Where the Bias Gets In, and What Nobody Looked For

The chapter uses two terms from chapter 3 constantly, so they are worth fixing now.

• Methodological objectivity. Knowledge comes from methods that work reliably and that anyone could run and get the same answer. The procedure is what makes it trustworthy.
• Agent objectivity. The person doing it should be detached, with no stake and no values in play. The researcher is what makes it trustworthy.

Both come from the same source, which the chapter calls the VFI. The value-free ideal is the view that good science has no social, political or moral commitments inside it anywhere. Every feminist position in this chapter rejects the VFI. Almost none of them give up objectivity, and that combination is the thing to be able to state.

And the chapter opens on a story that makes the stakes concrete. In 1912 the New York Times reported James Crichton-Browne, a prominent neuropsychologist who had worked with Darwin, on blood flow in male and female brains. Women came out with sensuous perception, rapid thought and emotional sensibility. Men came out with originality at the higher levels of intellectual work, calmer judgment and stronger will. The blood-flow theory is long dead. The thing it was used to prove, that cognitive differences between men and women are fixed by biology, has outlived every mechanism proposed for it: hormones, genes, brain structure. That survival is what the chapter wants explained.

You learn in school that the scientific method is: make a hypothesis, test it, conclude. The chapter says that is not false so much as useless, because the interesting decisions are hidden inside the word test. So it cuts inquiry into nine phases, following Anderson and Brown:

1. Choosing the research area or subject.
2. Framing the questions about it, which is where hypotheses come from.
3. Deciding which observations are relevant.
4. Collecting data based on those observations.
5. Analysing or interpreting the data.
6. Deciding when to stop collecting.
7. Drawing conclusions from the evidence.
8. Considering alternative explanations.
9. Testing the conclusions for robustness.

Two warnings they attach to the list. The phases are not a sequence: research loops, and a decision at phase 7 sends you back to revise phase 3. And they interact, so how you conceived the object in phase 1 decides what counts as relevant in phase 3, which decides what there is to interpret in phase 5. Separating them is a tool for looking, and never a description of a timetable.

Chapter 3 covered phases 1 to 3. This chapter is about 4 to 9, and the authors say why: those are the controversial ones. Hardly anybody denies that scientists choose their topics for human reasons. The claim that values reach into collecting and interpreting the evidence is the one people fight about.

Run the rod and frame test down the nine phases and you can watch the conclusion being assembled. Phases 1 and 2 were chapter 3's business. These are the six the authors say people fight about.

3. Which observations count. Researchers assumed the rod and frame test is a measure of visual-spatial ability. That assumption is the whole experiment, and nobody ever argued for it.
4. Collecting. They collected that men aligned the rod correctly more often. Correct, as far as it goes.
5. Interpreting. They read that as men having better spatial cognition. The authors make a point here worth copying: the researchers also shared the widely held belief that men and women differ cognitively, which made that reading the easy one and every other reading the effortful one.
6. When to stop. The chapter's key term lives in this phase, and it gets its own treatment below. The short version is that the expected result made less evidence feel like enough.
7. Concluding. They concluded the difference could only be explained by biology. The authors call this unwarranted for a specific reason: the differences are small, even where they are statistically significant. Reporting the significance while leaving out the size is itself a choice about what matters.
8. Alternatives, and this is the phase that does the damage. Social and environmental explanations existed. So did experimental error, meaning the test measures assertiveness. Neither was ruled out, because neither occurred to anybody. Concluding means inferring to the best explanation, and you can only compare the explanations you have thought of, so who is in the room changes how well supported the conclusion actually is.
9. Robustness. The missing alternatives point at confounding variables, meaning factors other than the one you are studying that move with the thing you are measuring. Name them and you can design controls for them. Later studies did, and the gap shrank.

Deciding you have enough evidence is a judgment about the risk of being wrong, and that is called inductive risk. How much evidence is enough depends on the probability the hypothesis is true and on what it would cost if you got it wrong. The second half is a value judgment. The authors run it both ways, which is what keeps it an argument:

• Wrongly accept that spatial ability is biologically fixed by sex, and you reinforce a stereotype, which bears on whether girls are encouraged into mathematics, engineering and architecture. Those are well-paid fields, so the cost shows up in women's earnings. And because the stereotype is already everywhere, and people weight evidence that agrees with what they already think, the error spreads easily.
• Wrongly reject it when it is true, and you lose the chance to understand a real difference and deal with it fairly, and you may back policies that cannot work.

