334779_en-US

liliconcom

The Gender Data Gap

Ever wonder what happens when entire segments of a population are ignored and gathering data about them isn’t even considered? Simply look around. Author of “Invisible Women,” Caroline Criado Perez joins us to explore the gender data gap and invisible bias that has a profound effect on women’s lives. After sharing case studies and research results, she will answer your questions about the hidden ways in which women are excluded from the very building blocks of the world we live in.

 

 

Auto-generated transcript...

 

Speaker	Transcript
Anne Milley, JMP	It is excellent and so important, this book. Can't recommend highly enough. So Caroline Criado Perez is a best-selling and award-winning author, feminist and activist.
	Her book, Invisible Women, has been translated into 24 languages and won numerous awards. Caroline herself has won several awards. Earlier this year, she was designated or named the OBE in the Queen's Birthday honors 2015.
	And earlier this year, she was the recipient of Finland's Hand Award for her social promotion of equality. We are so delighted to have her with us today. Please welcome Caroline Criado Perez.
Caroline Criado Perez	Hi, Anne. Thanks very much for that lovely introduction.
	I can't believe that the Hand
	Award was this year and it feels like so long ago, but yeah, it was actually March, just before we all went into lockdown. So I hope you're all busy voting on the poll, because I'm really, really keen to see how you will vote.
	Ideally, before I start talking too much and but I'm going to tell you a bit. What I'm going to talk about. And I'm going to try and talk as sort of quite quickly.
	And because there's a lot that I want to cover and obviously we want to leave time for questions. So what I'm going to talk about is a bias that affects us all, me as well as you, and I'm going to show you how this bias affects our collection and analysis of data, which in turn has
	a serious impact on how we allocate resources and design pretty much anything.
	And so before I start, let's just see what the results are of the first poll, which I'm very excited about. Are you aware of the data gender gap?
	That's very good. 65% aware of the data gender gap, 35% not aware.
	So you're going to all be aware by the end of this. Women in your design team? Wow, that's very high, very good. Is your design team gender balanced?
	Less good. Well, I'm going to make a very strong case for why your design team should be gender balanced. And so without further whatever I'm going to share my screen and let's get going.
	So I hopefully you can all see that. And so first of all, and when I'm talking about the gender data gap, I'm talking about gender...sorry, data.
	Let's define our terms to start off with. So what do I mean when I say data?
	What we tend to frame data as a tangible substance that has its own inherit properties, irrespective of how we engage with it. We think of data is some somehow sort of sitting out there in the ether fully formed, waiting for us to extract it,
	And which is understandable, right? People trust data because numbers feel real to us. They feel objective, they feel neutral. They're not sort of messy opinions.
	But data is actually much more tricksy than that because data doesn't actually exist apart from us, instead of suspended
	animation. Data is not a discrete entity, it is intrinsically, subjectively human and that means that we can't simply treat data as objective neutral facts, upon which we can base our objective neutral
	Decisions. Data, like the humans who have created it, is slippery and shaped by a whole host of factors that we must take into account, whenever we are dealing with it. Questions like, who has done the data collection and how have they done it?
	Who have they gathered the data from?
	What questions have been asked to elicit this supposedly neutral data? Or even before that, what data have we even chosen to collect? Right? What is
	the problem even that we think is really important, that we must solve, that has caused us to collect this data in the first place? And none of these
	are objective decisions, and they all have an impact on the data that ends up in our possession and the stories that those data tell us. So just to give you an example,
	labor force participation rates statistics in low-income countries have, for a long time, been notoriously unreliable when it comes to counting female labor participation.
	And because they have a history of undercounting them. And the reason for that is the way that the questions were framed and asked, in that they were framed around
	the way men have tended to work in the West. So that's paid employment that tends to take them out of the home for a substantial part of the day.
	But women in low-income countries don't oftentimes to work like that. They work often unpaid from the home, which includes unpaid care work, but can also include things like agricultural work, subsistence farming of that kind.
	Of course they carry out work that in the West, we have outsourced to utility companies. So going out to fetch firewood and water.
	But the questions aren't framed to capture this style of working, which means that women are often framed as economically inactive when that is
	of course, far from the case. So this obviously doesn't just affect female labor force participation rates. It also ultimately affects GDP, for example.
	And let me just show you another example of how data can be corrupted by poor collection methods.
	So this chart on the left, this is how transport planners have traditionally collected data. So as you can see, all paid employment data goes together in one block.
	But they haven't aggregated travel done for care work purposes. In fact, that's been split down with lots...within lots of separate categories like shopping, escorting visits. You know, so grocery shopping is mixed in together with
	food shopping for example, sorry grocery with SHOE SHOPPING (I clearly have food on the mind) and
	as you can see that makes it look like employment is by far the biggest reason that people are traveling. And looking at the table on the left, it makes sense, of course, to design
	public transport around travel for paid employment. But if you look at the table on the right and that tells us something different. And the table on the right was put together by an urban
	planner in Madrid who collected care-related travel data and and aggregated it all together in one block as we do for paid employment data and traditional travel data aggregation.
	And what she found was that the number of trips made for caring purposes almost equal those made for employment purposes.
	This is a gender issue why? Well basically because the vast majority of unpaid care work around the world is still being done by women.
	So globally women do three times the amount of unpaid care work men do. According to the IMF, this can be further subdivided into twice as much childcare and four times as much housework.
	And this affects how women travel. So when Sanchez-???, this urban planner in Madrid, further refined the data by sex, disaggregating it, she found the care was the single greatest purpose of travel for women.
