Before We Begin — Where We Left Off
Part Nine was about arithmetic. Across most of the world, and now across most of India, women are having fewer children than are needed to replace the people already alive. I went through what that number actually measures, why it fell, whether any government has ever pushed it back up, and who would have to pay if one tried.
The short version. The fertility rate is not a count of anybody's children — it is a snapshot of one year dressed up to look like a lifetime. It fell because children stopped dying, because children stopped being workers and became investments, and above all because the cost of a child rose sharply for the person who bears it. No country has ever restored replacement fertility through policy, though several have spent enormous sums trying. Coercion works in one direction only: it can push births down, and it cannot pull them up. And underneath the whole argument sat an unpriced assumption — that somebody would be available to do the physical work of caring for the old, which in every society on Earth means women.
Part Nine was written at the scale of nations. Millions of people, decades, national statistics. That scale is honest for the question it was asking, and it is also where this subject goes to hide. It is very easy to discuss birth rates for eighty pages without once picturing a person.
So this part goes the other way. All the way down.
Every part of this series so far has circled one specific claim without ever testing it head-on. The claim is that a woman who breaks the rules about sexual behaviour damages her own life — that she will struggle to marry, that her marriage will fail, that she will regret it, that she will be less healthy, less happy, less able to attach to anyone. It is the warning your grandmother gives. It is the warning a stranger gives a woman on the internet. It is offered not as a moral rule but as a kindness: I am telling you this for your own good.
That is a prediction. Predictions can be checked. This part checks it.
How this document is built
If you have read earlier parts you know the notation already. If not, here is the whole system. Six kinds of box, each doing one job, each looking different so you can see at a glance what you are about to read.
The first explains a hard word the moment it appears, so you never carry an unexplained term forward.
Association: two things going up and down together. Taller people tend to weigh more, so height and weight are associated.
Why it matters here: almost every finding in this document is an association. An association tells you that if you know one thing about a person, you can guess a little better about the other. It does not tell you that one caused the other, and the gap between those two sentences is what most of this part is about.
The second takes a number too big or too abstract to picture and turns it into something with a body.
A researcher says a factor “doubles the risk” of something. That sounds enormous. Then you learn the risk went from two in a hundred to four in a hundred.
Both statements are true. One of them makes you frightened and the other makes you shrug, and the second one is the one that tells you what to expect. Whenever you meet a multiplied risk in this document, I will give you the underlying numbers as well.
The third shows the actual evidence behind a claim, and then says what that evidence cannot show. It is the box that lets you decide how much to trust a sentence.
Most of what is known about sexual behaviour and later life outcomes comes from a handful of large national surveys, which ask people about their own pasts. In the United States that is chiefly the National Survey of Family Growth. In Britain it is the National Survey of Sexual Attitudes and Lifestyles. In India it is the National Family Health Survey.
What they are good at: size, repetition and consistency. Tens of thousands of people, the same questions across decades, so a change over time is a real change.
What they are bad at: they are asking people to report, honestly, on the thing they have most reason to lie about. Chapter One is entirely about how badly that goes.
The fourth is for places where informed people genuinely disagree. Each side gets its best case, not a version I find easy to knock down.
Every one of these has the same shape. 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 is the signature of this series. It does not argue with either side. It digs out what both sides are assuming without noticing.
Not caveats, not disclaimers. Unexamined premises sitting underneath an argument everybody is having. There are five in this part, and the last one is turned on this document.
The sixth closes every chapter, restating it in the plainest words available. A reader who read only these should still have the whole argument.
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.
Three warnings specific to this part
The first is about me, and it is heavier here than anywhere else in the series. I am an Indian man in my twenties. This part is about women's sexual histories and what happens to them afterwards. I am not a neutral instrument pointed at this question: I grew up inside the culture that issues the warning, I have heard it applied to women I know, and I have almost certainly repeated versions of it without examining them. There are two errors available to someone in my position. One is to find the warning confirmed because it is familiar. The other is to overcorrect and find it refuted because refuting it feels modern. I have tried to guard against both by writing down the effect sizes rather than the conclusions, and by giving the traditionalist case its genuinely strongest form in Chapter Nine rather than a version I can dismiss. Where I think I am at risk, I say so in the text.
The second is about tone. Nothing in this part is advice to any woman about what to do, and nothing in it is a judgement of any woman for what she has done. A document that reported “the research finds X” and then added “so be careful” would be smuggling a recommendation in behind a finding. I am not doing that. What each reader should do with this is genuinely their business, and Chapter Nine explains why I think handing down advice here would be a category error rather than modesty.
The third is about the evidence itself. This is one of the most politically loaded research literatures in the social sciences. Findings here are picked up within hours by people who want them to mean something. Studies get quoted with the confounders stripped out, effect sizes get inflated in the retelling, and null results get very little attention in either direction. I have tried to give you the original size of every finding, who produced it, and what it does not establish. Where a claim is popular but weakly supported I say so — in whichever direction that cuts, and in this part it cuts both ways more than once.
1What Is Actually Being Claimed
Before you can test a warning you have to know what it predicts. This one turns out to be four different warnings wearing the same coat, and the central number it depends on is among the least reliable in social science.
1.1 — Four warnings in one coat
The warning sounds like a single thing. She will ruin her life. Listen closely and it is four separate claims, and they do not stand or fall together.
The market claim. She will struggle to find a husband. Men will not want her. Her family will not be able to arrange a match.
The bond claim. Even if she marries, the marriage will be worse. She will compare. She will not attach properly. It will end.
The feeling claim. She will regret it. She will carry something heavy. She will be less happy, and the unhappiness will come from inside her rather than from anybody else.
The body claim. She will pick up infections. She will be less able to have children when she wants them.
Separate these and something becomes visible immediately. The market claim is not a claim about her at all. It is a claim about what other people will do to her. If it is true, the warning is not describing a consequence. It is describing a punishment, and it is being delivered by the same community that administers it.
The other three are genuinely about her. They say something happens inside a woman, or to her body, that would happen whether or not anybody found out. Those are the claims that can be tested against research, and they are what Chapters Two to Six are for. Chapter Seven then does the test that separates the two kinds.
1.2 — The number that cannot be true
Nearly all of this research rests on one variable: how many sexual partners a person has had. So before anything else, we should look at how well that number can be measured. The answer is: badly, in a way that is not a small technical problem.
Take any large national survey of sexual behaviour — the American National Survey of Family Growth, the British National Survey of Sexual Attitudes and Lifestyles, and their equivalents elsewhere. Ask men how many women they have slept with. Ask women how many men. Then add up each side.
The totals do not match. Men report substantially more opposite-sex partners than women do — commonly around twice as many, sometimes more, and the gap appears in survey after survey, decade after decade, country after country.
This is not a finding about behaviour. It is arithmetically impossible. In a population with roughly equal numbers of men and women, every heterosexual encounter is counted once by a man and once by a woman. The two totals have to be equal. They never are.
So at least one side is wrong, and probably both. Researchers who have investigated this find several things happening at once. Men with high counts tend to estimate rather than count, and estimates round upwards. Women appear to undercount, particularly encounters they would rather not include. Both sexes answer differently depending on how the question is asked, who is asking, and whether the answer is typed privately or spoken aloud to an interviewer. And some of the gap is definitional — people disagree about what counts.
Social desirability bias: the tendency to give the answer that makes you look the way you are supposed to look, rather than the true one. It is not usually deliberate lying. It is the pull of the expected answer.
Why it matters here: the pull runs in opposite directions for men and women on exactly this question. A high number is a credit to a man and a debit to a woman in most cultures on Earth. So the error is not random noise that averages out. It is systematic, and it is pointed.
The scale of that is easier to see if you take the arithmetic seriously for a moment.
Imagine a village of a hundred men and a hundred women who only ever sleep with each other. Every encounter involves one of each. If you ask everybody and add up the answers, the men's total and the women's total must come to the same number. They must. There is no arrangement of behaviour that makes them differ.
Now run the actual surveys and the men's total comes out roughly double. That is the size of the reporting problem sitting underneath every study in this part. Not a rounding error — a factor of two, in the single variable everything else is measured against.
Hold that in mind through the next four chapters. It does not make the research worthless. Errors of this kind mostly blur associations rather than invent them, which means a real relationship will show up weaker than it is, not stronger. But it does mean that precise-sounding numbers — “women with exactly two partners” — are being built on a variable that cannot bear that much weight.
Some people conclude from the above that the whole literature is unusable. That is a serious position and it deserves a serious answer.
You cannot build knowledge on a number that is provably wrong by a factor of two, and wrong in a direction that tracks exactly the social pressure the research is supposed to be studying. Worse, the misreporting is not evenly spread: the women most exposed to shame are the ones most likely to under-report, which means the survey category “women with one partner” quietly contains an unknown number of women with more. Every finding about partner count is therefore a finding about reported partner count, which is a measure of willingness to disclose as much as of behaviour.
Almost everything in social science is measured imperfectly — income, health, happiness — and we do not throw those out. What matters is whether the error is likely to create a false finding or hide a true one, and random or blurring error does the second. Where the same association appears across different countries, different decades, different survey methods and different research groups, it is unlikely to be an artefact of one questionnaire. And some of these studies do not depend on exact counts at all: comparing women with zero partners to women with many is robust to a good deal of misreporting at the edges.
What would settle it: validation against something that is not self-report. Some of this exists indirectly — infection rates, pregnancy records — but there is no way to independently verify a life history, and there never will be.
Why people care so much: because the finer the category, the more specific the advice that can be built on it. “Women with exactly two partners” makes a headline. “Women with several rather than none” does not.
There is one more thing about how this evidence is collected, and it is the largest fact in the chapter. It is not in the numbers. It is in what was never collected.
Every study in this part asks what happens to women. The warning is issued to women. The research question is asked about women. And, quietly, the data needed to ask it about men often does not exist.
The American survey that produced the best-known findings on premarital partners and divorce does not carry full information on men's premarital sexual behaviour. The researcher who published those findings said so himself, in the same paper. So when the results appeared and were used to argue that a woman's history predicts her marriage, the comparison that would have told you whether this is a fact about sexual history or a fact about women's sexual history had already been made impossible at the design stage.
That is not a conspiracy. It is what happens when a question feels natural in one direction and strange in the other. Nobody set out to make the male comparison unavailable. They simply never felt the need to collect it, because the interesting question — the one that seemed worth funding a survey to answer — was always about her.
The general form: a question asked about only one group, so the data that would allow comparison is never gathered, and the absence is then read as though the answer were known. It appears everywhere. Pain research conducted on men. Crash-test dummies built to a male body. Drug trials that excluded women for decades. In each case the missing half is not a gap in the findings — it is a gap in what was ever asked.
