Data Feminism, by Catherine D'Ignazio and Lauren F. Klein (MIT Press, 2020), is a way of thinking about data — its collection, analysis and communication — informed by intersectional feminist thought. It is not only about gender; it is about power.

Numbers feel objective. But every dataset embeds human choices — what to measure, which categories to use, whom to ask, what to ignore. Those choices carry the values and the power of the people who made them.

Legal scholar Kimberlé Crenshaw coined intersectionality in 1989 to show that a Black woman's experience is not simply 'racism plus sexism' — the forms of oppression intersect to create something distinct that neither alone captures.

Those with power decide what counts as worth counting. Historically, that has meant data about the powerful is rich and data about the marginalised is thin, distorted, or absent altogether.

Data feminism does not reject data — it demands better data: more representative, more accountable, more honest about its limits. The goal is data that serves justice, not data abandoned.

Data feminism is not new theory invented for the data age — it applies a long tradition. Its foundations sit in standpoint theory, Black feminist thought, and intersectional scholarship that asked, long before 'big data', whose knowledge counts.

South Asia runs some of the world's largest data systems — the Census, NFHS, Aadhaar, welfare rolls — touching over a billion lives. The bigger the system, the higher the stakes of who it counts well and who it counts badly.

The first principle is to examine power — to analyse how power operates in the world, and how it shapes the data we have. You cannot challenge what you have not first named.

D'Ignazio and Klein urge us to study not just the dataset but the data setting: the people, institutions, incentives and histories that produced it. A number is the visible tip of a social process.

Power is most effective when it is invisible — encoded in a 'standard' form, a 'normal' category, a 'default' user. Examining power means making those defaults visible and asking who they were built around.

Surveys often record the 'head of household' — frequently assumed, and recorded, as the oldest man. His characteristics then stand in for the whole family, and the women's circumstances are read through him.

Examining power is not about finding a villain. Power works through routine, well-meaning processes — a standard form, a default field, a 'best-practice' indicator — that quietly advantage some and disadvantage others.

Examining power is diagnosis; challenging power is action. The second principle commits data work to contesting unequal power and working toward justice — not merely describing the world, but changing it.

When institutions will not count a problem, people build the count themselves — mapping sexual harassment, logging manual-scavenging deaths, recording missing toilets — to force the issue onto the agenda.

Challenging power includes auditing the algorithms and datasets that increasingly decide who gets a benefit, a loan, a ration card or a welfare flag. Biased data produces biased decisions at scale.

India's MGNREGA social audits turn the scheme's own administrative records back on it: villagers read out muster rolls and payments in a public hearing, and those present testify whether the work and wages were real.

Power lives in framing. 'Women's low workforce participation' frames women as the problem; 'an economy that does not count or reward women's work' frames the system as the problem. Same data, opposite politics.

The fourth principle asks us to question the categories we count with. Binaries (male/female, formal/informal, literate/illiterate) and hierarchies are not facts of nature — they are choices that include some realities and erase others.

When a form offers only two gender boxes, anyone outside them is forced to misreport or vanish. The category does not just record reality — it constructs the official version of it.

India legally recognises a third gender (Supreme Court, NALSA judgment, 2014), and the Census has counted 'Others' since 2011. Yet many forms, schemes and datasets still offer only male/female — rendering trans and non-binary people invisible.

In 1990, economist Amartya Sen asked a feminist data question: given normal sex ratios at birth and survival, how many women should be alive in Asia — and how many are not? His answer: over 100 million 'missing women', lost to neglect, discrimination and sex-selective practices.

Classifications often smuggle in a hierarchy: 'head of household' above other members, 'skilled' above 'unskilled', 'productive' work above 'reproductive' work. The ranking shapes whose contribution shows up — and whose disappears.

Sen's insight is methodological as well as moral: some of the most important data is about people who are not there. The missing women, the out-of-school girl, the unregistered birth — absences that only a deliberate question can reveal.

Categories are not eternal. India's recognition of a third gender, the slow inclusion of disability in the Census, the debate over a caste census — each shows classification responding to demands for visibility.

The third principle challenges the idea that good data work must be cold and detached. Emotion and embodiment — feeling, lived experience, the body — are valid sources of knowledge, not contaminants to be scrubbed out.

In development data, a row is rarely an abstraction — it is a woman who walked three kilometres for water, a child weighed at an anganwadi, a worker whose wages went unpaid. The data-feminist asks us never to forget the body behind the number.

A bar chart of maternal deaths informs. A memorial that gives each death a name, a place, a face — that moves. Both are data; only one reaches the heart.

Embodiment means recognising that knowledge is held in bodies and lived experience, not only in spreadsheets. A community health worker 'knows' which hamlet is hungry before any indicator confirms it — that embodied knowledge is real data.

The sixth principle insists that numbers do not speak for themselves. A figure ripped from its context can mislead more than it informs. Context is not optional background — it is part of the data's meaning.

D'Ignazio and Klein echo Lisa Gitelman: 'raw data is an oxymoron.' Data is always already 'cooked' — cleaned, classified, shaped by choices — before anyone analyses it. Pretending it is raw hides those choices.