	And now designing public transport to support women's unpaid care work isn't just a matter of justice. It's actually an economic decision.
	So countries all around the world have for a long time been desperate to get more women into work again, why you need that accurate female labor force participation rate data,
	more people working equals more tax revenue equals higher GDP, higher productivity.
	But actually also very interestingly, it's been found that increasing the number of women in work increases GDP by more than increasing the number of men in work.
	They don't know why that is, but I mean, it's probably to do with, you know, declining
	rates of return. But nevertheless, you know, that is the case. There is a very, very clear economic case for increasing the number of
	women in work. And if you want to do that, you need to create a structure, whereby that unpaid work is being supported, which of course we're going very much backwards on
	during this pandemic, but I'll get into that a bit more later, assuming I don't talk too much and run out of time. Anyway,
	something you might have noticed about both those examples, was that the data was always skewed to the detriment
	of women, so women were the ones being undercounted and this is not a coincidence. In fact, most data sets in the world that you care to look at will share this problem of undercounting women.
	And that is simply because the vast majority of information we've collected globally and continue to collect, everything from economic data to urban planning data to medical data, has been collected on men.
	So that's male bodies, but also typical male lifestyle patterns like how men work, how men travel.
	And this means that pretty much everything in the world, from the office that you may not be working in right now but you're used to until a few months ago,
	and from the public or private transport that you might have used to get there, to the phone in your hand, to the apps on that phone, to the medical treatment you receive
	have been designed to work for men because that's the data that we have.
	And the result is that most things in the world just don't work as well for women. This is the gender data gap that I think 65% of you said that you'd heard of.
	So hopefully now, one hundred percent of you know what the gender data gap is.
	Now the first rule of the gender data gap, this is really important. If you take nothing away from this, please take this away,
	is that the gender data gap is not the product of a conspiracy. It is not deliberate. It is not the evil brainchild of a group of misogynistic data scientists who hate women and want us to die.
	It is simply the product of a bias, the bias that I mentioned at the beginning, but we all share, and which
	is itself, the product of a way of thinking that goes back millennia. And this this way of thinking is centered around a very good friend of mine called Reference Man,
	who is the world's worst superhero. But before I introduce you to Reference Man, I would like to start the second poll,
	And as you should be able to see this is specifically aimed at manufacturers and engineers. So those of you in manufacturing or in engineering, please answer these questions. And in the meantime, I'm going to introduce you to Reference Man.
	Here we go. Here's Reference Man. As you see, he's a Caucasian male, 70 kilograms, around 25 to 35 years old.
	As I said his superpower is (I might not have said this, but I'm telling you now) he has superpowers to represent everyone. He's not very good at this, because he is just one type of person.
	And anyway, Reference Man comes from, as I said, a long history of thinking about the average male as somehow the average human. And in fact, this way of thinking goes all the way back to Aristotle. Oops.
	There we go. Here's Aristotle. Aristotle was bemused by women and and generally felt that and yeah women...women were...women were very confusing. Women were...why did women exist?
	We were a sort of subtype of men, an aberration was was the word that he used. But the sort of best quotation, I guess you could say that really, really nailed down his thinking on Reference Man
	was to be found in his biological treaties on the generation of animals,
	in which he wrote, \!"The first departure from type is indeed that the offspring should become female instead of male.\!" So you see him very clearly articulating there that male is the type from which women deviate. That was just his standard position.
	Moving on, you find this way of thinking continuing through into the Renaissance, where there was this explosion of medical illustrations. People really got into it around the Renaissance.
	And you can see very, very clearly in the way that they represented the female body
	that they were still thinking of the female body is somehow a deviation from the male body. So it was literally the female body with the male body turned outside in.
	So the reproductive organs, why are the reproductive organs in females inside instead of outside? Well, they must have retreated inside and and their explanation for that was that
	it was a female deficiency in vital heat. Basically female bodies really cold and so the poor reproductive organs needed to hide up inside the body. And because they were just a female version of the male one, so you had
	as you might be able to see here, a picture of the female scrotum and the female testicles. But what sort of I find quite fascinating about that is you could easily turn that on its head, right, and say that this is
	this is just that it's actually men who are deviating from women, right? That men have too much vital heat and that's why,
	you know, these very sensitive reproductive organs are sort of dangling down outside the body, unprotected. Just, just a thought of mine.
	this form of thinking the female is somehow deviant continued all the way through into the 20th century, and this is Modular Man.
	And the human scale, Modular Man within the human scale, you might be able to see where I'm going with this human scale.
	Human scale was dreamt up by an influential Swiss architect in the 1920s called Le Corbusier.
	And as I said, he came up with this idea of the human scale in architecture. Humans are going to be in buildings; buildings should be designed around humans.
	The issue was that for him, human meant man, and specifically, six foot British police detective, which I was thinking is kind of charmingly specific.
	Actually someone did...there is evidence that someone did say to Le Corbusier, \!"Mate, women exist, maybe you'd like to consider
	including them in your human scale.\!" And so he did. He did consider women. But ultimately, the female body was rejected as a source of proportional harmony, which
	you know, I certainly recognize about my body. Now you may be thinking that this all sounds very silly, but it's a long time ago now, and surely we know better. We know to include women in design. Well, let's just see the poll results from
	the poll we just ran. It'll be very, very interesting to see
	what...I can't see the poll results. Well, I'm going to carry on and hopefully the poll results will pop up at some point.
	And so let's carry on to find out what
	when designing a product, do you collect sexless aggregated data as standard?
