Emily Oster
"No amount has been proven safe" is a logic error, not a scientific finding

1. One strange thing

In early 2025 the U.S. restricted the food dye Red No. 3. People came to Emily Oster to mourn a childhood: Swedish Fish would never be that red again.

Swedish Fish contain no Red No. 3.

They contain Red No. 40, which was not restricted. Except in California, where from 2028 Red No. 40 may not be used in food served at schools. Food schools sell as a fundraiser is exempt.

She copied the whole chain out as it stood, and then wrote:

Like I said, it's confusing, and there is a lot of misinformation. "Europe doesn't allow any food dyes" is a common refrain from people who think they should all be banned in the U.S. But in fact, it does. Most of the food dyes used in the U.S. — nearly all — are also approved for use in Europe.Should You Be Concerned About Food Dyes?, 2025-03-10

Notice what she does and doesn't do. She does not say dyes are harmless — she goes out of her way to add that Europeans using them doesn't automatically mean there's nothing to worry about. She simply puts the thing people are angry about and the thing that actually happened side by side on the table.

That is the move she has made over and over for twenty years. It is harder than it looks, because in public argument those two things almost always arrive in the same shape.

2. Who she is

Emily Oster, born 1980, economist, professor of economics at Brown, working in health economics. Both her parents are economists. Four books about pregnancy and parenting: Expecting Better (2013), Cribsheet (2019), The Family Firm (2021), The Unexpected (2024).

The detail worth keeping from that CV is this: at two years old, she was somebody's research subject. For a year her parents secretly tape-recorded the monologues she delivered to herself in her crib before falling asleep. A group of developmental psychologists analysed the 122 monologues, and Harvard University Press published the result in 1989 as Narratives from the Crib. The 2006 reissue carries a new foreword by the grown-up child from the crib, reflecting on what it is like to have been a research subject without knowing it.

The second thing happened in 2008. Her dissertation had a spectacular finding: China's skewed male-to-female birth ratio could be largely explained by hepatitis B rather than by the sex discrimination Amartya Sen had identified. It made her name. Then she got hold of a prospective Chinese liver-cancer cohort of 67,000 people, about 15% of them hepatitis B carriers, and checked the sex of their children. No effect at all. In 2008 she wrote the paper that demolished her own famous result. It is called "Hepatitis B Does Not Explain Male-Biased Sex Ratios in China."

So when she later tells American parents that the data does not say what they think it says, the first person she practised on was herself.

3. What she has been wrestling with

The question she hasn't let go of for fifteen years isn't "can pregnant women drink coffee." It is "who is entitled to make this decision."

She ran into it very concretely. Pregnant for the first time around 2010, she was handed a list at a prenatal visit:

There were certain things on it that I expected (like no tobacco) and even more that I did not — deli meats, raw eggs, unwashed vegetables, sushi, and on and on. As someone who loves data (and for some reason craved tuna-fish sandwiches when I was pregnant), this simple list drove me crazy.Is It Safe to Eat Deli Meats While Pregnant?, 2024-06-12

The point is that what drove her crazy was not the restrictions but the shape of the list. A column of prohibitions, same point size, same weight of voice, no numbers, no "roughly how likely," no "we're honestly not sure about this one," and nowhere at all that allowed for the possibility that she might weigh things differently from the person who wrote it.

She later turned that into a general diagnosis, and the diagnosis is the spine of this page: almost all pregnancy and parenting advice compresses three separate questions into one sentence.

The first is a question about evidence. The second is about magnitude. The third is not a medical question at all — it depends on how badly you want the coffee, how you feel about a risk of that size, what kind of year your household is having. In practice the third gets smuggled in with the other two, printed on the same list, and delivered in the tone of voice reserved for the first.

Reclaiming the third question is the whole project. Everything that follows is that one move applied to different material.

Worth noting that she was already doing it before she had children. The earliest piece in the archive is from 2010, titled "Are 'Feminine Problems' Keeping Poor Girls Out of School?", and it takes on a claim that was extremely popular in development circles at the time: that supplying sanitary products to girls in poor countries would substantially raise their school attendance. Her opening is pure Oster — tell the appealing story well, then add "as the story goes," then go and check. Her quarrel with good-sounding stories predates her kids.

4. Her three ideas, and how she ran into them

Idea 1: A rule is not data — "no amount has been proven safe" is a logic error

Start with where she noticed something was off. The heaviest item on the pregnancy list is alcohol, and the most-repeated sentence in American obstetrics is: "No amount of alcohol has been proven safe."

It sounds unimpeachable. It even sounds responsible — we're not sure, so we're being careful.