Why this one matters for everything that follows: a large share of the arguing in this field is about how big the effect on women is. Almost none of it is about whether the same effect exists in men. If it does, then this is not a rule about women at all, and the warning has been aimed at the wrong half of the population for a century. If it does not, that is a genuine finding — and one nobody has properly established, because the survey did not ask.
1.3 — What would count as the warning being true
Let me set the bar before we look at any evidence, so I cannot move it afterwards.
For the warning to be true in the way it is meant — as advice, as a kindness, as a fact about the world — three things would have to hold.
The association would have to be real. Women with more partners would have to do measurably worse on the outcomes named, in good data, more than once.
It would have to be caused by the behaviour, rather than by something that produces both. If the same trait or circumstance leads a woman both to have more partners and to have a rockier marriage, then changing the first would change nothing.
It would have to be large enough to matter to a person. An effect can be real, causal, and still too small to base a life on.
Those three tests structure the rest of this part. Chapter Two takes the first. Chapter Three takes the second. And the third — the size — is why every number in this document comes with its underlying figures rather than a multiplier.
The warning is four claims, not one: that she will not be able to marry, that her marriage will be worse, that she will feel worse, and that her body will suffer. They need different evidence and they have different implications.
The first of those — that men will not want her — is not a claim about her. It is a claim about what other people will do, delivered by the people who will do it.
The whole literature rests on self-reported partner counts, and men's and women's totals do not match. They cannot both be right; in a closed population the two sums must be equal. The gap is often around a factor of two.
The error is not random. It runs in opposite directions for men and women, because a high number is a credit to one and a debit to the other.
The data needed to ask the same question about men was frequently never collected — so the comparison that would tell you whether this is about sexual history or about women's sexual history has been unavailable from the start.
2The Divorce Question
This is the oldest finding in the field and the one the warning leans on hardest. The association is real. It is also smaller, stranger and less stable over time than almost anyone quoting it realises.
2.1 — Ninety years of the same result
In 1938 the American psychologist Lewis Terman published a study of marital happiness which reported that women who had had sex before marriage were more likely to have unstable marriages, and that more partners went with more instability. Terman was a pioneer of psychological testing. He was also an enthusiastic eugenicist, which is worth knowing, because the question of what a woman's sexual history does to her did not arrive in the research literature neutrally. It arrived carried by people who already had views about which women should be having children.
Since then the basic association has been found again many times, in different countries, with better methods. That matters. A finding that survives ninety years, several changes of research fashion, and a complete transformation in the behaviour being studied is not nothing.
So let us be clear at the outset: the association is real. Anyone who tells you the research shows no relationship between premarital sexual history and later marital stability is not describing the literature.
The interesting questions are what shape it has, how big it is, and what causes it. All three answers are surprising.
2.2 — The actual numbers
The most detailed public analysis comes from Nicholas Wolfinger, a sociologist at the University of Utah, using the American National Survey of Family Growth. He looked at more than ten thousand women and calculated the share whose marriages ended within five years, sorted by how many sexual partners they reported before marrying.
The National Survey of Family Growth is a large, repeated, nationally representative American survey run by the federal statistics agency. It asks about partners, marriages, cohabitation, births and separations, and it has run in waves for decades. Wolfinger used the three most recent waves available to him, collected in 2002, 2006–2010 and 2011–2013, and grouped women by the decade in which they married.
What that design is good at: sample size, national representativeness, and the ability to compare decades using the same instrument.
What it is bad at: everything in Chapter One. It is self-report, it depends on remembering and being willing to say, and — as noted there — it does not carry equivalent data on men.
One more thing you should know and weigh for yourself. These analyses were published through the Institute for Family Studies, an American organisation with a stated commitment to marriage. That does not make the numbers wrong; the underlying survey is public and the finding long predates the organisation. It does mean you are reading a result produced by people who were pleased to find it, and the honest thing is to say so rather than present it as though it came from nowhere.
Here are the figures for women who married in the 2000s. The number is the percentage whose marriage had ended within five years.
| Premarital partners | Divorced within five years |
|---|---|
| None (married as a virgin) | About 6 per cent |
| One | Just over 20 per cent |
| Three to nine | About 25 per cent |
| Two | About 30 per cent |
| Ten or more | About 33 per cent |
Read that table slowly, because it does not say what people think it says.
2.3 — The curve is not a line
The warning predicts a line. More partners, worse outcome, all the way up. That is not what the table shows.
Women with two partners divorced more than women with three to nine. In earlier decades the pattern was even stranger: among women marrying in the 1980s and 1990s, those with exactly two premarital partners were more likely to divorce than those with ten or more.
Wolfinger's own summary of this was that if you are going to have comparisons to your future husband, it is better to have more than one. Whether or not that explanation is right, the shape is the point. A dose-response relationship — more of the thing, more of the harm — is what you expect when something is genuinely causing damage. This is not that shape.
Dose-response relationship: when more of a cause produces more of an effect, steadily. More cigarettes, more lung cancer. It is one of the strongest signs that a relationship is genuinely causal.
Non-monotonic: a relationship that goes up and then down, or down and then up. The middle is worse than either end, or better.
Why it matters: a non-monotonic relationship is a warning sign that you are looking at several different groups of people mixed together, rather than one process operating at different strengths. Women with two partners are probably not “women with one partner, only more so”. They are likely a different kind of situation altogether.
The other feature of the table is the size of the first step. The gap between no partners and one partner is about fourteen percentage points. The gap between one partner and ten or more is about thirteen. In other words, roughly half of the total range in this table is crossed between zero and one — between never having had sex before marriage and having had sex only with the man you married.
That is not a story about how many. It is a story about whether at all, and it points at something other than accumulation. Women who marry as virgins in a country where five per cent of brides do are an extremely distinctive group — highly religious, usually marrying young into a community that supports the marriage heavily, and often facing a much higher barrier to divorce regardless of how the marriage is going. Chapter Three takes that apart.
Take a hundred American women who married in the 2000s having had ten or more partners. Within five years, about 33 of those marriages had ended, and about 67 had not.
Now take a hundred who had three to nine. About 25 ended; 75 did not.
The difference between those two groups is about eight women in a hundred. Real, and worth knowing. But the sentence “having many partners increases your divorce risk” and the sentence “eight women in a hundred” describe the same fact, and they do not feel the same at all.
And the majority outcome in every single row of that table is a marriage that was still going.
2.4 — The finding will not sit still
Now the part that should make everybody cautious, on all sides.
The behaviour being studied has changed beyond recognition within the span of the data. Among American women marrying in the 1970s, about 21 per cent married as virgins; by the 2010s it was about 5 per cent. The share who had had ten or more partners went from about 2 per cent to about 18 per cent. The share with exactly one premarital partner — nearly always the husband — fell from about 43 per cent to about 22 per cent.
When a behaviour goes from unusual to normal, the people doing it change completely. In 1970, having ten partners marked a woman out as very unusual in her community, with everything that implies about her circumstances and how she was treated. In 2015 it marks her out as ordinary. Any association between the behaviour and an outcome is therefore measuring something different in each era, even with identical questions.
And the numbers moved accordingly. Divorce rates fell for the shrinking group of virgin brides. They rose most for women with ten or more partners. The relationship between two partners and ten or more flipped. These are not the fingerprints of a stable biological or psychological mechanism. They are the fingerprints of a social process.
Both sides here accept the table. They disagree about what a person should take from it.
Ninety years, many countries, many research teams, one direction. The association survives controls for age, education, income and religion in study after study. It is bigger than plenty of risk factors that public health bodies act on without hesitation. And the effect on the extremes is not trivial: a third of marriages ending within five years, against six per cent, is an enormous difference in life outcomes by any standard. Dismissing it because the middle of the curve is untidy is holding this finding to a standard no other social science result is held to.
The shape is wrong for a causal story, the association moves around by decade, and half the range sits in the step from zero to one, where the comparison group is a tiny and unrepresentative religious minority who also face the highest barriers to leaving a marriage. A “divorce” is not the same event for a devout woman in a close community and a secular woman in a city, so the outcome variable is not measuring the same thing across the rows. And the whole table is built on a self-reported number that cannot be verified and is misreported in a patterned way.
What would settle it: a design that separates the woman from her circumstances — comparing sisters, or following people from before any of this happened. Some of that exists. It is in the next chapter, and it does not favour the simple reading.
Why people care so much: because this table is the closest thing the traditionalist case has to a hard number, and the closest thing the other side has to a target. Both have an incentive to read it as more decisive than it is.
2.5 — The exception that survives
One finding in this area has held up better than the rest, and it is worth stating precisely because it is the one most often mangled.
In 2003 the sociologist Jay Teachman published an analysis showing that women whose only premarital sexual partner — and only cohabiting partner — was the man they went on to marry had no elevated risk of divorce compared with women who had had no premarital sex at all.
Note what that does and does not say. It does not say premarital sex is harmless in general. It says that premarital sex with your eventual husband did not carry the penalty. Which means whatever the association is tracking, it is not the act.
The association is real. It has been found since 1938, across countries and decades. Anyone who says the research shows nothing is misdescribing it.
But the shape is wrong for damage. It is not a line. Women with two partners divorced more than women with three to nine, and in earlier decades more than women with ten or more. That is not what a dose of harm looks like.
Roughly half the whole range sits between zero partners and one — and women who marry as virgins are now about five in a hundred American brides, a tiny and highly distinctive group who also face the biggest barriers to leaving.
The finding will not sit still. As the behaviour went from rare to normal, the numbers moved and one comparison reversed outright. Stable mechanisms do not behave like that; social processes do.
And the cleanest single result in the field is that a woman whose only premarital partner was her future husband carried no extra risk at all — which means whatever this is tracking, it is not the act itself.
3The Machinery of Getting It Wrong
An association is a fact about two columns in a spreadsheet. Turning it into a fact about causes is where nearly everything goes wrong, and this literature contains one of the most honest failures in the field.
3.1 — Three ways to misread a true association
Suppose it is genuinely true that women with more premarital partners divorce more often. There are four possible explanations, and only one of them is the one the warning assumes.
One: the behaviour causes the outcome. Something about having had several partners damages a later marriage. This is the warning's version.
Two: something else causes both. Some third thing makes a woman more likely to have several partners and more likely to divorce, with no arrow between the two.
Three: the causation runs backwards. Not literally in time here, but in the sense that the same underlying situation produces both — a woman heading for a difficult marriage may already be living a life that produces more partners.
Four: it is an artefact of who ends up in which group.
Confounding: when a third factor causes both of the things you are comparing, creating an association between them that is real in the data and empty as a cause.
The standard example: ice cream sales and drowning deaths rise together. Ice cream does not drown anybody. Hot weather causes both.
Selection effect: when the people in a group got there for reasons that also affect the outcome, so you are comparing different kinds of people rather than the same people under different conditions.
Why it matters here: the two most powerful objections to the divorce finding are both of these, and neither has been answered.