	So that seems like not many do. Do you test your products on women as standard. Again, that's not an overwhelmingly positive response.
	do you always consider gender as a variable? And so I have to say that's not that surprising,
	because actually, that's extremely common, based on based on my research. But here is why you absolutely should always be considering sex and gender as a variable and including women in your product testing and and and also your product design. So
	back to the slides. Car crash safety. This is a classic of the genre, of the Reference Man genre.
	So for decades, the only car crash test dummy that was used was based on a 50th percentile male, so literally Reference Man,
	as if the 50th percentile male is representative of the 50th percentile human, if there is even as such a thing, which I would
	I would argue against. But the result is that cars just aren't designed to be as safe for the average woman. So seat belts have never been designed,
	have never been tested with breast tissue. We have yet to design a seat belt that is safe for women to use in pregnancy, particularly during the third trimester.
	And the result of this is that women are often found to be wearing seatbelts in what is called improperly, basically because you know it's cutting into their neck or they can't get it to lie comfortably across their breasts.
	And this means that they tend to be wearing them unsafely, basically if you're not wearing it right, you're not you're not being as protected as you should be. So it's a huge safety issue.
	There's also the issue of the seats, which are too firm. Now this isn't a matter of comfort, this is actually
	about the the body that the seats have been designed to absorb. So obviously the body of the 50th percentile male is bigger and heavier than the average female body and
	so because the seat has been designed to absorb the weight of that body, women get thrown further forward in the event of a crash.
	And and obviously that puts them at higher at higher risk, which is even worse given women are already tending to sit further forward
	in order to be able to reach the pedals and reach the steering wheel, you know, pretty important part of driving. But that puts them at a higher risk in a frontal collision.
	And when you add that to, you know, being thrown further forward, the seat belt possibly not being worn properly, and issues with the headrest (which I'm not going to go into right now, but there's some really interesting stuff if you look into that about
	attempts to introduce better headrest and how they make things better for men and worse for women, again because there wasn't this average female car crash test dummy),
	and in fact, the result of all of this is that if a woman is in a car crash, she's 47% more likely to be seriously injured and 17% more likely to die than a man in the same car crash.
	Now, there is some good news. Sorry, I'm just looking around to see if I can find a light, I just realized I'm going a bit dark here
	and that's probably not great for all of you. And so, yeah. The good news is that the EU in (I'm sorry, just trying to get a light on) the EU in 2015
	belatedly realized that women do, in fact, exist and
	Introduce what they call a female car crash test dummy.
	It's actually a very, very scaled down male dummy. And of course, the issue with that is that women are not just scaled down men, right. There are various other differences, including the aforementioned breast tissue.
	But also, you know, pelvic differences, there are differences in spinal column flexibility, there are differences in terms of weight...sorry...muscle mass distribution, so men carry it more on their necks and their
	shoulders, the upper body, which obviously makes a difference when it comes to biomechanics and car safety. But also this car crash test dummy,
	I should also mention it's it's very, very scaled down to about the size of a 12 year old
	child, so very much not representing the average female. It's only used in any case in one out of the five regulatory tests
	and only in the passenger seat. And so basically in the EU, very scaled down men can can sit in the passenger seat and feel fairly safe.
	And I just want to, to bring your attention to a rather fun twist I came across the other day.
	So this is from a study looking at the optimal comfortable seating...sleeping position, sorry, for someone driving in or being driven by an autonomous vehicle.
	And as you can see, they only use male subjects for the sake of homogeneity. So even here, when it comes to sleeping comfort, we are sticking with with just men,
	and I guess, you know, I guess, that safety for women can wait. I guess that's that's the, that's the position on that one. And we're going to move on from that, because it's a bit depressing, into personal protective equipment,
	which is something that I'm sure we've all heard a lot more of than we expected to this year.
	So before actually PPE and supply chains of PPE became such headline news,
	they were a very serious problem already for women who worked in industries where PPE is required.
	And, in fact, a study in the UK found that 71% of women who work in occupations where personal protective equipment is required,
	where PPEs had not been designed to fit their bodies and the effects of this can be serious. So the same report, in fact, found that the emergency services were worse for this, where only 5%
	of women said that their PPE never hampered their work. And this included things like body armor and stab vests. I'm sure you don't need me to tell you that if your
	stab vest doesn't fit you properly, it's not protecting you. So the two options, basically, that women were saying that they had was either be too big
	(and so you'd have all this extra material up here) or extra stab vest, I should say, because obviously stiff. It's not sort of folding over
	And that meant it wasn't fitting tightly to their body on their shoulders. One woman talked about being able to be picked up by her stab vest, but more importantly, it actually leaves
	your, your arteries exposed, which means that it's not being useful as a stab vest.
	The other, the other thing was that they might get one that...sorry...they would get a big one in order to accommodate their breasts. What I should have specified was that they're designing stab vests without
	accommodating breast tissue again. So you're going to get them big to fit your breasts, well, then you've got this problem.
	Don't accommodate your breasts and then they are too small and they ride up, because, you know, the boobs push them up, leaving the abdomen exposed, and again, sort of failing the stab vest test.
	And, and I think that's a pretty useful example of how
	wrong thinking this idea of Reference Man is, thinking of the male body is somehow this mutual body. I mean it isn't.
	And you would never ever, you know, design a stab vest with breast tissue accounted for and say, here you go, this is a unisex stab vest everyone can wear this. You know, that would be clearly ridiculous and yet somehow
	you know, because of this bias, but you know we all kind of have we do kind of think of the male body is somehow a gender-neutral body.