She won't have it, and her reason is purely logical:

At least in the U.S., one phrase I often hear is "No amount of alcohol has been proven safe." The implication, I suppose, is that we know that there exists a level of drinking that is bad, so we should assume all other levels are bad until proven otherwise. I think this is flawed logic.Will Alcohol During Pregnancy Alter My Baby's Brain?, 2022-12-02

Note that this is not an argument that you may therefore drink. It is an observation that the sentence is not a statement about evidence at all. It is a decision rule wearing the costume of a scientific finding. "Not proven safe" and "proven unsafe" are entirely different claims, and the sentence works by setting them next to each other until a value judgement — how cautious should we be under uncertainty — passes for a fact.

Then she does the second thing, and the second thing is the signature: she doesn't stop at the logic, she goes and finds the mechanism.

When alcohol enters your digestive system, your liver processes it into a chemical called acetaldehyde and then into acetate. The acetaldehyde is toxic to other cells… When you drink slowly, you metabolize much of the alcohol before it would get to the fetus. If you drink quickly, your liver cannot keep up and toxins are passed to the fetus. This is why binge drinking is so bad, but it also illustrates why negative effects of light drinking do not follow directly from negative effects of heavy drinking.ibid.

In the same passage she is unambiguous about the other half: heavy drinking in pregnancy is unquestionably harmful, and "if you are pregnant and are binge drinking, stop." She is not endorsing alcohol; she is dismantling an inference. The dose-response relationship isn't linear because the enzyme in the middle has a throughput ceiling — which is a specific, falsifiable biological reason rather than a shrug about what the data don't support.

She applies the same dismantling to the whole list. The risk class for sushi, the risk class for deli meat and the risk class for unwashed vegetables are three different pathogens, three different consequences and three incidence rates separated by orders of magnitude. Printing them in a column tells the reader "these are equally dangerous," and no study has ever said that.

The idea compresses to this: a list is itself a claim, and usually an unexamined one. It isn't that every item is wrong. It's that putting them on one list is wrong.

Idea 2: Confounding isn't a technical footnote — quite often it is the entire result

This is the one she has dug into deepest, and what she found is specific enough to stick.

It starts, again, with her. Three or four cups a day before pregnancy. She went looking for guidance and found the guidance fighting with itself:

My OB said under 200 milligrams (mg) a day (roughly two 8-ounce cups) was fine. A friend's OB said 300. The Mayo Clinic Guide to a Healthy Pregnancy ruled out caffeine entirely. … None of these sources could have been looking at the same data, or at least they were not interpreting it the same way. So I did what I always do: I went to the studies.How Much Coffee Is Safe During Pregnancy?, 2024-12-12

"So I did what I always do" is the whole methodology in six words.

The concern she was chasing was miscarriage. Down at the level of the human studies, she hit something:

Nausea.

Early-pregnancy nausea is, counterintuitively, good news — women who have it are less likely to miscarry; it signals a healthy pregnancy. Nausea also makes coffee undrinkable. Her own first-trimester routine of coffee before breakfast on an empty stomach became, in her word, revolting — and talking to other pregnant women, she found this completely typical.

But this means that women who drink a lot of coffee are probably those who are not experiencing nausea. This may make it appear that caffeine and miscarriage are linked, when they are really not. There is a third factor related to the health of the pregnancy that influences both.ibid.

What makes the example so good is that it isn't a minor technical blemish — it is enough to flip the conclusion outright. And the confounder is neither exotic nor obscure. It is the single most common experience of early pregnancy. It simply wasn't written down as a variable.

She has a pure-numbers version of the same skeleton. This is the problem she says she puts in every course she teaches:

Imagine you have a disease which affects 1 in 10,000 people. You've got a test which detects 99.9% of cases, with just a 2% false positive rate. Someone comes in and they get a positive test. What is the chance they have the disease?Bayes' Rule Is My Faves Rule, 2020-08-31

People usually answer 95% to 99%. She adds a detail it is hard to forget:

I use this example all the time. I think I've discussed it in every class I teach, regardless of the topic. I sometimes incorporate it into book talks. Once, I asked it to a group of University Trustees (one person got it right).ibid.

The answer is about 0.5%, or one in two hundred, and she walks every step. In ten thousand people, the one who has it is almost certainly caught. Of the 9,999 who don't, 2% test positive anyway — about 200 people. So roughly 201 positives, exactly one of them real.

What seems like — and is! — a small false positive rate amplifies because of the low baseline risk.ibid.