3.2 — The list of things that come bundled with a partner count
Consider what else is true, on average, about a woman who reports more premarital partners in an American survey.
She is less religious. She married later. She began having sex younger. She is more likely to have cohabited, and cohabitation carries its own association with divorce. She is more likely to have grown up in a household that itself broke up, which is one of the strongest known predictors of divorce entirely on its own. She scores differently on stable personality traits — higher on impulsivity and sensation-seeking, which are traits that affect how a person handles a marriage as well as how many relationships they form. She lives somewhere with different norms about leaving. And she is, on average, in a different economic position.
Every one of those independently predicts divorce. All of them travel with partner count.
Picture two women, both aged thirty, both married three years.
The first grew up in a devout household that stayed together, married at twenty-two into a community where her parents, in-laws, neighbours and priest all know the marriage exists and would all know if it ended.
The second grew up in a household that split when she was nine, left home at eighteen, moved cities twice, married at twenty-eight, and lives four hundred kilometres from anyone who knew her as a child.
They will have very different partner counts. They will also have very different divorce risks for reasons that have nothing to do with sex at all. Any study that compares them and attributes the difference to partner count has attributed an entire pair of life histories to one column.
Researchers control for these things statistically, and the association usually survives some controls. But statistical control only works for factors you have measured. Nobody has a good measure of how much a woman's community would punish her for leaving, and that is precisely the variable most likely to be doing the work.
3.3 — The sibling test
There is a way to get closer to an answer, and it is the best tool this field has.
In a sibling-comparison design, researchers compare brothers and sisters raised in the same household who differ in the behaviour being studied. Two sisters, same parents, same house, same religion, same neighbourhood, same income, largely overlapping genetics — but one had sex at sixteen and one at twenty-two.
Why it is powerful: every family-level confounder is held constant automatically, including the ones nobody thought to measure. You do not have to know what the third factor was in order to remove it.
What it has found: across studies of early sexual activity and later outcomes, associations that look substantial in ordinary data shrink considerably once siblings are compared — in some cases to nothing at all, and in some cases the surviving relationships do not run in the direction the popular story predicts.
What it cannot do: siblings differ from each other for reasons, and those reasons may be the cause. It removes what families share. It cannot remove what makes two sisters different.
This is the standard fate of this kind of finding across the social sciences. An association looks strong in raw data, holds up under a few controls, and then loses most of its size the moment somebody finds a way to compare people who are genuinely alike. It happens so reliably that it should be the default expectation, not a surprise.
3.4 — When the researcher tested his own explanation
Now the most honest thing in this literature, and the reason I take the researchers involved seriously even where I read their numbers differently.
Wolfinger — the same sociologist who produced the divorce table in Chapter Two, publishing through an organisation that would have been delighted by a clean mechanism — went looking for one. He tested the standard explanations for why premarital partners might damage a marriage: that a woman compares her husband unfavourably with earlier partners; that premarital sex leads to children outside marriage which strains later relationships; that people who accumulate partners are worse at choosing a mate; that it is all selection on attitudes and personality.
His conclusion, published in 2023, was that none of the commonly theorised mechanisms accounted for the relationship.
Sit with that for a moment. The best-known finding in this field comes with the researcher's own report that the explanations offered for it do not work. That is not a refutation of the association — it is still there in the data. It is something more specific and more useful: nobody can currently say what it is made of.
An unexplained association is a legitimate thing to keep studying. It is not a legitimate thing to give somebody as advice, because advice requires knowing which lever to pull.
This is the central methodological dispute in the whole part, and everything in Chapters Four to Six inherits it.
The association survives controls for age, education, religion, family background and income across multiple studies, and it has been found for ninety years in societies with different norms. Selection cannot be infinitely elastic — at some point “there is always another confounder” stops being an argument and becomes a refusal to accept any evidence. There are also plausible mechanisms that are hard to measure but not exotic: comparison with past partners is a real psychological phenomenon that people report directly, and a person who has left several relationships has practised leaving, which is a skill like any other. The fact that we cannot yet name the mechanism does not mean there is not one.
The shape of the relationship is wrong, the size moves by decade, sibling designs attenuate it, and the researcher who owns the headline finding reports that no proposed mechanism explains it. Meanwhile every unmeasured variable that would produce this pattern is exactly the kind that is hardest to measure — how much your community would punish you for divorcing, how stable your childhood was, what you are like as a person. And there is a decisive clue in the data itself: a woman whose only premarital partner was her future husband carries no extra risk. If the mechanism were about sexual experience, that group would show it. They do not.
What would settle it: nothing that anyone can ethically run. The experiment would require assigning young women to different numbers of sexual partners at random. That is not a limitation to be worked around; it is permanent. Everything in this chapter is the best available substitute for a study that will never exist.
Why people care so much: because “cause” licenses advice and “selection” does not. If it is causal, telling a young woman to have fewer partners is useful information. If it is selection, the same sentence is a moral instruction wearing a lab coat.
Before leaving this chapter there is something to say about the thing being measured on the other side of the equation, which nobody in the argument above has questioned.
Every study in this chapter uses divorce as the outcome. The traditionalist cites the divorce rate as evidence of damage. The critic disputes the divorce rate. Both are treating a marriage that lasts as the good result and a marriage that ends as the bad one.
But a divorce rate does not measure how good marriages are. It measures how many of them ended, which is a product of two things: how bad the marriage was, and how hard it was to leave. Those are not the same, and the second one varies enormously across exactly the groups being compared.
Look again at the table in Chapter Two with that in mind. The row with the lowest divorce rate is the row of women who married as virgins — a group who are, on average, the most religious, the most embedded in a community that regards divorce as a failure, and the most likely to face family opposition, financial dependence and social loss if they leave. Their marriages ended less often. Whether their marriages were better is a completely different question, and the survey does not ask it.
The measures that would answer it exist. Researchers can and do ask about marital satisfaction, about loneliness inside a marriage, about whether a person would marry the same partner again. Those results are messier and get quoted far less, because “divorce” is a clean, dated, recorded event and happiness is not.
The general form: the measurable proxy that quietly became the goal. It runs through every field that has to score something. Schools measure exam results and start teaching to them. Hospitals measure waiting times. Companies measure billable hours. In each case the proxy was chosen because it could be counted, and then it became the thing everyone was trying to produce.
What follows here is uncomfortable for both sides. If low divorce is partly a measure of high exit costs, then a rule that lowers divorce by raising the cost of leaving will look, in the data, exactly like a rule that makes marriages better. The two are indistinguishable on this outcome variable — and the entire evidential case in Chapter Two is built on it.
None of which makes the association disappear. It means we have been measuring the wrong side of it, with an instrument that cannot tell a good marriage from an inescapable one.
A true association has four possible explanations and the warning assumes only one of them. The others are confounding (a third thing causes both), selection (the groups are different kinds of people), and the same underlying situation producing both.
Everything that travels with a high partner count — later marriage, less religion, cohabitation, a broken childhood home, personality traits, living far from family — independently predicts divorce on its own.
Sibling comparisons are the best tool available, because they remove every family-level confounder automatically. When applied to early sexual activity, they shrink these associations considerably.
The researcher who produced the best-known divorce figures went looking for the mechanism behind them and reported that none of the standard explanations accounted for it. The association stands; its content is unknown.
And the outcome everyone is measuring — divorce — partly measures how hard it is to leave, which means a rule that traps women and a rule that makes marriages better look identical in this data.
4The Regret Asymmetry
One of the few findings here that has survived the hardest test anyone could design for it. Men and women do regret different things. The difference is real, it is smaller than you would expect, and what predicts an individual woman's regret is not her sex.
4.1 — What the finding is
Ask a large group of people about their regrets in the sexual domain and a consistent pattern appears. Men more often regret things they did not do — the opportunity they passed up, the person they did not approach. Women more often regret things they did — a particular casual encounter, a person they wish they had not slept with.
Action regret: regret about something you did. “I wish I hadn’t.”
Inaction regret: regret about something you did not do. “I wish I had.”
Why the distinction matters: everybody has both. The finding is not that women regret sex and men do not. It is that the balance between the two tips in opposite directions for men and women, specifically in the sexual domain.
The first systematic demonstration came from a team led by Neal Roese in 2006. It was replicated by Andrew Galperin and colleagues in 2013, who added a control that matters more than the finding itself: they also measured regret about non-sexual romantic matters — the relationship not pursued, the confession not made. There, the sex difference disappeared.
That control is what makes this worth taking seriously. If women simply reported more regret about everything, or reported feelings more readily than men, the pattern would show up everywhere. It does not. It is specific to sex.
4.2 — How big is it?
Here is where most retellings stop, and where the useful part begins.
Effect size: how big a difference is, as opposed to how confident we are that it exists. These are completely different questions and they are constantly confused.
With a large enough sample, a difference so small it could not matter to anybody will still be reported as statistically reliable. “Significant” in a research paper means “probably not zero”. It does not mean “large”.
Why it matters here: the sex difference in sexual regret is reliable. Its size is the thing you actually need in order to know what it means for a person, and it is the part that gets dropped in the retelling.
The size is moderate. It shows up as a clear difference between group averages, with very substantial overlap between men and women. Plenty of men report high action regret. Plenty of women report none at all.
A moderate group difference of this kind works like the height difference between men and women in reverse: real, visible in averages, and useless for predicting an individual.
If you know a person's sex, you can shift your guess about their sexual regret a little. If you want to actually know, you have to ask them — because the range within each sex is far wider than the gap between the two.
Put it as a practical test. Take a woman at random and a man at random. The woman will report more action regret than the man rather less than two times in three — which means it goes the other way often enough that the pattern tells you almost nothing about the person standing in front of you.
4.3 — The hardest test anybody has run
There are two broad explanations for a sex difference like this. One says it reflects something evolved and general — that the consequences of a sexual encounter have historically been asymmetric enough to shape different emotional responses. The other says it reflects social conditions — that women are punished more for casual sex, warned about it more, and taught to feel differently about it, so they do.
These make a testable prediction that differs. If it is social conditioning, then in a society where the conditioning is weakest, the difference should shrink or vanish.
A research team led by Leif Kennair and Mons Bendixen in Norway, working with the American psychologist David Buss, ran exactly that test. Norway is among the most gender-egalitarian, least religious and most sexually liberal societies on Earth — close to the best available real-world version of “remove the conditioning and see what happens”.
They asked 263 Norwegian students aged 19 to 37 about their most recent casual sexual encounter, or the most recent one they passed up, and how much they regretted it. A later study extended this to over five hundred Norwegians alongside an American comparison group.
What they found: the sex difference was not attenuated. It appeared in Norway at about the same size as in the United States. A follow-up comparing the two countries directly found that the cultural difference between them did not significantly change levels of sexual regret.
What this evidence is good at: it is a genuine prediction, made in advance, tested where the theory said it should fail. That is rare in this field.