	But this kind of thing really highlights how illogical that is, and, and, I mean, I personally find it really fascinating, but that we somehow think that, you know, why do we think that?
	And so anyway, as I said in the past few months, the issue of unisex PPE leaving women exposed has become even more acute. But before I go into that I'd like to run the final two polling questions.
	Okay. And these are just for people who work in the pharmaceutical industry. I'm very very interested to see the results of
	this poll. Anyway, so I'm going to talk a bit about Covid and PPE now. So basically, ever since the pandemic started and...
	so I just got a bit distracted and basically because my mom just got a positive Covid test result back. So there's suddenly all feels very, very real to me in a way that I guess it didn't
	12 hours ago. But anyway, women have been sending me pictures of themselves and messages saying that they their PPE does not fit them. It's just too big and women have been complaining about everything from
	scrubs to gowns to gloves, which are all very important. You know scrubs can be, and gowns can be a trip hazard.
	Gloves that are too big make it very difficult to do things like an intubation, which obviously is a very
	significant part of Covid care. But the number one thing that women have complained about is their masks and not be able to get a good seal on their masks, and of course, in a pandemic like this,
	it's extremely important, particularly for healthcare workers who are going to be very exposed.
	Women have said that they've really struggled to find a mask that passes a fit test on their face. I actually also did a freedom of information request to
	National Health Service Trust around England and Wales, and all the results that I got back on fit test did show that women were far more likely to be failing these
	fit tests, and when men were failing was and is because of stubble. It's another really big design flaw of these masks, is that if you even have
	you know, a day's worth of stubble, you're likely to fail the fit test.
	Um, but anyway. The problem is particularly acute for women, because they have been designed basically to fit the default male face. And even the the sort of the smaller ones,
	in order to get a seal, what what the workers are doing is just pulling them really really tight. Because the problem is, you know, you want to go in and be able to work and you can't work unless you pass your fit test. So they pull the mask
	really tight actually tighter than they probably would wear them in reality and and then that's how you end up with pictures like this going viral.
	And there is actually a fairly extensive body of research prepandemic, showing respirators are most likely to fit white male faces with everyone else trailing behind. I've actually been on the hunt of
	size specifications used by manufacturers and and the data that was used to determine those size specs, but
	so far, been unlucky. Regulators haven't got back to me, manufacturers have sent me to regulators or or pointed me towards regulations that only mentioned filtration rates, don't mention size specifications.
	And so I'm kind of starting to think that they don't actually exist. But anyway, if anyone has any information on that, I'd be extremely grateful if you would get in touch because I'm desperate to find out,
	you know, what are they? The data must exist somewhere. I did actually magnage to get a hold of one data set from the US, from the CDC.
	It was a male-dominated data set and the CDC told me that they didn't separate the data by sex.
	They simply used averages from the data verall, and you can obviously see what the problem there is if it's a male-dominated data set and you're not treating the male and female data separately.
	And meanwhile, the other day, this study came out, which again, really important study, great that they're looking at this, but why
	why are they testing maskes on a single man with a woman for a few comparisons. This is illogical
	in any situation, but particularly when it comes to health care, because globally, the healthcare workforce is heavily female dominated.
	It's around 70% female, goes up to 76, that's globally, goes up to 76% in the EU and the US.
	The figure is higher when it comes to nurses, that's around about 85 to 89% depending on which country you're looking at. Of course, nurses are the ones doing the more intimate up close to the body care.
	And and that's before you talk about things like carers, certainly in the UK, the epicenter...well in a lot of countries, the epicenter of the epidemic has been in care homes.
	And and in the UK, carers represented the number one occupational category for deaths from from Covid in the first...in the first wave.
	So if anyone should be the default when it comes to medical PPE, it should be women.
	I mean you shouldn't be using a default, you should be using representative data. But if you're going to insist on using a default, it really doesn't make sense in this context to be using men. Anyway, moving on and
	let us have a look at some...wait no, sorry, not the polls. Okay, fine. We'll look at the polls now.
	Now the polls have gone. Okay, I'm I was just going to go through these these medical illustrations. So here you go, here and you may remember the medical illustrations from the beginning of the talk. This is
	the modern version. As you can see, we've got standard pelvis, the standard genitals, female pelvis off to one side, where she belongs.
	Then you've got here the standard muscular system, the muscular system and then the female muscular system.
	Obviously we can't have a flayed person without gender stereotypes. The man is looking very strong a manly. She's looking very sexy, which I respect. I respect that. And anyway, I would like to see the poll responses now, if possible.
	And
	So do all your trials include females, as well as male? That is very good to see. You are an unrepresentative bunch, I think.
	Although I suppose it depends how many females. I should have asked that. But very, very naughty that it's about of those who this applies to 50/50 who are sex disaggregating your data. Always sex disaggregate your data. I'm going to try and convince you that now. So, um,
	Well, we don't have a lot of time for me to convince you, but just just accept you have to sex disaggregate your data.
	So the effects of Reference Man in the world of medicine is that the vast majority of our medical knowledge is based on the male body so female humans, animals, and even female cells
	have been historically, routinely excluded from clinical research. And this is mainly been historically an issue for women who suffer from more adverse drug
	reactions to drugs and vaccines and are more likely to be misdiagnosed for a whole range of diseases.
	But as we've seen in the Covid 19 pandemic, it is actually also an issue for men it turns out. So we know far less about the female immune system than we do about the male immune system. And so now we don't actually very well understand the mechanics
	behind why women have been surviving this disease better than men, right. Is it chromosomes? Is it sex hormones? Is it both? We just don't actually really know.
