The lost Marie Curies
01:20:41 Share

Welcome to the LSE Events Podcast by the London School of Economics and Political Science. Get ready to hear from some of the most influential international figures in the social sciences. Good evening. It's my great pleasure to welcome you all to LSE for this hybrid event. My name is Nava Ashrafa and I'm Professor of Economics at the LSE and Director of the Altruistic Capital Lab. This event that we are in today is part of our inaugural lecture series at the LSE.

which honors the professors from across the school. These events are a special moment in an academic's career, as it marks that they've now been promoted to professor or have recently joined the school. This is a way in which we can all celebrate the success of our community, and I can't imagine a better person to bring us into this 2025 inaugural lecture series than Xavier Giravel.

Xavier Gerval is a professor of economics at the LSE. His research explores innovation, inflation and inequality with related methodological work in applied econometrics. He serves as a member of the French Council of Economic Analysis, an independent advisory board to the French Prime Minister.

His contributions have been widely recognized, earning him the 2019 Philip Leverhulme Prize and the 2021 Prize for the Best French Economist Under 40. He's also a co-editor of the American Economic Journal Applied Economics. And according to REPAC, he is the most cited economist whose first publication appeared in the last 10 years.

So you can see how excited we are to have the inaugural series with him. Xavier will be talking tonight about how innovation is increasingly monopolized by a small entrepreneurial elite that is not representative of the population at all. It's urgent to involve everyone, especially women and people of underprivileged backgrounds in the innovation process, from the creation of technologies to their widespread dissemination.

What do we know and what should we do to find the lost Marie Curie's and lost Einstein's and give them their chance? For those Twitter users or exes who still want to tweet about this, the hashtag for today's event is #LSEevents. The event is being recorded and hopefully will be made available as a podcast subject to no technical difficulties.

As usual, there's going to be a chance to put your questions to Xavier. For our online audience, of which there are many, you can submit your questions via the Q&A feature at the top left of your screen. The questions are going to be submitted to myself right there. And please let us know your name and affiliation. We're particularly keen to hear from our students and alumni, so please let us know. For those of you in the auditorium, I will let you know when we'll open the floor to questions.

If you can raise your hand and wait for the microphone, I will ask to provide your name and affiliation before posting your question. So without further ado, I'm delighted to hand it over to Zebi Jarabel. Professor Jarabel. Thanks, Othnaba, for a very kind introduction. Thanks to all of you for being here. Great to see many familiar faces. So this is...

a talk about the lost Nercuries, which Navar already introduced very clearly. And I'm excited to share this with you today, I think, for two reasons. Number one is that I think it's a topic that could be eventually of great policy relevance. It's not there yet. And I think part of it is that the research hasn't disseminated as much in the policy circles as it could have, I think should have, given its potential importance. So I'll emphasize some of that.

And number two, since there are many economists in the audience tonight, even though it's a general audience talk, I think it's interesting because it's a topic where we can

combine different methodological approaches, more macro, more micro. So I'll show you some evidence from RCTs, some results from macro models, which together help us paint a picture of this topic of lost talents. That will be the theme for tonight, finding the lost meritorious is very important both to increase growth and reduce inequalities. So without further ado, let me be a bit more precise about the motivation for tonight's talk.

So the first motivation is that we often think of the dark sides of innovation, which is viewed as a driver of increasing inequality, including income inequality, wealth inequality, which then could also confer political power to some actors. At the same time, we know that we need innovation for many things, including fueling material prosperity,

In advanced economies, we need growth for sustainability of public debt, for sustainability of public services, and more broadly, we need innovation if we want to address challenges like aging or the green transition. So is it true that there's a collateral damage of inequality that is a necessary consequence of innovation? We need innovation to meet those challenges.

big question where I want to say no, but in fact we have many policies and the main one I want to talk about today is the finding the lost Merit Curie's that can help us simultaneously increase innovation and reduce inequality. And so I want to basically draw a list of what we know, what exactly is the evidence and in the spirit of this academic institution I would like you to be skeptical about

the evidence and then we can engage in the Q&A and see what it might take to convince you and what it might take to convince policymakers to make this an important part of the innovation policy toolkit, which today it isn't. So first, motivation, inequality, innovation, breaking that trade-off, collateral damage and saying, in fact, we can have the best of both worlds. Motivation two is putting inequality aside and just thinking about

the potential impact on productivity of finding these lost mercuries. And here my argument will be that there is so much lost talent that doing this well could be more important than generative AI or other recent technological revolutions. So to address these points, I'll first present a series of stylized facts to give you a sense about how much lost talent there might be.