A district reporting '90% institutional deliveries' looks like success — until you learn the facilities lack blood banks, the 10% missing are the remotest women, and the count includes a referral that ended in death elsewhere. Context flips the story.

Sometimes the ethical act is to refuse a comparison the data invites. Ranking communities by 'crime rate' without noting differential reporting and policing can stigmatise the over-policed and exonerate the powerful.

Part of considering context is tracing provenance: a crime statistic reflects policing as much as crime; a complaint count reflects access to grievance systems as much as grievances. The data measures the system, not only the phenomenon.

Every aggregation is a loss of context. A national average dissolves the district; a district average dissolves the village; a household figure dissolves the woman inside it. Each step up the ladder erases someone.

The seventh principle credits the work behind the data. Every dataset rests on labour — the surveyor in the field, the data-entry clerk, the cleaner, the respondent who gave an hour of their day — and most of it goes uncredited and unseen.

An ASHA worker filling registers, an anganwadi worker weighing children, a frontline enumerator building rapport — their data work is inseparable from care work, and like care work it is feminised, underpaid, and treated as 'natural' rather than skilled.

Beyond the data pipeline lies the largest invisible labour of all: unpaid domestic and care work, done overwhelmingly by women, and historically excluded from GDP and from most statistics — until surveys deliberately chose to count it.

An hour given to a survey is an hour not spent earning, cooking or resting — a real cost borne disproportionately by women, who are surveyed about the household yet rarely see anything return to them.

Behind every clean dashboard is invisible infrastructure labour: the engineer who maintains the database, the moderator who reviews content, the annotator who labels training data — work that is often outsourced, precarious and unnamed.

The fifth principle calls for many voices and forms of knowledge. The most complete picture comes not from a single expert viewpoint but from synthesising multiple perspectives — especially those of the people closest to the issue.

Local and experiential knowledge — a midwife's, a fisherwoman's, a sanitation worker's — is data too. Privileging only credentialed expertise discards the knowledge of those who live the problem daily.

Inviting many voices is not the same as listening to them. Token consultation — gathering views you have already decided to ignore — can be worse than none, extracting time while changing nothing.

Feminist philosopher Donna Haraway argued that all knowledge is situated — seen from somewhere, by someone, with a particular stake. There is no 'view from nowhere'. Embracing pluralism means combining many situated views into a fuller picture.

The deepest form of pluralism is letting communities help decide the questions, not just answer them. When women define what 'safety' or 'wellbeing' means in their own terms, the resulting indicators measure something that matters to them.

D'Ignazio and Klein use it as the analytic backbone of data feminism: a way to see how oppression is organised, and where data work can intervene.

Kimberlé Crenshaw (1989) showed that Black women fell through the gaps of laws that treated race and sex as separate — they were neither the typical 'woman' (white) nor the typical 'Black person' (male) the categories imagined.

In South Asia the most important intersection is often caste and gender. A Dalit woman faces a compounded exclusion that neither 'women's' data nor 'Dalit' data, read separately, can reveal.

Report only the gender gap and you miss caste; report only the caste gap and you miss gender. The Dalit woman sits at the bottom of the previous chart not by coincidence but by the compounding the matrix of domination predicts.

The matrix of domination is not a list of 'oppressed groups' — it is a structure everyone occupies. A person can be privileged on one axis (caste) and oppressed on another (gender), advantaged here and disadvantaged there.

Caste and gender are central in South Asia, but the matrix has many axes: religion, disability, sexuality, language, region, age and migration status all intersect. A disabled Muslim trans woman migrant sits at an intersection almost no dataset captures.

What is not counted is not budgeted for. The gender data gap is not a technical oversight — it is the privilege hazard at the scale of a whole statistical system.

In Invisible Women (2019), Caroline Criado Perez documents how a world built on data that treats the average man as the default harms women — from car-crash dummies to drug doses to phone sizes designed for men's hands.

India's Time Use Survey 2019 (NSO) was the first nationwide TUS in nearly two decades. By measuring how people spend their 24 hours, it rendered the unpaid care economy — overwhelmingly women's work — statistically visible.

NFHS-5 (2019–21) is a gender-data powerhouse: it measures women's anaemia, age at marriage, decision-making, experience of spousal violence, account ownership and mobile-phone access — asked directly of women.

NFHS asks about violence carefully and privately — yet even then, under-reporting is severe: shame, fear and normalisation keep many women from disclosing. The survey number is a floor, never a ceiling.

As services move online, a new gap opens: women in South Asia are markedly less likely to own a phone or use mobile internet. Data collected through digital channels then over-represents men — the privilege hazard, upgraded for the digital age.

Closing the data gap is essential — but being counted is not the same as being served. Surveillance counts the marginalised closely while denying them rights. The feminist test is not just visibility, but visibility that benefits the counted.

If data feminism had one operational habit, it would be disaggregation: never settle for the average. Break every headline number down by sex, then by caste, disability, region and wealth — and watch the hidden inequities appear.