	And that lack of knowledge is seriously hampering us when it comes to developing effective treatments. I was very interested to see this
	trial announced fairly early on in the pandemic. I haven't actually been able to find out the results of it.
	So I don't know if they're just still pending or what's happened there. But as you can see, they started injecting their male Covid patients with estrogen to see if that would help them
	survive. You know, that's the level of infancy we're at when it comes to how the mechanics, you know, the mechanics of the female immune system.
	And it really reminded me of this cell trial that I came across in the course of my research for Invisible Women. A 2016 cell trial where male and female cells were exposed to estrogen and then exposed to a virus.
	The female cells were able to use the estrogen to fight off the virus. The male cells were not able to use the estrogen to fight off the virus. So there is a very good example of why you must sex disaggregate your data. Another example
	that was very interesting I came across was this issue with
	stem cells from muscle tissue and where it was thought that they were unpredictable
	until someone disaggregated the data and realized that female cells promote regeneration, male cells don't. There was recently a really interesting study, I say recently -- recently when I wrote the book,
	study where...to do with circadian rhythms and survivorship in heart attacks. This was in animal studies and
	for years there had been one type...there was this gold standard idea about
	survivorship based on circadian rhythms and then this one study came along that up-ended the whole thing and everyone was super confused. What on Earth is going on?
	Well, it turned out that all the previous studies have been done in male mice and this study had been in female mice. Anyway, so sex disaggregate your data,
	which we are not doing during this pandemic, where it could not be clearer that there are sex-specific impacts of this of this disease.
	This was the state of play back in March. By the end of March, only six out of the 20 most effective countries were publishing properly sex disaggregated data.
	It took until April for the US and the UK to start reporting this data...
	to start reporting...sorry... sex disaggregated data on deaths, and until May for them to report sex disaggregated data on cases, which obviously you need. You need to know whether, you know, men were more likely to be contracting it
	or more likely to be dying once they contracted it. But because we weren't collecting and publishing this data, you know, there was a lot of time wasted at the beginning of the pandemic and
	things aren't still very good. By September only 30% of the world's countries were reporting sex disaggregated data on cases and deaths.
	And fewer than half of high-income countries are publishing this data. So it's not a matter of resources. It's a matter of priorities and also actually one in three...and sorry, some countries that were
	reporting sex disaggregated data have since stopped. Um, anyway. I, oh, I think I'm running over, but I really want to talk to you quickly about algorithms. So
	as you can see, only middle aged men get heart attacks. This is, I'm as I'm sure your pharma colleagues could tell you, very much not the case.
	Heart disease is in fact the leading cause of death for women in Europe and the US.
	Women from lower socioeconomic backgrounds are 25% more likely to suffer a heart attack than men in the same income bracket.
	Women are 50% more likely to be misdiagnosed if they have a heart attack than men, and more likely to die following one. Young women are almost twice as likely as men to die in hospital. And there have been concerns raised over commonly used risk prediction models,
	because they were developed on patient populations that were at least two thirds male.
	That concern was raised by the American Heart Association back in 2016. So why do the Google image search results present heart attacks as a middle aged men thing.
	Not because of conspiracy, but because that's the data it's been trained on. It gets a lot worse than some misleading Google images.
	So, towards the end of last year, and there was a study that was reported all over the UK news media, headlined \!"heart attacks
	to be predicted AI...sorry...AI to predict heart attacks, five years before they happen.\!" Right? It sounds amazing. This is what we dream of. This is what we're developing AI for.
	But when you look at the paper, actually the studies and...that that you know...the data that was used for all three studies was heavy male dominated.
	The paper itself provided barely any sex disaggregated data
	and it certainly didn't mention things like the fact that diabetes is a higher risk factor for women,
	that smoking is a higher risk factor for women. So what kind of job is this AI going to do if it doesn't even account for the sex differences when it comes to...when it comes to heart disease?
	And even more scary than that, of course, is that algorithms don't just reflect our biases back at us. Right. This isn't going to be just as bad as any other biased doctor, it's going to be worse because machine learning
	amplifies our biases, in fact, and by significant amount. So to give you an example in 2017, there was a quite worrying study that came out
	quantifying this amplification bias. What they did was they trained an algorithm on this very commonly used image data set.
	And this data set had this bias, whereby pictures of cooking were surprise, surprise, 33% more likely to include women than men.
	But by the time algorithm was done being trained on it, it was labeling pictures...it was connecting pictures of cooking with women 68% of the time. So, you know, it was labeling pictures of men
	as women just because they were standing in front of a stove.
	And the paper also found that the stronger...the higher the original bias, the stronger the amplification effect. Certainly in the medical world because of historical failure to include women in in research, as well as an ongoing failure in many cases, we have a pretty large bias.
	So, um, how do we address this?
	This is how you don't address it. The Henry Higgins effect. So the Henry Higgins effect, for those of you who aren't great big fans of musical theater, in which case I suggest you
	rectify that,
	this comes from My Fair Lady. Henry Higgins is a is the protagonist, I guess, one of the protagonists
	in My Fair Lady and he picks up Eliza, a flower girl, and says, I'm going to treat you...I'm going to train you to be posh and present you at the embassy ball and everyone will think you're a lady. I'll teach you how to speak proper,
	but she is much more sort of recalcitrant then he expects and he gets very frustrated and he sings this big number, \!"Why can't women be more like a man?\!"
