Invisible Women: Data Bias in a World Designed for Men

All too often, however, we don’t allow women to provide that link. And so we continue to treat too many of the world’s problems as insoluble. Like Freud, we continue to ‘knock our heads’ against what seem like riddles. But what if, like representing the hyperbolic plane, these problems aren’t insoluble? What if, like the problems in broadcast science competitions, all they are missing is a female perspective? The data that we do have is unarguable: as we continue to build, plan and develop our world, we have to start taking account of women’s lives. In particular, we have to start accounting for the three themes that define women’s relationship with that world.

The first of these themes is the female body – or, to be precise – its invisibility. Routinely forgetting to accommodate the female body in design – whether medical, technological or architectural – has led to a world that is less hospitable and more dangerous for women to navigate. It leads to us injuring ourselves in jobs and cars that weren’t designed for our bodies. It leads to us dying from drugs that don’t work. It has led to the creation of a world where women just don’t fit very well.

There is an irony in how the female body is apparently invisible when it comes to collecting data, because when it comes to the second trend that defines women’s lives, the visibility of the female body is key. That trend is male sexual violence against women – how we don’t measure it, don’t design our world to account for it, and in so doing, allow it to limit women’s liberty. Female biology is not the reason women are raped. It is not the reason women are intimidated and violated as they navigate public spaces. This happens not because of sex, but because of gender: the social meanings we have imposed on male and female bodies. In order for gender to work, it must be obvious which bodies elicit which treatment. And, clearly, it is: as we’ve seen, ‘the mere sight of a woman’ is enough for the viewer to ‘immediately elicit a specific set of associated traits and attributions’.8 To immediately class her as someone to speak over. Someone to cat call. Someone to follow. Someone to rape.

Or maybe just someone to make the tea. Which is where we run into the third trend, which is perhaps the most significant in terms of its impact on women’s lives worldwide: unpaid care work. Women are doing far and away more than our fair share of this work – this necessary work without which our lives would all fall apart. And, as with male violence against women, female biology is not the reason women are the bum-wiping class. But recognising a child as female is the reason she will be brought up to expect and accept that as her role. Recognising a woman as female is the reason she will be seen as the appropriate person to clear up after everyone in the office. To write the Christmas and birthday cards to her husband’s family – and look after them when they get sick. To be paid less. To go part-time when they have kids.

Failing to collect data on women and their lives means that we continue to naturalise sex and gender discrimination – while at the same time somehow not seeing any of this discrimination. Or really, we don’t see it because we naturalise it – it is too obvious, too commonplace, too much just the way things are to bother commenting on. It’s the irony of being a woman: at once hyper-visible when it comes to being treated as the subservient sex class, and invisible when it counts – when it comes to being counted.

There is one more trend I kept coming across while writing this book: the excuses. Chief amongst these is that women are just too complicated to measure. Everyone was saying this, from transport planners, to medical researchers, to tech developers: they were all knocking their heads up against Freud’s riddle of femininity and coming away baffled and defeated. Female bodies are too unharmonious, too menstrual and too hormonal. Women’s travel patterns are too messy, their work schedules are too aberrant, their voices are too high. Even when, in the early twentieth century, influential Swiss architect Le Corbusier was devising a standard human model for use in architecture, the female body was ‘only belatedly considered and rejected as a source of proportional harmony’,9 with humanity instead represented by a six-foot man with his arm raised (to reach that top shelf I can never reach).

The consensus is clear: women are abnormal, atypical, just plain wrong. Why can’t a woman be more like a man? Well, apologies on behalf of the female sex for being so mysterious, but no, we aren’t and no we can’t. And that is a reality that scientists, politicians and tech bros just need to face up to. Yes, simple is easier. Simple is cheaper. But simple doesn’t reflect reality.

Back in 2008, Chris Anderson, then editor of tech magazine Wired, penned an article headlined ‘The End of Theory: The Data Deluge Makes the Scientific Model Obsolete’.10 We can ‘stop looking for models’, Anderson claimed. There is now a better way. Petabytes [that’s 1,000 million million bytes to you and me] allow us to say: ‘Correlation is enough.’ We didn’t need to hypothesise about anything, we just needed to crunch the numbers – or, more accurately, ‘let statistical algorithms’ crunch the numbers. In the era of Trump, Brexit and Cambridge Analytica, this seems Pollyanna-ish to say the least, but even before these data scandals it should have been obvious that his claims were hubristic, because back in 2008 we had even less data on women than we have now. And when you’re missing out half the global population in the numbers you feed your statistical algorithms, what you’re actually creating is just a big mess.

Anderson holds up Google as an exemplar of what he dubbed ‘The Petabyte Age’, singing the praises of its ‘founding philosophy’ that ‘we don’t know why this page is better than that one: If the statistics of incoming links say it is, that’s good enough. No semantic or causal analysis is required. That’s why Google can translate languages without actually knowing them (given equal corpus data, Google can translate Klingon into Farsi as easily as it can translate French into German).’ Except, as we’ve seen, Google actually can’t translate very well at all, even ten years later. That is, if you care about women being erased from language.

So. Not so simple after all.

Anderson is right about one thing though. There is a better way. And it’s a pretty simple one: we must increase female representation in all spheres of life. Because as more women move into positions of power or influence, there’s another pattern that is becoming even more apparent: women simply don’t forget that women exist as easily as men often seem to.

Women in the film industry are more likely to employ women.11 Female journalists are significantly more likely to centre a female perspective and to quote women.12 Female authors do the same: 69% of US female biographers wrote about female subjects in 2015, compared to 6% of male biographers.13 The emphasis by women on female voices and perspectives extends to the academy. Between 1980 and 2007, female history faculty in the US rose from 15% to 35%14 – meanwhile across a similar time period (1975-2015), US history faculty specialising in women’s history rose from 1% to 10%15 – a tenfold increase. Female academics are also more likely to assign female authors to their students.16

Then there’s how women might interpret history: in a 2004 Guardian article comedian Sandi Toksvig wrote about how when she was studying anthropology at university one of her female professors held up a photograph of an antler bone with twenty-eight markings on it. ‘This,’ she said, ‘is alleged to be man’s first attempt at a calendar.’ We all looked at the bone in admiration. ‘Tell me,’ she continued, ‘what man needs to know when 28 days have passed? I suspect that this is woman’s first attempt at a calendar.’17

Caroline Criado Perez's books