Which error is worse is a value judgment, and somebody makes it every time. The practical rule the chapter offers: when the cost of error is high and nothing is urgent, collect more and avoid sweeping conclusions.

The move the whole chapter turns on. The traditional question is how does an individual know objectively. Feminists replace it with what conditions have to be met for knowing to be objective. The property shifts off the person and onto the arrangement.

The first answer, and the one most people start with. Spontaneous feminist empiricism is Harding's name for the position that the standard methods are fine and the problem is that they were applied badly. Follow the procedures properly and the sexist values get filtered out. It keeps both traditional kinds of objectivity intact and treats bias as a failure of execution.

Why the chapter thinks that is not enough. Practices thought to be good science can carry values inside them, where nobody sees them, which is exactly the rod and frame test. And if values are unavoidable at some phases, the job is no longer to remove them. It is to tell the legitimate uses from the illegitimate ones, and the obvious line, values are fine in phases 1 and 2 and not after, does not hold, because the phases feed each other.

Her account of evidence first, because the objectivity follows from it. Nothing is evidence on its own. A fact becomes evidence for a hypothesis only through background assumptions that connect the two. Change the assumptions and the same fact stops being evidence, or starts being evidence for something else. She calls this contextual empiricism: empiricist because only observable things count as evidence, contextual because what they are evidence of depends on the surrounding assumptions.

Then the objectivity. If what counts as evidence depends on assumptions, and assumptions are shared and invisible to the people who share them, then no individual can get out from under their own. The only thing that can is other people. So objectivity is a property of how a community is organised, and Longino names four conditions for what she calls transformative criticism:

1. Public venues for criticism, of evidence, methods, assumptions and reasoning.
2. Uptake. The community must not merely tolerate criticism but actually change its beliefs over time in response. This is the one that has teeth.
3. Public standards for judging theories and observations, so criticism has something to appeal to.
4. Tempered equality of intellectual authority. Take seriously anybody with the relevant expertise, and never dismiss them for gender, race or class. Tempered means expertise still counts: this is not a rule that every opinion weighs the same.

What this buys. Science is not value free and it does not have to be. The values get managed, by being made explicit and argued about in public, which Janet Kourany names the social value management ideal. And diversity now has an epistemic job, on top of the moral one: a community that all shares the same assumptions cannot see them, so difference is what makes criticism possible. Objectivity comes in degrees, since no real community meets the four conditions fully.

Two objections the chapter reports, and Kristina Rolin answers both.

1. It lets in too many people, including people who reject the norms themselves. Rolin's reply: the four norms already rule out anyone who will not keep them.
2. It costs too much. Answering every criticism would take all of a scientist's time. Rolin's reply: the norms put the duty to respond on the community, so its members can divide the work.

And the objection that matters most for this course. Kristen Intemann, one of this book's own authors, points out that the four norms ask a community to hold a range of values and say nothing about who is in it. So a department can satisfy all four while every member occupies the same social position, and the assumptions that come with that position never get raised by anybody, because there is nobody there who does not share them. The norms are silent about the thing that decides which criticisms turn up.

What her fix adds, in one line: who is in the room becomes an epistemic question. She keeps Longino, so a claim still answers to evidence and objectivity still belongs to a community working under the four norms. She adds standpoint theory's claim that people differently placed have access to different evidence, which supplies the answer Longino lacks to diverse in what respect. Diversity of social position stops being only a matter of fairness and becomes a condition on the evidence the community can reach. Intemann calls the result feminist standpoint empiricism: empiricist because evidence decides, standpoint because who is looking decides what evidence there is.

Her diagnosis, in her own words. The problem was that good science lacked the methodological resources to detect widely held sexist and androcentric assumptions. Androcentric means centred on men, treating male behaviour as the standard case. Good science, following the traditional standards, did not catch these, so the standards are what need fixing.

The engine of standpoint theory, which has a name. The inversion thesis: people who are socially, politically and economically subordinated are, because of that position, better placed to notice the assumptions researchers in dominant positions are making. Harding's line is uncompromising: a maximally critical study of scientists and their communities can be done only from the perspective of those whose lives have been marginalised by those communities.

Why that is more than a nice idea. Two mechanisms carry it, and both are concrete.