	That is what I've called Henry Higgins effect because it was something that I kept coming across in the course of my research. Very often, very well meaning attempts to fix
	any disparities, gender disparities, gender inequities was well let's train women to be more like men
	and that will fix the issue, whatever the issue may be, rather than looking at potentially systemic problems. So one example that that I thought was quite funny came from
	a voice recognition software supplier who talked about the fact that his voice recognition software was less likely to
	recognize women's voices. This is actually pretty common and it's because most voice data sets are heavily male, it's pretty clear what the fix is there. You just need to
	collect more female voices, which shouldn't be that difficult to do. But anyway, this guy said that perhaps women should undergo hours of lengthy training in order to fix the many issues with women's voices.
	But sadly, women didn't seem to want to submit to that, which is just women all over, isn't it? Then but, you know, you see it in in sort of things like
	workplace initiatives, really well-meaning workplace initiatives like let's train, train women to be more confident to ask for that pay rise, ask for that promotion.
	Well, actually the evidence suggests that women are just as likely to as men to be asking for a pay rise. They're just less likely to get it, which leads to one very important part of
	training women to be more like men, which is the actually women don't get treated the same when they behave
	like men. Actually there's a very good reason that women don't behave like men, it's not that we're illogical, irrational, it's that we know that we won't have the same reception
	if, you know, we seem to be pushy or bossy or strident all these words that that get used about women.
	But also, you know, what I find interesting is, again, this assumption that men have it at the right level, that what men are doing is the correct
	is the correct level and and you really need to watch out for that and correct it. Because often that leads you astray.
	And I, I was going to talk to you about three stone stoves, but I really don't have time.
	So you're just gonna have to buy the book and read about them because it's a very, very interesting case study when it comes to design and collecting data on women before you design something.
	So I'm just going to sum up with my three rules for considering gender in your decision making. So number one,
	This is a classic picture of women
	queuing for the toilet, by the way. Even in the Army, women have to queue for the toilet. Number one, remember that women are not failed men. Okay. So don't think of
	men as being the standard, the neutral, and and how do we get women to be more like that, or why is it that women are different from men. Just treat women is as perfectly rational
	Diversity, diversity, diversity. Let me just give you one example that I think really sums up why you want a diverse team in your...
	a diverse design team, which is one of the polling questions I had. So in 2014 with much fanfare, Apple released its health tracker app.
	And it was a comprehensive health tracker app, they said. And it was very comprehensive, you could track your
	molybdenum, you could track your copper intake. I'm really annoyed with myself and not thinking to do a poll question on that because I would really love to know.
	how many of you track your molybdenum and your copper, and also how many of you know how to, because I certainly don't know how to track either of them.
	But that was a missed opportunity on my part. And if someone can subtly spring up a poll question. And anyway, so you could track those but you couldn't track your period.
	Now I'm going to go on a limb here and say, far more people track their periods than track their molybdenumhem or their copper intake.
	Um, but the real question is why did Apple exclude it and there are two options. One,
	they hate periods, they think periods are disgusting, they don't want them in the health tracker app. Or they forgot periods exist. And I think it's more likely that they forgot periods exist. And why did they forget periods exist? Because they didn't have enough women on their design team.
	So that's just a little parable for you there. And finally, third and most important, always always always collect sex disaggregated data. You never know why, when it might matter.
	Just to close with one more example, even if it doesn't matter for your own study, it could matter for later meta analysis.
	One study I came across when I was researching Invisible Women was looking at a CRTD, which, for those of you who don't know, is it is a sort of
	basically very sophisticated kind of pacemaker. They found by doing meta analysis of the previous studies that had sex disaggregated data that the threshold levels were set for a male...
	were set at a male level and they were set too high for women, which meant basically that
	women had died. Basically women had died because the threshold levels were too high, until they did this meta analysis.
	of previous papers and were able to reset that threshold level. So on that note, I'm going to close and little bit over time. I hope you'll forgive me. Quite a bit over time, actually, I see now, I'm very sorry about that. I'm going to stop sharing my screen if I can find it.
Anne Milley, JMP	Thank you so much, Caroline. It was just fascinating.
	No, no, it's fine, delighted to hear you give such great examples and food for thought. We do have a number of questions that have come in from the audience, so thank you. And if you have more questions, please submit them through the Zoom Q&A
	window. But let's just jump in. Have you considered contacting the National Institute for Occupational Safety and Health for information on how face fit testing is being performed on humans? How different face dimensions are considered in the design process?
Caroline Criado Perez	Well, I actually haven't contacted them. No, actually I did contact them right in the beginning when I was trying to find the data and and
	then I was, I was looking specifically at the data. But yes, I do need to go back, go back to them and one of the reasons I looked at it...well, one of the reasons I looked at it was that I was very angry about the situation
	with healthcare workers and other frontline workers been needlessly put at risk by this sort of one-size-fits-all attitude.
	But also, I'm actually working on my next book at the moment, where I'm sort of wanting to look more at sort of solutions than than problems. And so, I haven't contacted them.
	Maybe you're asking me, because you know someone there. If you are, please do put me in touch. I would be very, very grateful for any contacts, because obviously these huge bureaucratic institutions, it's quite hard to get hold of a good email address or a good phone number.
Anne Milley, JMP	we know you specialize in women's data being left out of decisions. Are you also aware of big data gaps for other groups of people?
Caroline Criado Perez	Um, yeah, it is. It is basically a problem in all sorts of in all sorts of areas. There is absolutely, for example, an ethnicity data gap,
	which we are seeing more and more when it comes to AI. So, you know, classic examples like facial recognition software being less able to recognize women's faces and black people's faces. And so, in particular, black women's faces.