Number two, taking more macro and modeling approach without going into detail, but showing you some results about what is at stake in terms of economic growth. And then number three, thinking about some specific policy levers that seem to have great potential with evidence from RCTs.

So I think these three steps are what the recent literature, some of my work, but also many other people's work has been finding. So one, that there is a large pool of untaxed talent, people who would be interested in joining broadly this innovation sector, which includes science, entrepreneurship, invention, innovation, adoption, and diffusion of innovation.

That the effects can be very large for the macroeconomy and for inequality, and that we know by taking the status facts, putting them into models, and then comparing that to some empirical evidence that align well in terms of orders of magnitude. And then you'll see that effects can be very large.

And then number three, you might say, well, this might be very hard to change. But in fact, we have some results from program evaluation where it seems we can change things quite drastically, even in the short run. So I'll show you some evidence highlighting the importance of role models. And there might be many other tools. But it turns out that for certain tools, we have great evidence and things that we could be inspired by for large-scale policy reforms.

So the bottom line will be that you have this large potential by finding the lost meritorious to increase growth rates and reducing inequalities of opportunities and outcomes. It will be about gender, but also family background and where people grow up. And I'll focus on just a few papers, evidence from just a few countries, but most of what I show you has been shown in also other contexts, and that's also something we can discuss in the Q&A.

finish the motivation here by saying that there are several policies that have already been implemented trying to find the lost maritories but not in a very ambitious way, not with a great scope and so the message is to say that we should do this with much more ambition given what might be at stake both for growth and inequality. So let me start with a series of stylized facts. So I'll first use

a measure of innovation that uses patents in the United States. The results are similar if you look at scientists or entrepreneurs as alternative measures of innovation. So the graph here shows the relationship between the parent income and the fraction of children who become patent inventors at some point in their lives, so they obtain a patent.

And so you see that the rates are very different. If you grew up in a lower income family, say below the median of the income distribution, you have a rate of 0.84 children per 1,000 children who will one day get a patent. That rate is about 10 times larger if your family is from the top 1% of the income distribution. And this turns out to remain true when you start conditioning on measures of academic achievement.

So one graph is here where we look at early grades math test scores. This is the same sample as before. And here I'm going to show you two lines showing the relationship between these math test scores and then the fraction of people who become inventors. So you see that if you are from the lower income part of the distribution, you have a peer income below the 80th percentile before this became very convex.

Well, if you are very good at math, you have two standard deviations above the mean to the right of this graph. You do have an increase in the probability to become an inventor, but it's not very high compared to what you see if you have also high parent income, where you have this much steeper increase. And we see that with many other sociodemographic characteristics besides parent income.

One of them is gender, and this graph here puts this in a historical perspective. So the graph here looks at patent inventors by year of birth and the fraction of women among the patent inventors.

So parity would be 50% and see we're far from this. So inventors were born in the 1940s, we have about 8% of those who are women. For those from more recent codes born in 1980, they're close to 20%. So it's a slight slow increase in patenting rates for women. And at this rate, it would still take many decades to reach parity. So this changes, but it changes extremely slowly.

And there are other dimensions of heterogeneity that I won't show here, but for example in the US, race is going to be an important measure of heterogeneity with under-representation of racial minorities in addition to gender and parent income. The dimension I want to emphasize a bit more is geography.

Again, here sticking to the US, sticking to patent data, but you see very similar things if you look at the UK. I've also looked at this in France. Others have looked at this in Finland with measures like entrepreneurship instead of patents. So the graph here is representing where inventors grow up. So it is well known that innovation happens in particular places like Silicon Valley or around Boston. You have a lot of biotech companies. It is less well known that actually

Inventors grow up in those very same places. So if you grow up in the dark places, you have over three children per 1,000. When they get a patent, the light places, it's below 0.4 children per 1,000 children. So places like California or in Boston or in Chicago are much darker.

