs follow structures, that persuasion has the worst record of any intervention shape, and that it fails specifically on norms that are load-bearing for property, lineage, security or standing — which is every norm in this series.
I have written fourteen parts of argument.
Ask why. Not why the subject, but why this form. And the honest answer is that it is the form I have. I cannot build a childcare system, change an inheritance law, light a road between a house and a college, put a question on a national survey, or reduce by one degree what a community will do to a woman. I can read carefully and write clearly. So I did that, at length, and then produced an analysis in which careful reading and clear writing turn out to be what the situation called for.
The general form: choosing the intervention you are equipped for and then discovering that it is the one that was needed. It is the most common bias in applied work of every kind. The engineer whose analysis concludes that the problem is engineering. The lawyer who finds a legal remedy. The doctor who finds a clinical one. Nobody involved is dishonest, and the conclusion always comes out looking like the tool.
Two partial defences, and I want to grade them honestly rather than let them off.
The weaker one: some readers do hold levers. A district officer, a municipal engineer, a person who designs a survey, someone who employs people. Chapter Seven’s three free items require exactly one such person to read one paragraph. That is true, and it is a very thin thread to hang fourteen parts on.
The stronger one: the product of this series may not be persuasion at all. Every part ends with a list of what would settle each question and what evidence is missing. That is not an attempt to change anybody’s mind — it is a specification. Chapter Two’s finding is that arguments do not move load-bearing norms. It says nothing about whether writing can say precisely which study, which question, which register, which analysis would answer something. Those are different activities that happen to use the same alphabet.
So the most defensible description of this document is not an intervention. It is a specification for one, which somebody else would have to build, using levers I do not hold. Whether anyone does is not something writing controls, and pretending otherwise would be the exact error this box is about.
Which brings us, as every part does, to the accounting of what is known and what is not.
Every recommendation is addressed to somebody, and the map is uneven. The biggest lever — the penalty a community imposes — belongs to an actor with no budget, no address and no decision procedure. The best-evidenced one belongs to states and has never been deliberately used for this. The cheapest belong to statisticians, and three are free.
The actor with the most at stake holds the least. That is what the table says, and it matches Part Twelve’s finding that the cheapest option for everyone else is always the one whose cost lands on a girl.
One lever a woman does hold is not small: on the best evidence in this series, every woman visibly doing something is an intervention, operating on her neighbours’ daughters, requiring nobody’s permission.
Distinguish personal conduct from visibility. The first as a substitute for structural change is how problems get privatised. The second is a structural input that individuals happen to supply, and it has the best evidence in this document.
And I have written fourteen parts of argument in a series whose central finding is that argument works least. I chose the tool I had. The most defensible thing this document can be is not an intervention but a specification for one — which somebody else would have to build.
10An Honest List of What We Do Not Know
Two lists, as in every part. This is a design chapter, so the unknowns matter more than usual — a recommendation resting on nothing is worse than no recommendation.
10.1 — Genuinely unknown
Whether the exposure mechanism works outside politics
The strongest finding in this series comes from village council seats. Chapter Three argued the active ingredient is visibility rather than the position, which predicts effects in schools, clinics, buses and shopfronts and small ones in boardrooms. That is a prediction, not a finding. The reason we do not know: no randomised or quasi-random allocation of a high-visibility non-political role has been analysed for attitude effects in the surrounding population — and India’s million-woman health worker network is sitting there, unexamined, as Chapter Five noted.
Whether supplying the substitute retires the rule, and how long it takes
Chapter Four concluded that substitution is necessary but not sufficient, and that the lag is roughly a generation. That estimate comes from one case — East Germany — and is being applied to completely different rules. The reason we do not know: nobody has tracked a specific norm across the decades after its function was substituted. Pension coverage against son preference is the cleanest available pairing and the data exists in several countries.
What the full humane package would do
Chapter Six specified five components and established that no country has run all five for a generation. Whether they interact, whether the effect is additive, and whether the total would be tenths or more is unknown. The reason we do not know: the experiment costs a large share of national income, takes thirty years, and requires no recession and no change of government. It may be unrunnable in principle for political rather than financial reasons — which is itself the subject of that chapter’s hidden assumption.