1. You cannot see what you are seeing with. A person inside the dominant framework cannot inspect its assumptions, because those assumptions are the instrument.
2. Marginalised knowers can state their own interests, so they can propose a different account of what the object of study even is.

Two sociologists as worked cases.

1. Dorothy Smith found a mismatch between her subjects' lived experience and the concepts of traditional sociology. She called those mismatches lines of fault, and argued they are where the method should be rebuilt from.
2. Patricia Hill Collins found that neither mainstream sociology nor white feminist accounts fitted the Black women she studied. That is how standpoint theory leads into intersectional analysis, where a weaker version of it would merely have tolerated it.

And the definition to write down. The two words are a pair, and the second is defined against the first.

Harding's own summary of the second one. Strong objectivity requires that scientists and their communities be integrated into democracy-advancing projects, for scientific and epistemological reasons as well as moral and political ones.

She starts from a complaint about the word. Ian Hacking called objective an elevator word, meaning a word that does no work of its own and only raises the standing of whatever it is attached to. Calling a result objective says it does not suffer from some failing, without saying what it has.

So she makes it mean something. Objectivity is a property of the knowledge claim itself, where the tradition hung it on the researcher. A claim is objective to the degree it shows the epistemic virtues, meaning the properties that make a claim good as knowledge. The standard list comes from Kuhn and from Longino.

• Empirical adequacy. It fits the data.
• Inferential robustness. The conclusion survives a change in the assumptions it was drawn from.
• Explanatory power. It accounts for more of what needed accounting for.
• Internal coherence. Its parts agree with each other.
• Consistency with other established knowledge. It sits with what the field already holds.

And here is where standpoint gets in, which is the clever part. You can almost never maximise all the virtues at once. A model that fits the data best is often not the one that explains the most. So at every phase researchers must decide which virtues to prioritise for this project, and that decision depends on their purposes and goals, which depend on where they are standing. Standpoint improves objectivity by improving those judgments.

Her three theses, which are the tidiest summary of standpoint theory in the reading.

1. The generic situated knowledge thesis. All knowledge comes from some social position. There is no view from nowhere.
2. The systemic situated knowledge thesis. Social power is systematic, so the epistemic consequences are systematic too, and not random noise.
3. The thesis of epistemic advantage. Particular insights may be available to those who are marginalised.

And her status claim. Standpoint theory is a purpose-specific epistemic stance and not a full-service epistemology: a tool to be used on particular problems, and never a replacement theory of knowledge. Harding's early work treated it as an alternative epistemology, Wylie rejects that, and Harding's later work has moved toward Wylie's view. Intemann calls the merged position feminist standpoint empiricism.

Miriam Solomon sorts feminist empiricism into two kinds by what values are allowed to do.

• Gap feminist empiricism. Evidence never settles which theory to believe on its own, because theory is underdetermined by the evidence, meaning several theories fit the same data. Values break the tie. Longino's 1990 position is her example.
• Feminist radical empiricism. The stronger claim: values are themselves empirical, so evidence bears on them the way it bears on any other belief, and they can serve as evidence in turn. Lynn Hankinson Nelson, Sharyn Clough and Elizabeth Anderson. The picture behind it is Quine's web of belief, where everything you hold answers to experience as one connected whole, and never one claim at a time. On this view sexist science fails because it is worse science, less empirically successful, and the failure of the whole system is what tests the values inside it.

The chapter then reports two objections to the stronger view and endorses neither. Solomon doubts you could ever isolate one bad value inside a holistic web to test it. Audrey Yap's worry is practical: even where evidence does tell against a value, people holding it tend not to give it up, so this is unlikely to work as a way of removing sexism and racism from research.

The facts the chapter uses. Tarana Burke began MeToo on social media in 2006, encouraging people to make their experiences of sexual assault and harassment public. It became enormous in 2017, when the actress Alyssa Milano pushed the hashtag during the accusations against the film producer Harvey Weinstein.

What the authors say it did, which is more interesting than what it is usually credited with. It made visible something most people had not known: how common sexual assault and harassment are, especially at work.

And it did something sharper. It gave survivors a way to see their own experience differently. Women who had described an encounter with an employer as consensual, including Monica Lewinsky and several NBC employees who worked with Matt Lauer, later said the consent had come partly from fear of what refusing would do to their careers. The movement handed people a concept, and with it the encounter changed shape.

