I first became aware of Caroline Criado Perez through her campaign to keep a woman on the reverse of UK bank notes, after Elizabeth Fry was replaced by Churchill on the five pound note. This campaign was memorable firstly due to her success – Jane Austen now appears on the reverse of the £10 note – and secondly, from the deluge of threats, hate mail and acrimony she attracted through Twitter as a result. At the time, Twitter did almost nothing about this – the situation is (only) slightly better today due the changes they have made to the way abuse is reported.
Undeterred, Criado Perez published “Invisible Women: Exposing Data Bias in a World Designed for Men” in 2019. Her book is rather neatly summed up by the quote from Simone de Beauvoir she includes in the frontispiece:
“Representation of the world, like the world itself, is the work of men; they describe it from their own point of view, which they confuse with the absolute truth.”
The gender data gap
Criado Perez’s aim is to flag up the huge data gap that exists about the lives of half of humanity. She dubs this the ‘gender data gap’ and it crops up again in again in almost any context you can think of – medicine, product design, protective equipment, town planning, governance. The silence of women’s voices in these areas leads not just to irritations, such as too-cold offices or phones that don’t fit your hand but also to life threatening situations. From stab vests that don’t fit female police officers’ bodies, to cars that are 47% more likely to seriously injure women drivers, to medicines that do not work for women, or actively make them sicker, the assumption that the average male represents the average human is causing unnecessary harm.
Female-specific concerns that men (mostly) fail to factor in crop up repeatedly in the many areas that Criado Perez examines, but fit into three themes: the female body, women’s unpaid care burden and male violence against women. Males of course do experience violence, lots of it, but as Criado Perez says, “…it manifests itself in a different way to the violence faced by women.” Facilities such as suburban housing centres, travel networks, homeless shelters and refugee camps are usually planned by men, and do not take into account the types of activities women need to engage in, nor do they keep them safe while they do them.
There are multiple examples in the book of how women are missing from our data. Data is not only not collected about women, when it is collected it is then not disaggregated by sex. For example, few medical studies or trials specify the sex of the participants. When they do, participants are usually overwhelmingly male. If the sex of the participants is revealed, the results are not always then separated by sex. Even tests on animals or single cells are not often carried out on male and female animals or cells, even though research shows the results are likely be different between sexes. The most common adverse drug reaction in women is that the drug simply doesn’t work, putting their health and sometimes their life at risk as a result. It is likely that many drugs only make it to market because they are effective in men in early trials – anything that might have been a good treatment candidate for women alone is screened out at an early stage because it is not effective in men. And that is before you even address the woeful lack of research into conditions that principally affect women, such as period pain or endometriosis.
A scarily prescient section in the book describes how a lack of sex-segregated data can impact during a pandemic. We know from previous coronavirus epidemics, such as SARS, that symptoms can be more severe in pregnant women. During the last SARS outbreak in 2002-2004 in China, pregnant women’s outcomes were not consistently tracked. “Another gender gap that could so easily have been avoided, and information that will be lacking for when the next pandemic hits,” writes Criado Perez. Here we are, in the middle of the worst pandemic most of us can remember, still without this information. Is data being collected now on outcomes for pregnant women, or will we remain in the dark for the next one, and the one after that?
Gender blind is not always gender neutral
Another gender data gap exists where supposedly ‘gender blind’ neutral policies have an unintentionally discriminating effect against women. For example, US academics in the tenure track system have 7 years to achieve tenure. The years between completing your PhD and receiving tenure, ages 30 to 40, are when most women are likely to have their children. The result is that mothers with young children are 35% less likely than fathers to get tenure track jobs. A ‘gender blind’ policy to give all US parents an additional year to achieve tenure actually decreased mothers’ likelihood of being successful compared to fathers. The extra time gave fathers an advantage over their male peers, while the bulk of childcare and recovery from birth fell to mothers and comparatively decreased their chances.
We are seeing the same phenomenon appearing during the COVID-19 crisis – while everyone attempts to work from home and take on home education, according to Nature, women seem to be publishing far less compared to their male peers. The crisis seems to be gifting additional time to male academics to write up their research and submit grant applications, while at the same time robbing female academics of their chances, as they spend extra time caring for families, home-schooling and prioritising their students ahead of their own research interests.
The burden of unpaid care work
Academia is just one area where women do far and away the greater share of unpaid care work, to the detriment of their careers and to national productivity (GDP). A study of working patterns during the COVID-19 lockdown by the Institute for Fiscal Studies and University College London in the UK shows that due to the disproportionate childcare and housework burden, in households with home-working mothers and fathers, men have three times the uninterrupted work time that women do.
Even in normal times, the world cannot function without this care work – looking after children, elderly relatives, the vulnerable and disadvantaged. The vast majority of this work is unpaid and is carried out by women on top of their hours of paid work. They fit in multiple extra, short trips every day to support this unpaid work, dropping off children, doing shopping, seeing relatives. These journeys are poorly supported by the radial transport networks designed, largely by men, to serve the traditional daily commute from home to office. As we ‘clap for carers’ every Thursday to express our sincere and heartfelt gratitude for what care means during an international crisis, we shouldn’t forget that the caring burden is a daily reality for most women. For the moment, many of our commuter transport systems are empty. Post COVID-19, should we really go back to investing more and more money in systems that whisk us from home to office, but leave local neighbourhoods under-funded and under-served?
Building bias into the system
While we might be able to understand the presence of bias in humans, it can be tempting to rely on machines to fix the problem. Surely computers are neutral, with their artificial intelligence and gender blindness? Unfortunately, Criado Perez explains why this is not the case, because a large gender gap exists here as well. She describes how women are hugely under-presented in image and speech datasets. Speech recognition technology in smart speakers, phones, medical devices and cars are trained on male voices and struggle to respond accurately to women’s voices. Not only that, the images and text databases that AI systems train on are just as biased as humans, which is not surprising as they are generated by humans. So not only are datasets lacking in data from half the human race, the information that is in those datasets is biased towards gender stereotypes in the same way that humans are unconsciously biased. I encountered more research on this area at the Gender Summit in 2019. This has a real impact on outcomes for women when CV selection systems and even medical diagnostics are becoming increasingly automated using AI.
An individual perspective
If I have a criticism of Criado Perez’s book, it would be that the experiences of one person are sometimes used to make a point about the invisibility of women in general. Anecdotal evidence is still evidence, but does that person represent many? On the other hand, the whole point is that bulk data is lacking in many areas for all the reasons outlined above, so perhaps it’s understandable.
I found reading this book an eye-opening but ultimately rather sobering experience. Getting into a car to drive, will I feel as safe having read it? I certainly won’t stop feeling absurdly irritated by the smart speaker at home that responds instantly to my husband’s voice but stubbornly ignores mine until the third or fourth attempt. I’ve lost count of how many times I’ve nearly dropped my new phone trying to use it to take a photo one handed. As a materials science student at university, the protective equipment we used for welding or casting metals practically drowned me – it seems unlikely it made me safer if I could hardly move without tripping over it. Who knows how many times my CV didn’t make the cut for a science job due to skewed AI algorithms? I need to work flexibly and part-time to fit round my roles as mother of a special needs child, school governor and fundraiser for the National Autistic Society .Realistically, this limits my career options. Minor points on their own perhaps, but over a lifetime they add up, they really do.
Read this book – it will certainly make you think.
KatieGee 