What it is not: these are student samples in two Western countries. Norway is the strongest available test of the social explanation, not a perfect one — Norwegian women still live in a world with the rest of the world in it.
I want to be straight about this, because it cuts against the reading I would find more comfortable. The social-conditioning explanation made a clear prediction, the prediction was tested in the place most favourable to it, and it did not hold. That is real evidence, and it is the strongest single piece of evidence in this entire part that something here is not purely a matter of what women are taught.
4.4 — But look at what predicts it
Now the finding that reorganises the chapter, and it comes from the same research group.
Having established that the sex difference is robust, they asked a further question: within each sex, what predicts how much regret a person reports? The answers were not about sex at all.
The strongest predictors of regret after a casual encounter turned out to be worry — about pregnancy, infection, reputation — disgust, feeling pressured into it, and not having enjoyed it physically. A person's dispositional orientation towards casual sex mattered too.
Sociosexuality: a measured personality trait describing how comfortable a person is with sex outside a committed relationship. People at one end feel fine about it; people at the other find it distressing. It varies widely within both sexes.
Why it matters: in these studies, a person's sociosexuality predicted their regret better than their sex did. A woman comfortable with casual sex reported less regret than a man who was not.
And the researchers were explicit that the sex differences they found, while robust, were modest — and smaller than the differences produced by sociosexuality and by physical gratification.
That last item deserves a sentence of its own. One of the things most strongly predicting whether a woman regretted a casual encounter was whether she had enjoyed it. This is not a trivial observation, because the enjoyment gap in casual sex is itself large and well documented: research on American college students found women reporting orgasm far less often in first-time casual encounters than in established relationships, while men's rates barely moved.
Both sides now have strong evidence, which is unusual and makes this one of the more honest disagreements in the part.
The prediction was made in advance and tested in the most egalitarian society available, and it held. It is specific to the sexual domain and absent for non-sexual romantic regret, which rules out a general reporting difference. It replicates across research groups and countries. And there is a straightforward reason it might exist: for most of human history the possible consequences of a single sexual encounter were catastrophically different for the two sexes, and emotions that track consequences are exactly what you would expect to differ.
Look at what actually predicts regret: worry, pressure, disgust and lack of enjoyment. Every one of those is a condition, not a preference, and every one of them falls disproportionately on women — because the pregnancy risk is hers, the reputational risk is hers, the pressure is more often applied to her, and the physical enjoyment is measurably less likely. If women encounter more of the things that produce regret, they will report more regret, and no separate psychological difference is needed. Norway is more equal in law and attitude; it has not equalised who can get pregnant or who is more likely to be pressured or who has an orgasm.
What would settle it: comparing regret in encounters matched on worry, pressure and enjoyment. If the sex difference survives when all of those are equal, the first side is right. If it disappears, the second is. The measures exist; the study has not been done properly.
Why people care so much: because “women feel differently about sex” is doing heavy lifting in the wider argument this series is about. If it is fixed, the old rules can be presented as accommodating a fact. If it is produced by conditions, then the rules are partly producing the fact they claim to accommodate.
One last observation, which belongs to neither side. The warning says a woman will regret it. The research says some women regret some encounters, and that whether they do is predicted mainly by whether they were worried, pressured, or disappointed. The warning does not distinguish between a woman who regrets an encounter she was pushed into and a woman who does not regret one she chose. It treats those as the same event. The research does not.
Men more often regret what they did not do; women more often regret what they did, in the sexual domain specifically. The difference vanishes for non-sexual romantic regret, which is what makes it worth taking seriously.
The difference is moderate, not large, with heavy overlap. Knowing someone's sex shifts your guess a little and tells you almost nothing about the individual.
The social-conditioning explanation made a clear prediction — that it would shrink in an egalitarian society — and the Norwegian test did not shrink it. That is the strongest evidence in this part that something here is not purely taught.
But what predicts an individual woman's regret is not her sex. It is worry, feeling pressured, disgust, and whether she enjoyed it — and her own comfort with casual sex predicted her regret better than her sex did.
Every one of those predictors is a condition rather than a preference, and every one of them can change.
5The Mind
Does sex outside a committed relationship make women unhappy? The literature is genuinely mixed, which is itself informative — and one finding inside it reorganises the question entirely.
5.1 — A claim that sounds obvious to both sides
This is the claim most people feel certain about without having read anything. One side is sure that casual sex leaves women depressed, empty and damaged. The other is sure it is harmless and that the distress is manufactured by shame. Both are stating a research finding they have not checked.
The actual literature is a mess, and it is a specific kind of mess that tells you something. Studies find associations between casual sex and lower wellbeing. Studies find no association. Studies find casual sex associated with higher wellbeing. All of these are published, in reputable journals, by competent people.
When a field looks like that, there are two possibilities. Either the studies are too small and noisy to see a real effect, or the effect genuinely differs between people — which means the average is hiding the answer rather than revealing it.
Moderator: something that changes the size or direction of an effect for different people.
An example without any politics in it: alcohol has a moderator called body weight. The same two drinks affect a large person and a small person differently. If you studied “the effect of two drinks” across everybody and reported the average, you would produce a number that described nobody.
Why it matters here: this whole question turns out to have a strong moderator, and once you know what it is, the contradictory studies stop contradicting each other.
5.2 — The finding that sorts it out
In 2014 Zhana Vrangalova and Anthony Ong published a study asking not whether casual sex affects wellbeing, but for whom. They measured people's sociosexuality — the trait from Chapter Four describing how comfortable a person is with sex outside a committed relationship — and then looked at wellbeing in those who had casual sex and those who had not.
The result was clean. For people comfortable with casual sex, having it was associated with higher wellbeing. For people uncomfortable with it, having it was associated with lower wellbeing.
The effect did not have one direction. It had two, and which one you got depended on whether the behaviour matched the person.
Two women, same university, same week, same encounter with a stranger.
The first thinks casual sex is fine, has thought so for years, and would say so to her friends. Afterwards she feels good.
The second was raised to believe it is wrong, still believes it somewhere she cannot argue with, and did it anyway. Afterwards she feels terrible.
The act was the same. The outcome was opposite. What differed was the fit between what she did and what she believes — and where her beliefs came from is not something she chose.
This does not settle the argument. It relocates it, and the relocation is the interesting part. If harm tracks the mismatch between behaviour and belief, then the harm is real — the second woman is genuinely suffering, and telling her she should not be will not help. But the source of the harm is not the act. It is the collision between the act and a set of beliefs supplied by her upbringing.
Which means both sides can read this finding as a victory, and both readings are coherent. That is worth sitting with rather than resolving too quickly.
5.3 — Which way is the arrow pointing?
There is a further problem underneath all of this, and it is at least as serious as the confounding in Chapter Three.
Most studies in this area are cross-sectional: they measure the behaviour and the wellbeing at the same moment. That design cannot tell you which came first, and here the arrow plausibly runs both ways.
Distress leads to behaviour, not only the reverse. People who are depressed, lonely, drinking heavily, or in the aftermath of a bad relationship are more likely to seek out casual encounters — that is not a moral claim, it is a well-documented pattern. So a study that finds unhappy people having more casual sex has found exactly what you would expect if the unhappiness came first.
The better designs are longitudinal: measure wellbeing at one point, behaviour later, wellbeing later still. Where these have been done, the prospective link running from earlier distress to later casual sex is at least as strong as the link running the other way — and often stronger.
What this cannot show: it does not prove the behaviour has no effect. It shows that the raw association is substantially explained by an arrow pointing in the opposite direction from the one everybody assumes.
There is a broader factual correction that belongs here too, because the whole discussion is usually framed against a picture of the young that is simply wrong. In several rich countries, young people today report fewer sexual partners than the generation born in the 1950s and 1960s did at the same age, and are having less sex overall, later. The moral panic about a generation of hookups is being conducted about a generation that is, if anything, more cautious than its parents.
The evidence above is compatible with two quite different overall pictures.
The mismatch account does not exonerate the behaviour; it identifies who is at risk. A large share of women — in India, the overwhelming majority — hold exactly the beliefs that make the mismatch damaging, so for most women in the world the association with distress is not a statistical curiosity, it is a prediction about them. Nor is it obvious that the beliefs are the thing that should give way. The distress may be tracking something accurate: that an encounter without commitment carried a real risk, was often not enjoyable, and left her exposed in ways she correctly registered. Calling that “shame” and proposing to remove it is proposing to remove a signal rather than a cause.
If the same behaviour produces higher wellbeing in one woman and lower in another, the behaviour is not the cause; the collision is. And the beliefs on one side of that collision were installed, not chosen — by families, by communities, by exactly the warning this part is testing. That makes the warning partly self-fulfilling: it teaches a woman that this will damage her, and the teaching is a substantial part of what does. Meanwhile the arrow in the raw data mostly runs from distress to behaviour, not the other way. The strongest reading of the evidence is that unhappy people do this more often, and that the ones who suffer afterwards are the ones who were told they would.
What would settle it: longitudinal studies that track beliefs, behaviour and wellbeing over years, in more than one culture. The design is entirely feasible and the studies are mostly not being funded, because the question is politically radioactive in every direction.
Why people care so much: because the mismatch account puts the culture on trial alongside the behaviour. If the harm comes from the collision, then anyone who installed the beliefs is implicated in the damage — and that is an uncomfortable position for a parent who thought they were protecting their daughter.
Let me say plainly where I think the honest reader is left, because it is not where either camp would like.
The distress is real. Women who go against beliefs they hold feel worse, measurably, and this is one of the better-supported findings in the part. Whether that distress is evidence that the beliefs are correct, or evidence that the beliefs are costly, is not a question research can answer. It is the same question the whole series keeps arriving at, and Chapter Seven is where it finally gets a proper test.
The literature on casual sex and wellbeing is genuinely mixed — findings in all three directions, published by competent people. That pattern usually means an effect that differs between people rather than an effect that is not there.
It does. The strongest moderator is how comfortable the person is with casual sex. For those comfortable with it, wellbeing was higher; for those uncomfortable, lower. Same act, opposite outcomes.
The causal arrow largely runs the other way from the one assumed. Distress predicts later casual sex at least as strongly as casual sex predicts later distress.
And the background picture is wrong: young people in several rich countries report fewer partners than their parents' generation did at the same age.
The distress is real and should not be argued away. But it tracks the collision between what a woman does and what she was taught — and she did not choose what she was taught.
6The Body
Everything so far has been contested. This chapter is not. The physical claim is true, the mechanism is known, and the risk rises with the number. This is where the traditionalist case is strongest — and where it turns out to be aimed at the wrong person.
6.1 — Say the true thing first
More sexual partners means more exposure to sexually transmitted infection. This is not an association that might be confounded. It is a transmission mechanism. Each new partner is a new draw from a pool that may contain something, and more draws means more chances.