And the same hashtag hid things. The surge of use by wealthy white celebrities made less visible the thing Burke's original movement was about: that women of colour, Indigenous women, trans women and other vulnerable women are targeted disproportionately.

The field. Agnotology is the study of ignorance: how it is produced and maintained. It got going in the 1990s with Robert Proctor, and feminists, critical race theorists and decolonial theorists developed a related line called epistemologies of ignorance. Sullivan and Tuana's definition: identifying the different forms of ignorance, examining how they are produced and sustained, and asking what role they play in knowledge practices.

Ignorance is not automatically bad, and the chapter spends real time on this. It is a tool as often as a failure.

• Blinding a clinical trial. Nobody is told who got the drug, precisely so that knowing cannot bend the result.
• Knowledge it would be wrong to get. You cannot force patients into a randomised trial of a surgical procedure, so some evidence will never exist, by agreement.
• Knowledge that is dangerous. How to build a biological weapon is restricted on purpose.
• Ordinary life. There are things said about you that you would rather not know.

And the claim that makes it a field. Some say ignorance is simply the absence of knowledge, or the lack of true beliefs, or uncertainty. The chapter's position, from Sullivan and Tuana and Proctor, is that this misses the interesting cases. Social and cultural forces leave gaps, and they also make certain things hard or impossible to know. Linda Alcoff's phrase is the one to keep: ignorance as a substantive epistemic practice in itself, and never merely a neglectful one.

ignorance is not something simple: it is not a simple lack, absence or emptiness, and it is not a passive state. Ignorance of this sort, the determined ignorance most white Americans have of American Indian tribes and clans, the ostrich-like ignorance most white Americans have of the histories of Asian peoples in this country, the impoverishing ignorance most white Americans have of Black language, ignorance of these sorts is a complex result of many acts and many negligences.Marilyn Frye, 1983, p. 118

Where it comes from. Nancy Tuana built the taxonomy out of the women's health movement of the 1970s and 1980s, which she describes as a movement of epistemological resistance: getting women knowledge about their own bodies, and attacking the production of ignorance about them. The five are in the order she gives them, and the order matters because it runs from passive to active and then out the other side.

1. What we do not know but do not care to. Since the 1960s hormonal contraception has been aimed at women and tested on women. Male hormonal contraception was considered and dropped. The reasoning was that men would be less motivated to use it and less willing to accept side effects, with reduced libido the example. Nobody worried about reduced libido in women, because of the standing assumption that women did not have much desire to begin with. Pharmaceutical companies also assumed preventing conception was women's work, so it looked unmarketable. We still have no safe effective male hormonal contraceptive, and what stopped it was a decision about who would buy it.
2. What we do not know that we do not know. Hard to spot, and in hindsight it looks like the first. Tuana's case is the anatomy of the clitoris. Study of the genitals was organised around reproduction, the clitoris was taken to play no part in reproduction, and so it was not a thing to be ignorant about. Nobody rejected it as unworthy. It was never an object of enquiry at all.
3. What they do not want us to know. The pharmaceutical industry concealing known risks and side effects of oestrogen in contraceptives, to protect its margins. The chapter notes that this kind is not automatically wrong, since restricted dangerous knowledge is also somebody deciding what you may not know.

What we do not know and do not want to. This is the only one of the five where the ignorant person is doing something, namely employing active strategies to avoid knowing, including neglecting evidence that would force the knowledge on them.

The mechanism Mills describes. Some white people do not know about their own privilege, and the claim goes further than that. They refuse to understand the conditions non-white people live under, and the institutions and beliefs underneath them. Where a society is structured by inequality, the dominant group has an interest in seeing the world as a place where inequality is justified or absent. People with advantages have an interest in believing the world is fair and that what they have is deserved, and no interest at all in learning otherwise, because of what that would imply.

White America remains unable to believe that Black America's grievances are real; they are unable to believe this because they cannot face what this fact says about themselves and their country.James Baldwin, 1985, p. 536

Elizabeth Spelman's analysis of how that works in practice. It is a failure of interest and a gap in experience, and then something more active. When confronted with testimony or evidence, people immunise themselves against anything that would correct them. Evidence of police brutality gets discredited, the suspect must have been doing something, or dismissed, the force was reasonable because they did not comply. Only in extreme cases with inescapable evidence, such as the videos of the murder of George Floyd in 2020, is a wrong widely acknowledged, and even then it is treated as exceptional, where it could be read as evidence of a pattern.