The coffee case and the test case are two faces of the same thing. What a piece of information is worth is never set by how accurate it is, but by the surface it lands on. She had been calling this principle "context" for years — then decided the word was too soft and said what she actually meant was Bayes' rule.

In practice it becomes a question she carries around. When she sees "A is associated with B," she doesn't ask how large the coefficient is; she asks what determines whether someone ends up in group A in the first place. That is exactly how she handles Jonathan Haidt's stack of literature on phones and adolescent mental health. She doesn't deny the studies exist:

The problem is that the vast majority of this evidence does not have any credible claim to causality. It shows correlations that have many interpretations (mental health could drive social media usage, for example). Although there are huge numbers of these studies, they aren't convincing — individually or together — about anything.What the Data Says About Phones and Social Media, 2024-07-08

She then picks out the strongest evidence herself — all of it from economists, because those are the papers with real identification strategies: staggered Facebook rollout across campuses, a four-week randomised discouragement experiment, a six-week randomised deactivation of Facebook or Instagram before the 2020 election. She isn't arguing that phones are fine; she is refusing to let a pile of unidentifiable correlations support any direction at all. The position annoys both camps, and she has held it from 2024 through 2026 without softening.

Idea 3: Data doesn't decide for you — "data isn't bossy"

The first two ideas are demolition. The third is the genuinely original one, and the one most often misread.

She says the question she gets most is: when should I follow the data and when should I follow my gut?

My answer: I reject this premise. Data isn't bossy, but your gut shouldn't be either.Sometimes You Just Have to Choose, 2021-05-13; reissued in 2026 as How to Make the Right Decisions

What she takes apart is the word "gut."

Saying "my gut feeling is that day care is right for us" is just another way to say "I would prefer to have my child in day care." The phrase "gut feeling" or "instinct" makes this seem like some kind of special, deeply rooted value, but it's not fundamentally any different than a preference that you'd have about your job or what to eat for lunch.ibid.

It is a hard move, and a useful one. "Instinct" is a trump card that cannot be answered; "preference" is a quantity you can put on a scale next to other quantities. Once you admit it's a preference — a strong one, perhaps — you have allowed the possibility that some piece of information could override it. You'd prefer that day care, but it has a two-year waitlist and costs double, so you reconsider. If it were an instinct, you'd just have to grit your teeth.

Having taken it apart she offers a replacement, and the replacement comes from an odd place: corporate management. The Family Firm argues for treating a household as an organisation that needs governing. The central tool is the Four Fs — Frame the question, Fact-find, Final decision, Follow-up. It reads like an MBA deck, but her reason is practical: what stops people isn't missing information, it's never having defined the question, and never letting any one discussion count — revisiting and revisiting until time makes the decision for you.

The most concrete thing she offers is from her own house. She observes that a great deal of parenting frustration is really frustration with a partner, and that most people handle conflict by deferring it:

The core reason: disagreements do not go away if you ignore them. Instead, they just stay and worsen and then, usually, blow up when you're not ready for it.Why You Should Surface Family Conflicts on Purpose, 2021-09-23

Her example is domestic. You feel strongly that your seven-year-old is in bed, lights out, at eight. You know your partner doesn't care as much, so you never raise it — you do bedtime anyway. Then you're away for work, you call at nine, and the child is up having a snack. You lose it; they shout back: this is your rule, not mine.

The conflict hasn't gone away; it's just waiting. And that moment — the moment when you're both tired and yelling — is just about the worst moment for this to surface.ibid.

Her fix is to put the conflict on the calendar:

In my household, we have a fixed weekly time called "Hard Conversation Time." This is a 30-minute time period at a time when we do not expect to be bothered, which is reserved for bringing up difficult issues. Often, there is nothing on the agenda.ibid.

"Often, there is nothing on the agenda" is the most honest line in the whole system. It concedes that the tool idles most of the time — and idling is precisely how it works. A mechanism that only activates when something goes wrong will always be activated at the worst possible moment.

By 2026 the thread reaches a new place. She begins to suspect that the things she has spent years helping people optimise are the wrong order of magnitude.

The fact is that there is little evidence that this kind of higher-volume involvement actually makes a difference to child outcomes, and, for parents, it can feel exhausting and unsustainable. The problem is, we have no idea what we should let go of.Let's Try "Deliberate Parenting" Instead, 2026-04-19

She cites a number: in 2018, working mothers spent more hours a week with their children than non-working mothers did in 1975. And then gives the shortest summary she has written in years:

In a short sentence: obsess less, but plan more.ibid.