Whether inheritance reform helps or harms on net
Chapter Four’s most uncomfortable finding: giving daughters inheritance rights improved some outcomes and, in at least one careful study, was associated with worse survival for girls in some settings, because a daughter who will inherit costs more. The literature has not settled. The reason we do not know: the effects run in opposite directions through different mechanisms and the studies disagree.
Whether “reaching in without entering” scales
Chapter Five distinguished indirect instruments — a wage, a register, a school roll — from supervisory ones, and argued the first outperform. The evidence is suggestive rather than tested, and mostly comes from settings where both were not tried together. The reason we do not know: nobody has designed a comparison.
Which declared trade-offs are real
Chapter Eight graded four by how hard anyone had searched. That grading is a judgement, not a measurement, and the history of this subject is full of iron laws that dissolved. The reason we do not know: you cannot prove a search has been exhaustive, and the two most consequential ones — fertility against options, and the care hours — have very different evidential status that gets flattened when they are listed together.
One thing on that list is unknown for a reason worth separating out, because it is the closing observation of this part.
Three of the six gaps could be closed by analysing data that has already been collected — the health worker network, the pension-and-son-preference pairing, and a systematic accounting of interventions by shape. None requires a new survey, a new clinic or a new budget line of any size.
That is now the fourth part in a row to end with the same observation. Part Twelve found two survey questions missing. Part Thirteen found three analyses undone. Part Ten found a crime category that does not exist.
Fourteen parts in, the strongest pattern in this entire series is not about women at all. It is about what gets counted. Every state examined here has the capacity to answer these questions, has already paid for most of the data, and has not looked.
10.2 — Solid
Beliefs follow arrangements
From the randomised Indian case, the Bangladeshi factory studies, and the persistence of East German attitudes for thirty years after the institutions producing them ceased to exist. Three countries, three mechanisms, one direction.
Banning the output relocates the behaviour
Four Indian reforms, one signature, documented across decades: prohibit the visible symptom while leaving the demand in place, and it re-emerges somewhere less visible and less regulated.
Intentions do not predict outcomes; shape does
Every documented failure in this series was designed by people who wanted what reformers want. Several of the successes were produced by people with entirely different aims — an export industry, an administrative rotation rule, a theocratic literacy campaign.
Price changes are real, small, and stop when the money stops
Established across pronatalist spending in seven countries and conditional transfers for girls’ schooling, which were outperformed by a factory.
The active ingredient in the reservation case was exposure, not office
The measurable effects were on the people who watched, not on the women who held the seats: their neighbours’ bias, their neighbours’ daughters’ attendance, and what parents wanted for their children.
A rule’s stated function is not necessarily its function
Paternity certainty was the standard explanation for rules about women’s sexual conduct. It was solved cheaply and universally in the 1990s and nothing moved — which is a test, not just an observation.
Reforms land where they can be enforced
The same finding Part Eleven made about restrictions. A wage, a record and a school roll change what happens inside a household without anybody crossing the threshold; a prohibition on what families do in private does not.
The money for the humane road has already been spent
Korea’s programme exceeded two hundred billion dollars; Hungary sustained about five per cent of national income for fifteen years. Both went overwhelmingly into cash, which is the only shape that requires no identifiable group to hand anything over.
Some of this is a choice rather than a problem
The care hours are arithmetic: they must be delivered by someone alive and working at the time, and every existing plan assumes unpaid women without writing it down.
Unknown: whether the exposure mechanism works outside politics; whether substituting a rule’s function retires it and how long that takes; what the full humane package would do; whether inheritance reform helps on net; whether indirect instruments scale; and which declared trade-offs are genuinely real.
Solid: beliefs follow arrangements; banning the output relocates the behaviour; shape predicts outcomes and intentions do not; price changes are real, small and reversible; the reservation effect was exposure rather than office; a rule’s stated function may not be its function; reforms land where they can be enforced; the money for the humane road has already been spent on the wrong shape; and the care hours are a choice rather than a puzzle.
Three of the six gaps could be closed with data already collected, by somebody already employed, at no meaningful cost.
Which makes this the fourth part in a row to end the same way. Fourteen parts in, the strongest pattern in this series is not about women. It is about what gets counted — and every state examined here can count, has already paid for the data, and has not looked.