It is dose-responsive in exactly the way Chapter Two's divorce finding was not. It replicates everywhere, in every culture, because bacteria and viruses do not care about norms. And unlike everything in Chapters Two to Five, no amount of controlling for religion, personality or family background makes it go away.
So: on the body claim, the warning is right. I am not going to soften that, and a document that buried it would not be worth reading.
Human papillomavirus (HPV): a very common family of viruses passed on by skin-to-skin sexual contact. Most sexually active people acquire some type of it at some point, and in the large majority of cases the body clears it within a couple of years without the person ever knowing.
A small number of types are different. If one of those persists for years, it can change the cells of the cervix — the opening at the lower end of the womb — and those changes can become cancer.
Why it matters: persistent infection with these types causes virtually all cervical cancer. It is one of the very few cancers with a single known infectious cause, which means it is one of the very few that can be prevented outright.
6.2 — What this costs in India
India carries an enormous share of the world's cervical cancer. Roughly a fifth of global deaths from it occur here. The annual figures run to something in the order of 120,000 new cases and around 75,000 deaths — it is among the leading cancers in Indian women, and it kills them at ages when they are usually raising children.
Seventy-five thousand deaths a year is about two hundred Indian women a day, from a cancer that is caused by a virus we can vaccinate against and detect years before it becomes dangerous.
Put it against the thing this series is about. Every honour killing reported in India in a year, every one that goes unreported, every reputational catastrophe, every ruined engagement — put all of them together and the number is a fraction of this. The great physical harm attached to sex in Indian women's lives is not the one the warning talks about, and it is the one nobody warns about at all.
6.3 — What actually reduces it
Here is where the chapter turns, and it turns on evidence rather than on argument.
There are four things that reduce a woman's risk of cervical cancer and other infection-related harm. Three of them are medical and one is behavioural, and they are not equally powerful.
Vaccination against the high-risk HPV types prevents the great majority of cervical cancer if given before exposure. India now produces its own vaccine.
Screening finds the cell changes years before they become cancer, when treatment is simple and highly effective.
Barrier contraception and treatment reduce transmission of most sexually transmitted infections substantially. Condoms reduce HPV transmission but do not eliminate it, because it passes by skin contact rather than fluid. For chlamydia and gonorrhoea — which, untreated, can scar the fallopian tubes and cause infertility — testing and a course of antibiotics resolve the problem entirely.
Fewer partners reduces exposure. It genuinely does.
The relevant Indian number is not about behaviour at all. India's own National Family Health Survey asked women aged 30 to 49 whether they had ever been screened for cervical cancer. Around two in a hundred had.
That is the finding to hold on to. In a country where cervical cancer kills tens of thousands of women a year, and where a screening test can catch it a decade early, virtually no one is screened.
What this evidence is good at: it is a simple, direct, nationally representative count of a medical service, not a self-report about behaviour, so it carries none of the problems from Chapter One.
What it establishes: the gap between what is killing Indian women and what is being done about it is not a gap in their sexual conduct. It is a gap in access to two interventions that have existed for decades.
Both of those facts are agreed by everybody. What follows from them is not.
Everyone accepts the biology. The dispute is about what follows from it.
Everything else in this part is contested, and this is not. Exposure rises with partners; the mechanism is understood; the harm is severe and sometimes fatal. A woman with one lifetime partner has a far lower risk than one with twenty, and no vaccine is perfect, no screening programme catches everything, and no condom eliminates skin-to-skin transmission. If a grandmother's warning turns out to be right about the one thing that can actually kill you, that is not a small vindication. Public health bodies advise on partner numbers routinely for exactly this reason; only here does the advice get treated as an insult.
Compare the sizes. Vaccination prevents the large majority of cervical cancer regardless of behaviour. Screening catches nearly all of it early. Two per cent of Indian women in the relevant age group have ever been screened. If the goal is fewer dead women, the enormous lever is medical and the small lever is behavioural — and the behavioural one has been pulled for a thousand years while the medical ones sit largely unused. There is also a decisive fact about who is at risk: a woman's exposure depends on her partner's history as much as her own. Indian women with exactly one lifetime sexual partner, married, faithful, die of this disease every day. The rule cannot protect them, because it was never aimed at the person whose behaviour determines their risk.
What would settle it: nothing needs settling. This is one of the rare places in this part where the evidence is clear and the disagreement is entirely about what to do.
Why people care so much: because the health argument is the most respectable form of the warning, and therefore the one that gets used to carry the rest of it. A great deal of advice that is really about reputation arrives dressed as advice about infection.
6.4 — The fertility claim, which is a different claim
The last piece of the body claim is that a woman who delays will find she cannot have children when she wants them.
The underlying biology is real and is not disputed by anybody. Female fertility declines with age, gradually from the late twenties and more steeply from the mid-thirties, and the decline is not fully reversible by medicine. A woman who plans to have children at forty is taking a genuine risk that she will not be able to.
But look carefully at what that has to do with the warning. It is a claim about timing, not about number. A woman with fifteen partners who marries at twenty-six has lost nothing. A woman who has never had sex at all and marries at thirty-nine faces the full decline. The two variables have been quietly welded together in a way the biology does not support.
There is one genuine bridge between them, and it is the one from earlier in this chapter: untreated infection can damage the fallopian tubes and cause infertility, and infection risk does rise with partners. So there is a real path from partner count to fertility — and it runs entirely through an infection that a test and a course of antibiotics would have resolved.
On the body, the warning is right. More partners means more exposure to infection. It is dose-responsive, the mechanism is known, and no statistical control makes it disappear. This is the strongest part of the traditionalist case and it should be stated plainly.
Persistent infection with certain HPV types causes virtually all cervical cancer. India carries roughly a fifth of the world's deaths from it — in the order of seventy-five thousand a year, about two hundred women a day.
Four things reduce that risk: vaccination, screening, barrier protection and treatment, and fewer partners. The first two are far more powerful than the last. About two in a hundred Indian women aged 30 to 49 have ever been screened.
A woman's infection risk depends on her partner's history as much as her own. Married Indian women with one lifetime partner die of this disease. No rule aimed at her can protect them.
And the fertility claim is about timing, not number — those are different variables, and they have been welded together by a warning rather than by the biology.
7Does It Travel?
Part One promised a test that would separate a cost caused by an act from a cost caused by a punishment. Here it is, run on every finding in this part. The results do not sort neatly onto either side.
7.1 — The test, restated
In Chapter Five of Part One I set out a test and said the series would come back to it. This is where.
The portability test: if a harm is caused by the behaviour itself, it should show up wherever the behaviour happens — in every country, under every set of rules, in every religion.
If it is caused by the community's reaction, it should show up only where that reaction exists, and it should shrink or vanish where it does not.
Why it matters: this is the only tool available for telling a consequence from a penalty, and telling those apart is the whole question underneath this series. It is not a perfect tool. It is what there is.
The logic is simple. A virus does not know which country it is in. A community's contempt does. So if you find a harm that appears in Oslo and Patna and São Paulo at the same size, it probably belongs to the act. If you find one that is enormous in Patna and absent in Oslo, it probably belongs to the rule.
7.2 — Running it
The marriage-market penalty. This one is not close. Whether a woman's sexual history damages her marriage prospects varies from catastrophic to irrelevant depending entirely on where she lives, and within India it varies by state, city, caste and family. In some communities a rumour ends an engagement. In others nobody asks. There is no version of this that travels. It is a penalty, and it is administered by people.
Infection and cervical cancer. The opposite. Transmission works the same way in every country on Earth. The risk rises with exposure regardless of what anybody believes about it. This travels completely, and it is the clearest case in the part of a cost that belongs to the act.
The regret asymmetry. This is the difficult one, and I flagged in Chapter Four that it cuts against the reading I would find comfortable. The prediction was that it would shrink in an egalitarian, secular society. Tested in Norway, it did not shrink. On the face of it, it travels — which points towards something real rather than taught.
Wellbeing after casual sex. Here the test gives a clean answer in the other direction. The effect does not have a fixed direction at all: it depends on whether the woman believes the behaviour is wrong. Beliefs vary by community. So the harm travels with the belief, and the belief travels with the rule.
The divorce association. The honest answer is that we largely do not know, and that is itself informative. Nearly all the best evidence is American. The finding has not been established with anything like the same rigour in Europe, in East Asia, or in India. And within the American data it moved substantially as norms moved — the relationship between two partners and ten or more actually reversed across decades. A relationship that changes shape when the surrounding culture changes is behaving like a social process, not a mechanism.
Cross-cultural comparison is the strongest tool in this chapter and it has a specific weakness worth naming.
What it does well: when the same study design is run in two societies with genuinely different norms and produces the same result, that is real evidence about the act. The Norwegian regret work is a good example — a prediction made in advance, tested where the theory said it should fail.
What it does badly: there is no society on Earth without rules about women's sexual behaviour. Norway is less strict than India. It is not a control group. Norwegian women still face pregnancy risk, still encounter pressure, still have friends and mothers and an internet. So “it appeared in Norway too” means “it survived a weaker version of the rule”, not “it survives the absence of the rule”.
What this means in practice: the portability test can strongly identify things that belong to the rule. It can only weakly identify things that belong to the act, because the untreated condition does not exist anywhere.
7.3 — The scoreboard
Putting it together.
| The claimed harm | Does it travel? | Best reading |
|---|---|---|
| Infection, cervical cancer | Completely | Belongs to the act. Reduced far more by vaccination and screening than by conduct. |
| Regret asymmetry | Yes, so far as tested | Direction robust, size moderate. But the predictors of an individual's regret are conditions — worry, pressure, enjoyment. |
| Reduced wellbeing | No — it reverses | Belongs to the collision between behaviour and installed belief. Direction depends on the woman. |
| Marital instability | Unknown, and unstable within one country | Association real in American data; no mechanism found; moves as norms move. |
| Marriage-market damage | Not at all | Belongs entirely to the community. A penalty, administered by people who can stop. |
That is the intellectual result of this part, and it is untidy in a way I did not expect when I started. The warning is not simply false. One of its claims is straightforwardly true and physical. One is robust and modest. Two belong to the rule rather than the act. One is unexplained.
Anybody telling you this is all made up is wrong. Anybody telling you the research vindicates the warning is also wrong. The findings sort into different bins, and which bin a claim falls into changes what could be done about it.
Before leaning on that table, the tool itself deserves scrutiny.
Comparison across societies is how anthropology, epidemiology and economics all separate the universal from the local, and it is not controversial in any of those fields. When a finding reverses direction depending on what a woman believes, that is about as clean a demonstration as social science produces that the belief is doing the work. And the test has integrity precisely because it does not always favour one side: it delivered a result for the traditionalist case in Chapter Six and another in Chapter Four. A tool that only ever produces the answer you wanted is not a tool.