This is a self-correction. She spent fifteen years teaching people how to get each decision right; now she is saying that most decisions have consequences too small to be worth the agony, and that the effort belongs instead in building structure in advance so you don't have to agonise in the moment. "Obsess less" and "plan more" only look contradictory. They are two ways of spending the same budget.

5. How she thinks

With a long-running writer, the transferable part is usually not the conclusions. Four things here can be taken straight off the shelf.

One: go to the source; don't argue in the second-hand layer. In 2024 a new panic went round social media: rushing your kids gives them anxiety. She didn't respond to the posts. She went looking for where the claim came from — and found The Hurried Child, a 1981 book by the psychologist David Elkind that has sold over half a million copies. She read the book, said it had real value and had in fact anticipated a great deal of what Haidt and others say now, and then cut:

However — and I cannot emphasize this enough — this book has nothing to do with rushing your kids out the door. This is about a global, existential hurrying, not about trying to get to stuff on time. Telling your 4-year-old child to hurry up and clean up your toys so we can have a bath is not the same as telling a sixth grader that if they don't make the travel soccer team they can forget about a college scholarship.Can Rushing Kids Give Them Anxiety?, 2024-10-21

She could have stopped there. Instead she added a short analysis of why the distortion keeps happening:

Smart and thoughtful research on how to develop a growth mindset in children is translated to "never tell your child 'good job.'" One study about how music improved test scores in college students resulted in parents everywhere piping Mozart into the womb.ibid.

Her explanation is a little brutal and quite accurate: parents are desperate for guidance that is concrete, finite and actionable, and the original research cannot supply that — what it asks for is a society-wide change in how children are regarded. So the market translates it into an action on your behalf. Playing the Baby Mozart CD at least feels like doing something. The distortion isn't caused by ignorance. It's caused by demand.

Two: state your own position and your own choices first. Her piece on childhood vaccines opens, before any data, by saying that she fully vaccinated both her children and that she believes the AAP schedule is right for virtually all healthy children. Then, immediately:

However, I also completely understand the instinct to ask questions about our kids, and I think the job of public health should be to meet these with answers. Like all medical treatments, vaccines carry some risk, and being unwilling to discuss that is not productive.Fact Check: Vaccine Studies, 2025-09-16

Three: say "I don't know" out loud, and specify which kind of not-knowing it is. Asked on the podcast anniversary about melatonin in pregnancy, she gives the available evidence and then adds a methodological note:

Sometimes the reason for conflicting advice is that the answer is: we don't have perfect evidence. And so the conflict is, how do you read the limited evidence that we have, not that people have different evidence.It's a Podcast Anniversary Q&A!, 2024-09-26

Four: structural honesty — leave an exit for readers who don't want the argument. Near the top of almost every long piece sits a button reading "Take me to the bottom line," and every piece closes with a numbered set of bottom lines. It looks like layout; it is a position. She knows most readers came for the number, and she refuses to charge them the price of reading her reasoning first. Those who want the reasoning scroll up; those who don't take the answer and leave. She holds neither in contempt.

A last exchange serves as a footnote to all of it. On the anniversary episode her producer Tamar offers her a lifeline — poll the audience, whose answers might be a bit more anecdotal, "but enough anecdotes become data."

Emily Oster: It's not how it works, but okay. I like the idea. I like the concept.ibid.

6. Where she may be wrong

First, alcohol. This is her most famous position and the one met with the weightiest opposition. In 2013 Susan Astley of the University of Washington's Fetal Alcohol Syndrome Diagnostic and Prevention Network wrote a rebuttal to Expecting Better titled "The Conventional Pregnancy Wisdom is Right — Do NOT Drink while Pregnant"; the National Organization on Fetal Alcohol Syndrome has criticised her publicly as well. The objections are twofold: that as an economist she can run the statistics but lacks the clinical context of the field; and that individual responses to identical exposure vary genetically, so no risk on average across a population does not mean no risk to your child. I have not read the text of Astley's document, so I report the bibliographic fact only and will not paraphrase an argument I haven't read. It is worth saying, though, that the criticism and her logic do not quite collide: she attacks the form of the inference in "not proven safe," while the critics attack whether the population average is the right statistical object. Both can be true at once.

Second, COVID and schools. Through 2020 and 2021 she was among the loudest American advocates of reopening; she built a school case dashboard (criticised at the time for its methods) and launched the COVID-19 School Data Hub in September 2021. In October 2022 she published a piece in The Atlantic arguing for a "pandemic amnesty," asking people to forgive one another for what they said and did in the dark. It produced the largest backlash of her career, and one strand of it is hard to answer: the person proposing the amnesty tends to be the person who needs it, and those who lost the most — families of disabled children cut off from therapies and in-school support — did not hold an equal position in that mutual forgiveness.