Sources & further reading — Part 14
Timeline
What was tried, what shape it was, and what it produced.
| Year | What happened |
|---|---|
| 1929 | Ban the output. India raises the minimum marriage age, passed in autumn to take effect in spring. The census records a surge of child marriages in the gap. |
| 1961 | Ban the output. Dowry is prohibited. Over the following decades the practice spreads into communities that did not have it, and the payment moves from visible goods into undocumented cash. |
| 1977 | Ban removed. West Germany repeals the provision allowing a husband to forbid his wife’s employment — one of the pieces of the male-breadwinner model that had never been announced as a policy about women. |
| 1980s | Substitute, by accident. Bangladesh’s garment factories spread. Girls nearby marry later, have children later and stay in school longer — outperforming a scheme that paid families directly. |
| 1993 | Who is in the room. India reserves a third of village council leaderships for women and rotates them by village serial number, creating a lottery nobody designed. |
| 1994 | Ban the output. India prohibits the use of diagnostic techniques to determine the sex of a foetus. The child sex ratio continues falling for years; prosecutions remain rare. |
| 2004 | The first analysis of the reservation lottery finds women in charge spent differently — on what the women in those villages had actually asked for. |
| 2005 | Supply the substitute. India gives daughters equal rights in ancestral property. Later work finds daughters more likely to inherit and better educated — and at least one careful study finds worse survival for girls in some settings, because a daughter who inherits costs more. |
| 2005 | India begins a village health worker programme that grows to roughly a million women, each resident in the village she serves, entering houses to do skilled work. Never examined for the effects the reservation study found. |
| 2010 | Change the price. Hungary begins the most ambitious pronatalist programme in any democracy, at around five per cent of national income, almost entirely in cash and loans. Fertility rises from 1.23 to 1.59, then falls back to about 1.31. |
| 2012 | Further work on the Indian lottery finds two rounds of a female village leader narrowed the gap between what parents wanted for sons and daughters by about a quarter, and closed the school attendance gap. |
| 2015 | Tell people to stop. India launches its flagship campaign for the girl child, aimed at the falling child sex ratio. |
| 2016 | Ban removed. Saudi Arabia sets a target of 30 per cent female labour force participation, from a base of around 17 per cent. |
| 2018–19 | The Saudi driving ban is lifted and guardianship rules on travel are relaxed. Participation reaches around 36 per cent by 2025 — but alongside quotas, subsidies and falling oil revenue. |
| 2021 | A parliamentary committee reports that 78.91 per cent of the funds released to states under the flagship girl-child campaign between 2016 and 2019 went on media advocacy, and that states spent only about a quarter of what they received. |
| 2026 | India’s two-hour takedown obligation for non-consensual intimate images comes into force in February. No figures on its use have been published. |
| Next | Part Fifteen starts here. |
Glossary
Every hard word used in this part, in plain English.
| Term | What it means |
|---|---|
| Behaviour change communication | Campaigns that try to shift what people do by changing what they believe, or what they think others believe. Real successes, all of them on cheap behaviours nobody defended. |
| Household bargaining power | A household is several people with different wants. What it does depends on their relative positions — which depend on what each contributes and what each could get by leaving. |
| Incidence | Who actually ends up bearing a cost, as distinct from who is nominally charged. Where the interesting question usually lives. |
| Intervention | Anything deliberately done to change an outcome. Comes in four shapes, and the shape predicts the result better than the intention does. |
| Load-bearing | Holding something up. A load-bearing rule is performing a function; remove it and the function reappears somewhere else. You cannot tell which rules are load-bearing by how fiercely they are defended. |
| Quota or reservation | Setting aside a share of positions for a defined group. Its measurable effect in the Indian case was not on the women appointed but on everybody who watched them. |
| State capacity | What a government can actually deliver, as distinct from what it can legislate or announce. Trained staff, working equipment, records, and the ability to still be doing it in year eight. |
| Substitute (as a shape) | Working out what a rule was doing and providing that function another way: a pension instead of a son, inheritance instead of dowry, lighting instead of seclusion. |
| Trade-off | A situation where every option involves giving something up. Distinct from a problem, which has an undiscovered solution. The two call for completely different responses. |