And the distinction worth having, because it is subtle. Self-deception is convincing yourself something is true because you want it to be. Willful ignorance is different: you fear it might be true, you have an interest in not believing it, so you avoid the question altogether and build ways of minimising whatever evidence turns up. Tuana's phrase for it is an active ignoring of the oppression of others and one's own role in it.

What we cannot know. The one kind that is no failing at all. There are limits on what humans can perceive, limits of time and money, and limits on understanding experiences you have not had. Maria Lugones and Mariana Ortega call the right response loving ignorance: accepting that some experience cannot be fully shared, while refusing to ignore, dismiss or stereotype the people whose it is. Lugones calls for a loving perception of those who have been marginalised. Ortega sharpens it: many white feminists think reading about the experiences of women of colour is enough, and that reading produces its own arrogance, a confidence that you now understand. What she asks for instead is sustained engagement and real relationships.

Cutting across the five kinds, the chapter names five strategies. These answer how, where the taxonomy answered what.

Passive when a question never occurs to anyone, active when somebody chooses to look away. Three cases from the chapter:

• Fossils in the ancient Mediterranean. Adrienne Mayor shows the accounts existed. The history of science passed over them because scholars disregarded non-elite sources and filed them as folklore, so modern palaeontologists were ignorant of the earliest fossil discoveries.
• What the news covers. Media choose which stories to run, which is part of why people on different news diets believe different things.
• Which parts of United States history get taught. The chapter's longest case, and it notes that neglect happens across the political spectrum: the Democratic Party's 2016 decision to concentrate on urban and suburban voters left its own officials systematically ignorant about rural conditions.

Four routes, and only the first is the obvious one.

• Active. Bans on teaching certain topics, or classification of dangerous material.
• Scientific. Data or methods are not shared, under commercial confidentiality or claimed national security.
• Chosen by the marginalised themselves. Mayor records Native American groups withholding fossil knowledge from outsiders, and many Navajo avoiding anything to do with death, including dinosaur fossils on their land.
• Language and form. Latin American feminist philosophy stays invisible to Anglophone scholars, and Indigenous oral histories get excluded as unreliable.

And the fifth route, citation, which has the chapter's sharpest example. Cuvier wrote up Iroquois and Delaware accounts of mastodon fossils, calling the mastodon the grandfather of the buffalo. When his work was translated into English twenty years later that entire section was left out, which made North American fossil discovery look European and fed the story that nobody was there before.

Attacking the knowledge. Indigenous and local knowledge gets dismissed as myth for not arriving as written-up hypotheses, despite Micronesian navigation systems, Indigenous pharmacology that was then exploited, and conservation findings repeatedly rediscovered from traditional ecological knowledge. The chapter names the assumption underneath this the scientification of knowledge: that only the products of Western science count. Midwives are the clearest case, their experiential knowledge dismissed as unscientific and unskilled as obstetrics professionalised in the nineteenth century, with licensing restrictions that still exist.

The main technique is uncertainty, which is the tobacco strategy. Executives knew they did not have to show smoking was safe, only that the science was unsettled. They funded conflicting research, exaggerated the uncertainties, and put non-scientists on television to attack the science.

Attacking the knower works through dismissal for an irrelevant feature: Indigenous people as savages, women as incapable, and more subtly the senators who said they believed Christine Blasey Ford had been attacked while doubting her memory of by whom. It can also target institutions, since claims of fake news aim to leave people not knowing what to trust at all.

What is new is the speed and the scale. Fabricated stories, sites and images spread worldwide in seconds and keep spreading after a platform flags them. Three things amplify it: lack of trust, social networks built on shared values, and algorithms that feed you what your clicks predict you will like. Add confirmation bias, the tendency to weight evidence that agrees with you, and you get information silos, echo chambers and what C. Thi Nguyen calls epistemic bubbles.

Misleading with things that are true. Four cases, and the second is the one to remember:

• The party. A teenager says there will be adult supervision and leaves out that the chaperone is a friend's twenty-one-year-old sibling.
• The drug. A company reports that patients taking it for five years had a 95 per cent survival rate, and does not mention that the rate was also 95 per cent at two years, so patients buy three extra years of a drug for nothing.
• The heavy metals. Chemical companies funded studies showing less testicular cancer in rats at low doses, implying low doses are good for men, while never checking whether other cancers rose at the same doses.
• Political advertising, which does the same with quotations and voting records.