There is no society without these rules, so the comparison is always between strict and less strict, never between rule and no rule. Worse, a genuine act-caused harm might only become visible under certain conditions and would then look local when it is not — and a rule-caused harm might be near-universal because the rule is near-universal, and would then look like a fact of nature. Norway is not a laboratory. It is a slightly different arrangement of the same species under a slightly different version of the same rule.
What would settle it: a society with genuinely no norms governing women's sexual behaviour. None exists, none has existed in the historical record, and the absence of even one is itself a piece of evidence — though what it is evidence of is exactly what Parts Two and Six of this series spent their length arguing about.
Why people care so much: because this test is the hinge. If a harm belongs to the act, the rule is at worst clumsy protection. If it belongs to the rule, then the community is causing the damage it points at as justification. Almost everything else in the argument is downstream of which of those is true.
There is something both sides of that argument are taking for granted, and it is the thing an actual woman would notice first.
This entire chapter has been sorting harms into two bins: caused by the act, caused by the punishment. The traditionalist says the cost is real, so be careful. The reformer says the cost is imposed, so it is unjust. They are having a genuine and important disagreement.
Now consider a twenty-year-old woman in a district where the punishment is severe. What does the distinction give her?
Nothing. There is no version of her behaviour available to her that comes without the punishment attached. She cannot select the Norwegian package. The cost that reaches her is the sum of both bins, and it arrives as one number. Being told that seventy per cent of it is socially constructed does not reduce it by seventy per cent.
So the distinction is real, important, and useful only to someone deciding what a society should do — a legislator, a reformer, a parent choosing what to teach, a writer producing a document like this one. It is close to useless to the person living inside the situation. And the argument is conducted almost entirely by people in the first group, about people in the second.
The general form: a distinction that is real at the level of causes and inert at the level of the person. It runs through every applied field. Whether a patient's pain is physical or psychological in origin matters enormously to the doctor and not at all to the patient, who hurts either way. Whether unemployment is structural or personal matters to the economist and not to the man without work.
Why I am flagging it here rather than in the last chapter: because Chapter Nine is going to ask what a person should do with all this, and this box is the reason my answer there is not the one either side wants.
That is the state of the evidence, sorted as honestly as I can sort it. The next chapter asks what anybody is supposed to do with it.
The portability test: a harm caused by the act should appear everywhere; a harm caused by the punishment should appear only where the punishment does.
Run on the five claims, the results split. Infection travels completely — it belongs to the act. Marriage-market damage does not travel at all — it belongs entirely to the community. Wellbeing effects reverse direction depending on what the woman believes, so they belong to the rule. The regret asymmetry travels, which is the strongest evidence in this part for something real and untaught. And the divorce association is essentially untested outside America and moves as norms move.
So the warning is neither vindicated nor made up. Its claims fall into different bins, and the bin determines what could be done.
The test is asymmetric: strong evidence when something fails to travel, weak when it does — because there is no society without rules about women, so there is no control group.
And for the woman living inside a strict community, the distinction does not help. She cannot buy the version of her own behaviour that comes without the punishment attached.
8India
Everything in the last six chapters was measured mostly in America and Northern Europe. This chapter is about the place where the penalty is largest, best organised, and least studied — and about a count that nobody keeps.
8.1 — A different order of magnitude
The research in Chapters Two to Five was conducted in societies where the consequence of a woman's sexual history is, at worst, some private unhappiness and a somewhat higher chance of divorce. Read those chapters and it is easy to conclude the whole subject is a fuss about effect sizes of eight per cent.
In India the consequence can be the end of a marriage prospect, expulsion from a household, a beating, or death. That is not the same subject with the numbers turned up. It is a different phenomenon, and almost none of the research in this part was designed for it.
Izzat: usually translated as honour, which is misleading. It is closer to standing — a family's public credit, held collectively, that determines the marriages it can arrange, the deals it can make and how it is treated.
The crucial feature: it is held by the household, not the person, and it is far easier to lose than to build. One daughter's rumoured conduct can affect the marriage prospects of her sisters, her cousins and her brothers.
Why it matters: this is why the pressure on an Indian woman is not primarily moral. It is economic and collective. Her family is not policing her soul; it is protecting an asset that her behaviour can destroy and that she does not own a share of.
8.2 — The market that actually exists
To understand why sexual history carries the weight it does in India, you have to see what marriage still is here.
The great majority of Indian marriages are arranged by families rather than chosen by the couple. A substantial share of women report meeting their husband for the first time at or shortly before the wedding. Marriage outside one's caste remains rare — on the order of one marriage in twenty. And surveys of Indian attitudes have found that around two-thirds of respondents regard it as very important to stop women in their community from marrying outside it, with similar figures across religions.
In that system, a woman's marriage is a transaction between two families, and what is being exchanged includes a claim about her. A rumour is not gossip. It is a defect discovered in the goods, and the market prices it.
Consider the arithmetic from the family's side, stated as coldly as they would never state it.
A household with three daughters is arranging three marriages. Each match depends on the family's standing. If a rumour attaches to the eldest, the other two matches become harder and more expensive — dowry demands rise, the pool of acceptable families shrinks.
So the family is not making a judgement about one daughter's conduct. It is making a calculation about three futures, and the daughter in question is one input into it. That is why the response is so violently out of proportion to the act: the act is small and the exposure is large.
It is also why the pressure does not lift when a family becomes more educated or richer. Money does not remove you from the market. It raises the value of what is at stake in it.
8.3 — The examination
Where a claim about a woman's history has commercial value, someone will try to verify it. India has two documented forms of this, and both are worth naming precisely.
The first is community virginity testing: the practice, documented in some Indian communities, of examining a bride after the wedding night and reporting the result to the family or the caste council. It has been challenged from inside — a campaign led by young members of one Maharashtra community brought the practice to national attention in recent years and led to arrests.
The two-finger test: a manual examination once performed on women reporting rape, in which a doctor recorded whether the vagina admitted two fingers, and drew conclusions about whether the woman was “habituated to sexual intercourse”.
It has no scientific validity of any kind. It cannot establish sexual history, and the hymen is not a record of anything. Its function in a courtroom was to shift attention from what the accused did to what the complainant had done before.
Where it stands now: India's Supreme Court held in 2013 that the test violates a woman's dignity and bodily integrity, and in 2022 went further, holding that anyone who conducts it is guilty of misconduct and directing that it be removed from medical curricula. In 2018 the World Health Organization, together with UN Human Rights and UN Women, called for the elimination of virginity testing worldwide, noting it was still practised in at least twenty countries.
I include this because of what it reveals about the structure. If a woman's sexual history were simply a private matter with private consequences, there would be no reason to examine her, and certainly no reason for a court to have to ban the examination twice. Institutions do not build verification procedures for things that do not have value. The test existed because the information had a price.
8.4 — The count nobody keeps
At the far end of this system is lethal violence against women — and men — who marry or partner against a family's wishes.
India's National Crime Records Bureau publishes an annual count of crimes, and it includes a category for killings connected to honour. The recorded figures are very low — typically a few dozen a year for a country of 1.4 billion people.
Almost nobody who studies this believes that number. Researchers and civil society organisations who have gone through case files and press reports consistently arrive at figures far higher. The reason is in how the deaths get recorded: they are entered as suicides, as kitchen accidents, as ordinary murders with no motive attached, or they are not reported at all — because the people who would report the crime are frequently the people who committed it.
Now treat that absence as evidence rather than as a gap. India runs a large, functioning statistical apparatus. It counts births in sample villages twice over. It surveys hundreds of thousands of households on their health. The capacity exists.
So a state that can count almost anything is producing a figure for this that its own researchers do not accept. That is not a measurement failure. An institution that can count and does not has made a choice, and the choice is the finding.
With that on the table, the central Indian disagreement can be stated properly. It is not about whether the penalty exists — both sides agree it does. It is about what the system is for.
This is the most delicate argument in the part, and both positions are held sincerely by enormous numbers of people.
Given the world as it actually is, a family that restricts a daughter is reducing her exposure to a genuine catastrophe. The penalty is real — the ruined match, the lifelong precarity of an unmarried woman in a society with no independent place for her, the violence. A parent who says “do not do this, you will be destroyed” is not inventing the destruction. They are reporting it. In a country where a woman's economic security still runs almost entirely through marriage, arranging a good one is the single most valuable thing a family can do for her, and protecting the conditions for it is care, not commerce. Calling this an asset trade insults people who are frightened for their children.
Test it against the mechanism rather than the intention. If the object were her safety, the response to danger would be to shield her. Instead, when a violation occurs or is merely rumoured, the response is very often to punish her — to confine her, to marry her off urgently, to cast her out, in the worst cases to kill her. Protection that responds to harm by destroying the person it protects is not protection. The pattern also tracks the wrong variable: the controls tighten where the family's standing is most exposed, not where her risk is highest. And the clearest test of all is what happens to a woman who is assaulted rather than willing. If this were about her welfare, she would receive maximum support. She frequently receives the opposite, because what was damaged was the asset, and the asset does not care how it was damaged.
What would settle it: systematic data on how families respond when the two interests diverge, particularly in cases of assault. Some exists in case studies and legal records; nothing at national scale, and see the box above for why.
Why people care so much: because almost every Indian reading this is describing their own parents, or themselves. That is not a reason to soften the finding. It is a reason to state it carefully.
That argument turns on something both sides accept without examining, and it is the one I want to leave this chapter on.
Every defence of the system says it is for her. Every critique argues about whether it works for her. Both are treating her welfare as the thing the rule is aimed at, and then disputing the effectiveness.
But a rule's purpose is not what it says. It is what it optimises for, and you can read that off its behaviour. Look at where these controls are tightest and where they relax. They are tightest where the family's marriage-market exposure is greatest — a daughter of marriageable age, with unmarried sisters, in a community that talks. They relax when she is no longer on the market, or when the family's standing is secure by other means, or when she is far enough away that nobody local will hear.
Now notice what that pattern is not tracking. It is not tracking her physical risk. It is not tracking her health, her happiness or her chance of ending up with a man who treats her well. If it were, the controls would tighten around the dangerous marriage and relax around the safe boyfriend, and in practice they do the reverse.
The general form: reading a rule's stated beneficiary as its actual one. Employment rules that say they protect workers and track employers' liability. Licensing that says it protects consumers and tracks incumbents' competition. In each case the stated beneficiary is real, is often genuinely helped, and is not what the rule is organised around — and you can tell, because you can watch what the rule does when the two come apart.
The reason this matters more than the usual version: the people applying these rules are not cynics. An Indian parent restricting a daughter is not privately thinking about assets. They love her and they are frightened. The mechanism does not need anybody to understand it in order to run — which is exactly why it is worth writing down.
None of this is a description of bad people. It is a description of a machine that good people operate without seeing the shape of it, which is the only kind of machine that lasts.
The findings in Chapters Two to Five were measured where the penalty is small. In India it is an order of magnitude larger, and almost none of that research was designed for it.