	I was interviewed actually by a BBC journalist Kim ???, who told me that she had problems
	getting those electric...sorry, not electric...those automatic taps, you know, you get like an airports and stuff, they just don't
	recognize her hands. You know, because they're darker hands and they just don't see them. So absolutely, there are there are data gaps around there. It's difficult to find good
	stats about the extent of them, and that is because we don't disaggregate our data. So when I was writing Invisible Women, you know, I really wanted to be able to talk about the differences, for example, between white women, able-bodied women, women of color, you know, but
	it was basically really difficult because way you had any kind of disaggregation,
	it was usually, you know, you would have maybe sex disaggregation, you would have maybe ethnicity disaggregation, but you'd never have them together. So if you're looking at, for example, trying to find out how many
	black minority ethnic female professors there are in the UK,
	that data when I was trying to write the book was not available because you could find out how many BME professors there are, how many female professors there are, but not the two together. So that in itself is a really frustrating issue.
Anne Milley, JMP	Yes, indeed. Um, so many questions coming in. So I want to get to as many as we can. Caroline has listed the gender data gap and has shown a lot of examples. Is she aware of conscious efforts by these big companies like Apple, Google, etc., where there they are adding this into their AI efforts?
Caroline Criado Perez	And I, well, I'm not aware but that doesn't mean it's not going on. I have actually spoken to Google and and again with these big companies, it's quite difficult to know
	who exactly is going to give you a straight answer. You get sent to the PR department,
	and obviously, they tell you that they are, but I don't know to what extent it is actually happening. Apple. I've never managed to speak to. That said, they have released
	or announced today they're releasing a smaller smartphone that has you know as high specs as as the other one that they're releasing, which is
	quite exciting. I don't know if it's because I complained about huge smartphones in my book, but I'm certainly taking it as a win for me.
	So yeah, it's really difficult to say because it's difficult to get a straight answer. You know, as I said, I've been trying to find out about whether
	there's sex disaggregated data for masks. And, you know, manufacturers aren't telling me and regulators even, which really surprised me, arejust ignoring me.
	So it's difficult. It's really difficult. You, you only really find out when you see the effects of a product and and that of course is, is one of the big problems when it comes to AI. I mean, I'm a strong proponent of regulation when it comes to
	to AI because there just isn't any sense that tech companies really understand the social impact of what they're doing. You know, the fact that we are literally currently coding inequality into the future and and and also that
	I've lost my train of thought. I'm sorry.
Anne Milley, JMP	That's ok. We have so many. If it comes
	back, you can jump on it.
	Okay, next one is in the last few years, the idea that gender is a social construct is growing, and descriptors such as non binary, gender fluidity etc, are becoming more accepted. Does this movement threaten your work? Some scientists,
	e.g., Deborah ??, are noting that any work describing gender differences are not being funded.
Caroline Criado Perez	Well, I think that you need to distinguish between sex and gender. So sex being your biology and gender being. I think most people would agree...well, certainly I would advocate, is a social construct. I don't think you're born, you know,
	being subservient if you're a woman, right, or being, you know, liking pink. That's, that's kind of what you mean by gender. And when you're talking about sex, it's talking about biological differences that are, you know,
	absolutely clear in all sorts of medical and research, but also, you know, just by sort of looking at humans, right. There are some humans who have boobs. And there are some humans who don't. I would say when it comes to
	to trans people in medical research in particular, I think that is a huge gap but that sort of relates back to the question someone asked
	earlier about other other gaps.That's a really concerning gap, I think, because there is very little research on, for example, you know, what
	what the right treatment is for someone who is on sex hormones. How those hormones might interact differently with the medication. Are they treating someone
	in the right way? Are they treating them according to their gender? Are they treating them according to their sex? Are they taking into account all these differences? Because one thing we know is that cell sex
	does affect the progression of diseases and all sorts of treatments and but also so do hormones. And if we're not tracking those...
	well, essentially I think that there's probably going to be a medical malpractice suit down the road unless someone and well someone...people...everyone the medical community really gets the grip on it. It's really concerning. There was this...
	there's been a few things actually and recently about
	kidney disease, in particular for trans men, and how to treat kidney disease, basically because there have been
	trans men who have not received the right treatment and have actually almost died
	because there wasn't this factoring in of them being trans and how these different treatments might interact with their bodies and what the different thresholds maybe should be.
	So I think it's more of a concern from that perspective and the perspective of, you know, we can't sex disaggregate our data.
Anne Milley, JMP	I think that's a really important point for our audience is that there are variation concerns in so many dimensions and these things are not being measured,
	the variation in that data and what that means for questions we're trying to answer and solutions we're trying to find. So thank you for that commentary.
	So we have so many more questions, let's see if we can get to a few more. How should data be disaggregated for someone who identifies as a different sex.
Caroline Criado Perez	How should data be...well, I think that it's very important to record sex at birth, and then also
	gender identity. I mean, you should have both always and because it's, you know, you need to know both. I think it's very simple. You need to know both, particularly in a medical sphere and but also in in other areas to be able to treat people properly.
Anne Milley, JMP	Yes. Good.
	There's one here. Are there any car manufacturers that take into account the biomass of women's bodies in their engineering?
Caroline Criado Perez	Well, that is a very interesting question. And so, Volvo certainly claims that they do. However, they would not provide me with any data to back this up.
	So yeah, as far as I'm aware, no, and I think it would be quite difficult for them to do that because you the average female car crash test dummy does not exist. And although
	car manufacturers are moving more towards computer modeling, actually, when you look at those computer models, they tend to still be based around the
	the dummies rather than human anthropometric measurements, because the dummies are what are used in the tests and so they need to design the cars to pass the tests.