Third, and most structural. In October 2024 Anna Louie Sussman argued at length in The Nation that Oster has turned pregnancy and parenting into an individual optimisation problem, and that the framework itself makes collective action invisible. The Family Firm teaches a household how to secure the best position for itself inside a bad system; it asks no one to fix the system. It hands well-resourced families a competitive toolkit and has nothing to say to families without the resources.

This criticism is more lethal than it sounds, because it lands on her most attractive claim. "Data isn't bossy; the choice is yours" holds only if you actually have choices. You have to be able to afford that day care, live in that district, take the leave, and have a partner who will sit down for a thirty-minute Hard Conversation Time. Her tools operate on a feasible set; when the feasible set has one element, optimisation is not a meaningful word. In "Why Schools Always Call Moms" she does propose an institutional fix — the school forms should have a line asking who to call first, and then schools should honour it — but that passage is the shortest in the piece, and the advice immediately after it returns to asking whether the pattern is a problem for your family. Handing a structural pattern back to the individual is her best move and also her most exposed one.

Last, her own selection bias. The problem she is best at spotting now grows on her own distribution channel. ParentData has gone from one person's free newsletter to a company with a paywall — as of this page's review, an anonymous reader can read about 3% of her output in full. Which means the population that keeps reading her complete arguments is selected on ability to pay a subscription. That is precisely the kind of selection she has spent twenty years telling everyone else to watch for. This is not hypocrisy; it is the cost of scale. But since she taught her readers to ask what determines who ends up in a group, the question should be turned on her too.

7. Sources

• How Much Coffee Is Safe During Pregnancy? (2024-12-12)The nausea-as-confounder piece. If you read one, read this — the whole method runs end to end here: official guidance contradicts itself, go to the primary studies, find a third variable nobody recorded, and an entire literature may be pointing the wrong way.
• Bayes' Rule Is My Faves Rule (2020-08-31)One in 10,000, 99.9% detection, 2% false positives, answer 0.5%. She says she puts the problem in every course she teaches; in a room of university trustees, one person got it right. The free portion happens to contain the problem and the full solution.
• Will Alcohol During Pregnancy Alter My Baby's Brain? (2022-12-02)The logic error in "no amount has been proven safe," plus the acetaldehyde throughput argument for why heavy-drinking harm does not imply light-drinking harm. Her most famous position and the one with the weightiest opposition.
• Can Rushing Kids Give Them Anxiety? (2024-10-21)She reads the 1981 book The Hurried Child all the way through, shows the internet panic is about something else entirely, and then explains why research always gets translated into the wrong action — parents want an executable instruction, and the market supplies one.
• Why You Should Surface Family Conflicts on Purpose (2021-09-23)The most concrete artefact of the Family Firm apparatus: the standing thirty-minute weekly "Hard Conversation Time" in her own house, where usually there is nothing on the agenda. Free in full.
• Fact Check: Vaccine Studies (2025-09-16)Her complete demonstration of how to handle a hostile topic: declare your own choices, concede the part of the worry that is factually correct (the absence of true saline placebo arms really is a fact), then ask what the evidence on the other side actually is. The site made this one free in full.

The remaining quotations on this page come, in order, from Should You Be Concerned About Food Dyes? (2025-03-10), Is It Safe to Eat Deli Meats While Pregnant? (2024-06-12), Sometimes You Just Have to Choose / How to Make the Right Decisions (2021-05-13, reissued 2026-01-26), What the Data Says About Phones and Social Media (2024-07-08), Let's Try "Deliberate Parenting" Instead (2026-04-19), Why Schools Always Call Moms (2023-07-17), It's a Podcast Anniversary Q&A! (2024-09-26), The Four Fs (2024-03-31), Safety Lasagna (2020-09-17) and Are 'Feminine Problems' Keeping Poor Girls Out of School? (2010-04-27). Biographical facts and criticism come from Narratives from the Crib (ed. Katherine Nelson, Harvard University Press, 1989; the 2006 edition carries a new foreword by Emily Oster), Chen & Oster, "Hepatitis B Does Not Explain Male-Biased Sex Ratios in China" (NBER w13971, 2008; later in Economics Letters, 2010), Susan Astley, "The Conventional Pregnancy Wisdom is Right — Do NOT Drink while Pregnant" (University of Washington, 2013), and Anna Louie Sussman, "Emily Oster and the Optimization of Parenting" (The Nation, 2024-10-29). Blog home: parentdata.org.