Izzat is not honour, it is standing — a household asset, held collectively, that one daughter's rumoured conduct can damage for her siblings and cousins. That is why the reaction is out of proportion: the act is small and the exposure is large.
Most Indian marriages are still arranged, inter-caste marriage runs at about one in twenty, and around two-thirds of surveyed Indians say it is very important to stop women marrying outside the community. In that market a claim about a woman's history has a price — which is why verification procedures exist, and why the Supreme Court has had to ban the two-finger test twice.
India records a few dozen honour killings a year and almost no researcher believes the figure. A state that can count and does not has made a choice, and the choice is the finding.
The controls track the family's exposure, not the daughter's risk — and you can see it in the one case where the two come apart, which is what happens to a woman who was assaulted rather than willing.
9What a Person Could Do With This
Eight chapters of findings. Now the question everybody actually wants answered — so what should she do? I am going to give the traditionalist case its strongest form, lay out three answers, and then explain why I am refusing to pick one.
9.1 — The case for the warning, put as well as I can put it
Here is the argument for the traditional advice, made without a single appeal to religion or tradition. It is stronger than most people who dismiss it realise, and it uses this part's own findings.
Start with what survived. The health claim is true and physical. The regret asymmetry survived the hardest cultural test anybody has run. The divorce association has been found for ninety years across many research teams. That is not nothing, and a person who acts as if all of it were invented is acting against the evidence.
Then take the costs that are socially imposed. Chapter Seven established that the marriage-market penalty belongs entirely to the community. Now notice that this does not help her at all. A penalty that is unjust still lands. A woman in a district where a rumour ends an engagement faces exactly the same outcome whether the penalty is written into nature or into her neighbours. She cannot decline to live in her society, and advising her as though she could is not liberation, it is abandonment.
A prudential reason is a reason of self-interest: do this because it will go better for you.
A moral reason is a reason about right and wrong: do this because it is what you owe.
Why the distinction matters here: the traditional warning is almost always delivered as a prudential reason — I am telling you for your own good — while functioning as a moral one. The two get conflated constantly, in both directions. Someone who intends a moral instruction reaches for the prudential language because it is harder to argue with. And someone rejecting a moral instruction often answers the prudential version instead, which is a different argument.
Add the asymmetry of consequences. If the warning is wrong and she heeds it, she has lost some experience. If the warning is right and she ignores it, in the strictest environments she may lose her marriage prospects, her family, or her life. Those are not symmetrical, and under genuine uncertainty a rational person weights the tail.
And finish with humility about the evidence. Every chapter in this part reports uncertainty. In conditions of uncertainty, an inherited practice that many generations have converged on deserves some weight — not because old is good, but because a rule that survived a long time may encode observations that were never written down. Demanding a peer-reviewed effect size before respecting accumulated practical knowledge is its own kind of error, and this series named it in Part One.
That is the case. I do not think it is decisive, but anyone who cannot state it in that form has not earned the right to reject it.
9.2 — Three answers
Every position below is held by serious people who have looked at the same evidence.
Describe the world as it is. She lives in a specific place with specific penalties, and the useful thing is accurate information about them plus a recommendation that minimises her risk. Withholding that so as not to endorse the system is a luxury belief: the person paying for it is her, and the person feeling good about it is you. Doctors advise patients to avoid genuinely dangerous behaviour without first litigating whether the danger is socially constructed. This is the same, and pretending otherwise costs real women real outcomes.
The marriage-market penalty exists because it is expected to be obeyed. Every woman who complies confirms that the threat works, which sustains it for the next one. Advice that adapts to an unjust system is one of the mechanisms by which the system persists — it converts a punishment into common sense, and common sense needs no enforcer. And this counsel of prudence has been offered in every generation about a world that then changed, usually because people stopped taking it.
Both of the above accept that the only variable is her behaviour. But look at what this part actually found. Regret is predicted by worry, pressure and lack of enjoyment. Wellbeing damage is predicted by the collision with installed belief. Death from cervical cancer is prevented by vaccination and screening. Not one of those is fixed by a woman having fewer partners, and every one of them is addressable — contraception, a screening programme, protection from coercion, and not installing the belief in the first place. The whole framing of “what should she do” has already conceded that she is the only adjustable part of the system.
What would settle it: nothing empirical. This is the point in the series where the evidence genuinely runs out and a value judgement has to be made by whoever is making it.
Why people care so much: because the first answer is what most Indian parents say, the second is what most of the young people arguing with them say, and the third is what almost nobody says, because it requires spending public money and changing institutions rather than instructing a girl.
9.3 — Why I am not picking one
I said in the front matter that nothing here would be advice, and that refusing was a position rather than modesty. Here is the reasoning.
The strongest single finding in this whole part is a moderation finding: the effect of the behaviour on a woman's wellbeing depends on what she herself believes and wants. That is not a caveat. It is the result. And a result of that shape does not license a general recommendation, because there is no general person to make it to.
Put the whole part into one woman's situation, and watch the answer change as she changes.
A woman in a metropolitan Indian city, financially independent, secular, in a family that would not disown her: nearly every cost documented here is small for her, and the largest real one is infection, which is addressed by a vaccine and a screening test.
A woman in a district where a rumour would end her sister's engagement, dependent on her family for everything, who herself believes what she was taught: nearly every cost documented here is large for her, and none of them is addressed by anything she can do alone.
The evidence in this part is identical in both cases. The answer is not, and anyone giving one answer to both is not reporting research — they are expressing a preference.
The second reason is about who is speaking. Chapter Eight described a system in which a woman's conduct is decided, discussed and adjudicated mostly by other people. I am, at this moment, another person discussing it. The one contribution available to me that does not simply add to that pile is to hand over the information disaggregated — this cost is real and physical, this one is imposed by your neighbours, this one depends on what you believe, this one nobody can explain — and let whoever is living the situation do the applying. That is a smaller deliverable than a recommendation. It is the one I can honestly produce.
9.4 — Where I have to be careful
The front matter said I would flag this at the point where it applies.
Writing Chapter Eight, I noticed something in myself worth reporting. It was easier to write the section on the two-finger test than the section defending the family, and the ease was suspicious. Condemning a banned medical practice costs a writer nothing and reads as courage. Stating fairly why a frightened Indian parent restricts a daughter — and taking seriously that they may be right about the world even if wrong about the remedy — is the part that requires actual effort, and it is the part I was tempted to hurry.
There is a matching temptation in the other direction, and I want to name it too, because I am writing for an audience that includes many people who agree with the warning. It would be very easy to end this part with a sentence that lets the traditional view feel vindicated by the health chapter, because the health chapter is the one place I could do that honestly. I have tried not to let Chapter Six carry more weight than its own findings support.
Whether I have managed either is not something I can judge from inside the document. What I can do is tell you the two directions the error would run, and note that both temptations were present.
This whole part has treated the warning as a prediction. I unpacked it into testable claims in Chapter One, checked them in Chapters Two to Six, ran a portability test in Chapter Seven, and have just spent this chapter asking what follows if the prediction is partly true.
That framing assumes the warning is in the business of predicting. Consider the possibility that it is not.
Here is the test. Suppose every effect size in this document came back at exactly zero tomorrow — no divorce association, no regret difference, no wellbeing effect, and a vaccine that eliminated the health risk entirely. Would the warning stop being issued?
Obviously not. Not in India, not anywhere. The aunt would still say it. The internet would still say it to the woman in the photograph who started this series. Which tells you what the warning is: it is a norm, and it wears a prediction the way a claim wears evidence when evidence is the currency of the room. The predictive dress is not the thing. It is what the thing puts on to be argued with.
So I have spent ninety pages auditing the one component that was auditable, and reporting the result as though the result were the point. That is not useless — a reader now knows which costs are physical, which are imposed, and which nobody can explain, and that is genuinely worth having. But I should not pretend it settles anything with anyone, because the part I checked is not the part doing the work.
The general form: auditing the testable component of something whose function is not to be tested. It happens constantly. A policy justified by an economic forecast that its supporters would back regardless of the forecast. A dress code justified by safety. A rule justified by evidence that arrived long after the rule. In each case the evidence is genuine and detachable, and checking it is worth doing — as long as nobody mistakes checking it for having addressed the rule.
What actually addresses the rule is what Parts Two, Three and Six of this series were for: where it came from, what it was built to do, and what its defenders are really defending. This part checked the receipt. Those parts were about the purchase.
So this is where Part Ten ends: with the findings sorted, the advice withheld, and an admission that the sorting may matter less than it looks.
The traditionalist case at full strength does not need religion: the health claim is true, the regret asymmetry survived the hardest test, the socially imposed costs still land on her, and under uncertainty the downside of ignoring the warning is heavier than the downside of heeding it.
Three answers exist to “what should she do”, and they agree on all the evidence. They differ on who should bear the cost of a system that is unjust and real at once. That is a moral disagreement, and no study will settle it.
The strongest finding in this part is a moderation finding — the effect depends on what she herself believes and wants. A result of that shape cannot license a general recommendation, because there is no general person to give it to.
So I am handing the findings over sorted rather than summarised into advice: this cost is physical, this one is imposed by your neighbours, this one depends on what you were taught, this one nobody can explain.
And the assumption underneath the whole document: I have treated the warning as a prediction. If every number in here came back zero tomorrow, nobody would stop issuing it — which means the part I checked is not the part doing the work.
10An Honest List of What We Do Not Know
Two lists, as in every part. What is genuinely unknown, with the reason it is unknown — which is usually more interesting than the gap. And what is solid enough to build on.
10.1 — Genuinely unknown
What the divorce association is made of
The association between reported premarital partners and marital dissolution has been found for ninety years. Nobody can currently say what produces it. The researcher who published the best-known figures tested the standard explanations — comparison with past partners, children outside marriage, poor mate choice, selection on personality — and reported that none of them accounted for it. The reason we do not know: the study that would answer it cannot be run, and everything else is a substitute.
Whether any of this is true of men
The largest gap in the field, and it is a gap in what was collected rather than in what was found. The survey behind the headline divorce numbers does not carry equivalent data on men's premarital behaviour. So the comparison that would tell you whether this is a fact about sexual history or a fact about women's sexual history has never been available. The reason we do not know: nobody thought to ask.
Whether the divorce finding exists outside America
Nearly the entire evidence base is American. It has not been established with comparable rigour in Europe, East Asia or India. Given that the American finding itself moved substantially as American norms moved, this is not a small gap. The reason we do not know: the surveys that would support it either do not ask, or do not ask consistently enough across decades.
Whether the regret asymmetry survives matched conditions
The sex difference is robust in direction and moderate in size. The proximate predictors identified so far — worry, pressure, disgust, physical enjoyment — are conditions rather than dispositions, and they are unequally distributed. Nobody has compared regret in encounters matched on all of them. The reason we do not know: the study is entirely feasible and has not been done.