	So it's really a regulation issue, actually. Again, it needs to come from the regulators. They need to recognize that female bodies exists, that female bodies are going to be in cars and driving cars
	so that the manufacturers can design for those bodies. Because at the moment, if they're going to pass the test, they are going to have to focus on these very specific body types that are represented in the dummies.
Anne Milley, JMP	Okay, very good. So there's been recent changes in the NIH (National Institutes of Health) and animal research requiring cage co housing and use of male and female subjects.
	This has been a part of grant reviewer comments for the past two years now. Do you have data that may support this trend?
Caroline Criado Perez	That I'm not aware of any data sharing what the difference has been. I do know that when the NIH first instituted including female subjects
	in human trials, the General Accounting Office did a review a number of years later and actually heavily criticized the NIH for not enforcing the regulations. And certainly
	some experts I've spoken to complain that the same thing as being done now, that these regulations are there, but they're not specific enough so people can fudge it and also that they're not being followed up properly, which is really frustrating.
Anne Milley, JMP	Hmm.
	Wow, any tips for ensuring we think carefully about overcoming this gender bias when designing studies, experiments and products?
Caroline Criado Perez	Sorry, could you repeat the question?
Anne Milley, JMP	Do you have any tips for ensuring we think carefully about overcoming this gender bias when designing studies, experiments and products?
Caroline Criado Perez	Yes, I do have tips. My tip is include women.
	And there is a very good evidence to show that having
	a diverse team means you are more likely to do gender analysis, to sex disaggregate your data. This is, this was an analysis...a
	review of studies going back several years. I can't remember off the top my head exactly how many years, but you know it's a pretty robust study finding that there was this strong correlation between having women in your team and remembering to include women in your in your research.
	And then there's also the issue of, you know, going back to that Apple health tracker app, you know that this isn't a sort of malicious issue. This isn't
	men deliberately excluding women. It is a case of not necessarily being aware of all the issues you might want to look into. And so I...there was this really interesting
	Kathy O'Neill article from quite a few years ago where she interviewed this...
	it was a recruitment AI firm and they develop this AI that was meant to look at for...
	it was recruitment for coders. Rather than just looking at your CV and your job history, they also looked at your social data. They looked on the internet for you, basically. The person that she interviewed said that one of the things that they found was
	that people who visited this particular Japanese manga site were...tended to be better coders.
	And that was presented as this, oh, isn't that interesting. Such a fun little correlation. This is something you wouldn't find in a CV.
	But as Kathy O'Neill pointed out, you know, manga sites are very heavily male dominated, it's much more likely that you're going to be a man if you're visiting a site like that. Well, why is that? For a number of reasons. One, they're not necessarily that
	welcoming to women, sites like that. But also women have less time. Women in the US have five few hours of leisure time than men.
	So unless you have an awareness of gender, right, of the disproportionate shouldering by women of the unpaid care work. The fact that women have longer working hours when you factor in the paid and unpaid care work together. The fact of, you know, misogyny on the internet and how that might
	prevent women from accessing certain sites, you're going to end up designing something
	that might end up completely unwittingly discriminating against women, when what you're intending to do, you know...this was meant to be a gender-neutral product. They wanted to find
	the best coders, but actually what they did, because they didn't include gender in analysis was potentially find the best male coders.
	And I just don't think that would have happened, you know, had there been more women on the team to be able to point out those kinds of gaps in perspective. Another example is things very sort of useful in this
	in this area, it's not about research. But it's that an anecdote that Sheryl Sandberg wrote about in her book,
	Lean In, where she was pregnant when she was working at Google.
	And she was in a senior enough position to walk into the head of Google and say you need to put in pregnancy parking. You know, this walking across the car park is killing me.
	And and his response, I thought, was fascinating, which was, yeah, of course. It had never occurred to me. And I just thought, well, of course not. Why,
	if you've not been pregnant, you don't. I wouldn't occur to me either, I've been pregnant.
	But it needed a woman who was in a senior enough position to be able to make that call,
	because I'm sure there had been pregnant women at Google before, right? I'm sure, there'd been pregnant cleaners and pregnant account managers, whatever it is that they have there, but they couldn't
	go and request it. So that's just a way of saying that, actually, it's not just having a certain number of women. It's also making sure you have them in the right level so that they can point out these gaps and they can make these changes.
Anne Milley, JMP	Right, very good advice. And the other thing you can do to think more carefully is read your next book when it comes out.
Caroline Criado Perez	Absolutely. You're a much better marketer than I am.
Anne Milley, JMP	Well, I also wanted to mention that another past Discovery Summit Europe plenary speaker David Hand also cites your book is excellent examples of dark data, which is the name of his latest book, so maybe you guys can collaborate on another one.
Caroline Criado Perez	Well, maybe.
Anne Milley, JMP	A lot of dark data out there.
Caroline Criado Perez	Yeah, well, maybe, maybe you could put us in touch.
Anne Milley, JMP	Okay.
	Thanks. Thank you so much for sharing all your fascinating research and we look forward to your next book,
	but in the meantime, I recommend to the audience please pick up her book if you haven't. We have some that I think we can give away as well. And thank you for joining us for this final...for the plenary of the day.
Caroline Criado Perez	Thank you very much for having me.
	Thank you.
Caroline Criado Perez	Bye bye.

KeynoteRe: The Gender Data Gap

This was a fantastic talk, and thanks again to the JMP team for sending out copies of her book ... just started reading last night!