Which way the wellbeing arrow runs
Distress predicts later casual sex at least as strongly as casual sex predicts later distress. How much of the raw association is each is unsettled. The reason we do not know: most studies measure everything at one moment, and the long-term studies that would separate them are expensive and politically unattractive to fund from any direction.
How much lethal violence there actually is in India
The official count is a few dozen a year and almost no researcher accepts it. The true figure is unknown within an order of magnitude. The reason we do not know is the interesting part, and Chapter Eight set it out: the deaths are recorded as something else, often by the people involved.
What women actually do
Every behavioural number in this part is self-reported, and men's and women's totals cannot both be true. The direction of the error tracks exactly the social pressure being studied. The reason we do not know: there is no way to independently verify a life history, and there never will be.
Whether beliefs can be changed without cost
Chapter Five found that the damage tracks the collision between behaviour and installed belief. That suggests the beliefs are a lever. It does not establish that removing them is free — a person who loses a framework may lose things attached to it. The reason we do not know: nobody has studied what happens to women who change their minds about this, as opposed to women who hold one view or the other.
One absence on that list is permanent rather than temporary, and it is worth stating precisely because it changes how everything else should be read.
The design that would settle nearly every question in this part is a randomised trial: assign young women at random to different numbers of sexual partners, follow them for twenty years, compare. That study will never be run, in any country, by anyone. It is not merely unfunded. It is unethical in a way that no review board or change of politics would ever alter.
What follows from that: every finding in Chapters Two to Five is observational, permanently. Sibling comparisons, controls, longitudinal tracking and cross-cultural tests are the best available substitutes, and they are substitutes. Anyone claiming certainty about causes here — in either direction — is claiming something the evidence cannot deliver and never will.
Why this counts as knowledge rather than a complaint: knowing that a question is permanently underdetermined is itself a finding, and it tells you how much weight any answer can bear.
10.2 — Solid
The reported partner counts cannot all be right
In every large national survey, men report substantially more opposite-sex partners than women — often around double. In a closed population the two totals must be equal. This is arithmetic, not interpretation, and it has been replicated across countries and decades.
The divorce association exists in American data
Reported premarital partner count and five-year marital dissolution are related in the National Survey of Family Growth. Found since 1938, replicated by many teams. Whatever its cause, the pattern is there.
The relationship is not a line
Women with two premarital partners divorced more than women with three to nine, and in earlier decades more than women with ten or more. About half the whole range sits between zero partners and one. These are features of the published figures, not of anyone's reading of them.
Premarital sex with only the future husband carries no elevated risk
Established by Teachman in 2003 and the cleanest single result in the field. Whatever the association tracks, it is not the act.
Sex differences in sexual regret are real, moderate, and not attenuated in Norway
Men report more inaction regret, women more action regret, specifically in the sexual domain and not in non-sexual romantic matters. The prediction that it would shrink in a highly egalitarian society was made in advance and tested. It did not shrink. The size is moderate, with heavy overlap between the sexes.
An individual's regret is predicted by conditions
Worry, feeling pressured, disgust and lack of physical enjoyment predict regret better than sex does. A person's own comfort with casual sex predicts their regret better than whether they are a man or a woman.
The wellbeing effect is moderated and reverses
For people comfortable with casual sex, having it was associated with higher wellbeing; for those uncomfortable, lower. Same behaviour, opposite outcomes, depending on the person.
Infection risk rises with exposure, and cervical cancer is preventable
This is biology, it is dose-responsive, and it holds in every society. Persistent infection with certain HPV types causes virtually all cervical cancer. Vaccination prevents the large majority of it; screening catches nearly all of the rest early. Roughly two in a hundred Indian women aged 30 to 49 have ever been screened.
A woman's infection risk depends on her partner's history as well as her own
Married women with a single lifetime partner develop and die from cervical cancer. No rule governing only her conduct can prevent this, because her exposure is not determined only by her conduct.
The two-finger test has no validity, and India has banned it twice
It cannot establish sexual history. The Supreme Court held in 2013 that it violates dignity and bodily integrity, and in 2022 that conducting it is misconduct. In 2018 the World Health Organization and two United Nations bodies called for the elimination of virginity testing worldwide.
Unknown: what the divorce association is made of, whether it is true of men at all, whether it exists outside America, whether the regret difference survives matched conditions, which way the wellbeing arrow runs, how much lethal violence India actually has, and what women actually do — because the core variable is self-reported and provably misreported.
Solid: the reported counts cannot all be true; the divorce association is real in American data and is not a straight line; premarital sex with only the future husband carries no penalty; the regret asymmetry is real, moderate and survived Norway; individual regret is predicted by conditions rather than by sex; the wellbeing effect reverses depending on the person; infection risk is real and dose-responsive; cervical cancer is preventable and almost nobody in India is screened; and a woman's risk depends on her partner's history as much as her own.
One absence is permanent. The experiment that would settle the causal questions is one nobody will ever ethically run, which means every finding here is observational forever.
So the honest summary of ninety years of research: one claim in the warning is physically true and addressable by a vaccine, one is real and modest, two belong to the community rather than to the act, and one has never been explained by anybody — including the people who found it.
Sources & further reading — Part 10
Timeline
Ninety years of research into a question that has almost only ever been asked about women, and the Indian legal thread running alongside it.
| Year | What happened |
|---|---|
| 1938 | Lewis Terman publishes the first systematic finding that premarital sexual experience is associated with marital instability. Terman was a pioneer of psychological testing and an enthusiastic eugenicist; the question did not enter the literature neutrally. |
| 1948–53 | The Kinsey reports establish that reported sexual behaviour and publicly stated norms differ enormously, and that large surveys can measure it at all. |
| 1960 | The contraceptive pill is approved in the United States, separating sex from pregnancy risk for the first time at scale — and changing one of the conditions this literature keeps rediscovering. |
| 1979 | The Mathura case verdict in India, in which a young woman's alleged prior sexual conduct is treated as relevant to whether she consented, triggers a national campaign. |
| 1983 | India amends its criminal law in response, introducing a presumption of absence of consent in custodial rape cases. |
| 2003 | India removes the provision that had allowed a rape complainant's general character to be used against her in court. In the same year Jay Teachman publishes the finding that women whose only premarital partner was their future husband carry no elevated divorce risk. |
| 2006 | Neal Roese and colleagues publish the first systematic demonstration of sex differences in sexual regret: men regret inaction, women regret action. |
| 2012 | Research on American students documents the large gap between women's reported physical enjoyment in casual encounters and in relationships. Sibling-comparison work on age at first sex begins attenuating associations that had looked strong in raw data. |
| 2013 | Galperin and colleagues replicate the regret asymmetry and show it is absent for non-sexual romantic regret. India's Supreme Court holds in Lillu v. State of Haryana that the two-finger test violates a woman's dignity and bodily integrity. |
| 2014 | Vrangalova and Ong show that the effect of casual sex on wellbeing is moderated by the person: higher for those comfortable with it, lower for those not. |
| 2016 | Kennair, Bendixen and Buss test the regret asymmetry in Norway and find it not attenuated. Wolfinger publishes the partner-count divorce table and the non-monotonic curve. |
| 2017 | A direct United States–Norway comparison finds that the cultural difference between the two countries does not significantly change levels of sexual regret. |
| 2018 | The World Health Organization, UN Human Rights and UN Women jointly call for the elimination of virginity testing, noting it is still practised in at least twenty countries. A campaign led by young members of a Maharashtra community brings community virginity testing to national attention in India. |
| 2021 | A large survey of Indian religious attitudes finds around two-thirds of respondents saying it is very important to stop women in their community from marrying outside it. |
| 2022 | India's Supreme Court holds in State of Jharkhand v. Shailendra Kumar Rai that anyone conducting the two-finger test is guilty of misconduct, and directs its removal from medical curricula. |
| 2023 | Wolfinger tests the standard explanations for the divorce association and reports that none of them accounts for it. India begins deploying a domestically produced HPV vaccine. |
| Next | Part Eleven starts here. |
Glossary
Every hard word used in this part, in plain English.
| Term | What it means |
|---|---|
| Action regret | Regret about something you did. The counterpart of inaction regret, which is about something you did not do. |
| Association | Two things going up and down together. It lets you guess a little better about one if you know the other. It says nothing about cause. |
| Cross-sectional study | A study that measures everything at one moment. Cheap, common, and unable to tell you which thing came first. |
| Confounding | When a third factor causes both of the things you are comparing, producing an association that is real in the data and empty as a cause. |
| Dose-response relationship | More of the cause, more of the effect, steadily. One of the strongest signs a relationship is genuinely causal. |
| Effect size | How big a difference is, as opposed to how confident we are that it exists. “Significant” means probably not zero; it does not mean large. |
| HPV | Human papillomavirus. A common family of viruses passed by skin-to-skin sexual contact. Persistent infection with certain types causes virtually all cervical cancer. |
| Izzat | Usually translated as honour; closer to standing. A household's public credit, held collectively, which determines the marriages it can arrange. |
| Longitudinal study | A study that follows the same people over years, measuring things in sequence. Expensive, slow, and the only way to see which came first. |
| Moderator | Something that changes the size or direction of an effect for different people. If an effect has a strong moderator, the average describes nobody. |
| NFHS | India's National Family Health Survey. A large repeated household survey, and the only Indian source that breaks results down by religion, caste and education. |
| Non-monotonic | A relationship that goes up and then down, or the reverse. Usually a sign that several different groups are mixed together. |
| NSFG | The American National Survey of Family Growth. The source of nearly every headline finding on premarital partners and divorce. |
| The portability test | If a harm is caused by an act it should appear everywhere; if by a punishment, only where the punishment exists. The main tool for telling a consequence from a penalty. |
| Prudential reason | A reason of self-interest — do this because it will go better for you. Distinct from a moral reason, though the two are constantly conflated. |
| Reverse causation | When the effect is actually causing the supposed cause, or when both come from the same underlying situation. |
| Selection effect | When people ended up in a group for reasons that also affect the outcome, so you are comparing different kinds of people rather than the same people under different conditions. |
| Sibling-comparison design | Comparing brothers or sisters raised in the same household who differ in the behaviour studied. It removes every family-level confounder automatically, including unmeasured ones. |
| Social desirability bias | Giving the answer that makes you look the way you are supposed to look. Here it pulls in opposite directions for men and women on the same question. |
| Sociosexuality | A measured trait describing how comfortable a person is with sex outside a committed relationship. In these studies it predicted regret better than sex did. |
| The two-finger test | A discredited manual examination once used to suggest whether a woman was “habituated to sexual intercourse”. It has no validity and has been prohibited by India's Supreme Court. |
| Virginity testing | Examination of a woman to determine whether she has had sex. Scientifically impossible; called for elimination by the World Health Organization and two UN bodies in 2018. |