And in this work with the co-authors, we show that a lot of this heterogeneity reflects a causal effect of places. So we arrive at this conclusion by looking at people who move at different points of the life of the child. And so if you basically move when your child is five, the child is going to converge to the probability of the local population to become inventors, in fact, also become inventors in particular

topics, so say biotech or modulators or semi-modulators or oscillators. At a very granular level, people end up doing things that are likely in their immediate surroundings. So it seems to be these causal effects of environments.

which later I'll also show you more evidence suggesting that this might have to do with role models, exposure effects, what people have on their mind, what they view as a natural horizon that they will imagine themselves to pursue later in their career. And so what this means is that the innovators of today nurture the innovators of tomorrow, but then you have also huge inequalities across places.

And importantly, this heterogeneity is still true if you just look at people who live in the same place today. This is not just a mechanical thing where if you grow up in California, you're going to have to live then later in California and you have more innovation opportunities. It's also true if you look at two adults living in Boston today, but one grew up as a child in Boston, another grew up in Alabama, and then the second one is going to be much less likely to engage in innovation later. That's the other dimension of--

heterogeneity and potentially lost talent that I wanted to highlight. That we lose talent in terms of gender, territorial, childhood environments where people grew up, and parental income and race.

But, Geography of the Atrogen 80, here is a graph that highlights that actually the Atrogen 80 is really, really local. So it's not just California versus Alabama. Even within California you can have vast differences. And so here, it's a figure that ranks the commuting zones of the United States, looking at the top 10 in terms of the fraction of children who become inventors to the left. And then to the right you have the bottom 10.

And so you see that you have some places in California, like San Francisco, San Jose on the left. But you also have places in California on the right, Modesto, Fresno, which are just three hours away from San Francisco. So it really seems to be something quite local. And again, that speaks to this idea of people being inspired by the immediate environment that they grow up in and the type of people that they interact with. And then I'll show you later some of this with policy interventions, bringing role models to people.

And then the last stylized fact I wanted to introduce is this one, which is trying to assess the types of people that we might be missing out on. So one tradition in economics is to use

models where agents are rational. So in this case, some agents might face barriers like discrimination or other forms of barriers that make it more difficult for them to enter certain professions like innovation. But then if someone is

really has really high ability in these fields, by the logic of these models they should always go into these fields, Marie Curie, whatever the barriers, by the logic of these models, would always go into science, would always get a Nobel Prize because no matter what she would do it. And so according to these models then you should see that people who do make it to the innovation sector

should be on average better than the others because they're positively selected. In fact, in the data we don't see this at all. If you look at parent income here or ethnic minorities or by gender,

People from the minority groups, or either racial minority, or lower parent income, or female inventors, if anything tend to have lower outcomes than in the majority group, which is either that there is strong discrimination on the job, and then that reduces their ability to have an impact, so then Marie Curie is not able to thrive for full potential.

Or it might be also that we miss out on some of the best and brightest who could have enjoyed going into these professions but who did not because they were not doing rational calculations, they were influenced by many other types of factors, for example sociological factors, and they were just not imagining going into these careers. And so I can show you a bit later some other evidence that really corroborates that the effects of role models might also

be important even for the very best people in the highest ability in certain professions. And so that's why we might be missing out even on the best and brightest, hence the term loss mercury. So I mentioned earlier that I wanted also to use this evening to put the spotlight on different types of methods that we use in economics. So this so far was just a range of statistics, really descriptive statistics.

giving you a sense that there is a lot of lost talent, which was just stylized facts. So now I want to change gears a little bit and without going into detail, show you the results from a macroeconomic model. So we might think we are losing a lot of talent, say by gender, because we have under 20% female inventor instead of 50%. So if we were able to relax these barriers, we would really increase the size of the inventor pool.

In fact, if you combine gender and parent income, you see that we could have four times as many inventors in the US than we currently have. If people from these underrepresented groups went into these innovation careers at the same rate. But now you might think for the macro economy, there will be lots of complicated displacement effects. If we bring in more women, they might be replacing some of the male inventors. So how do these things play out? And what is the overall impact on growth?

So we have models that precisely are aimed at capturing this type of crowding out effect. So let me show you the result of the model and then give you some idea as to why I think the magnitude is actually plausible. So this is the result of a model where we calibrate the model to the US economy, where the baseline annual productivity growth rate is 2%.

And then the counterfactuals to the right are what happens if we relax the bears. So these are pretty ambitious counterfactuals because we're going to give full access to the population of women. So we go from today, in the full population, you have about 12% female inventors. And now you would go to 50%. So it's a long way away from where we are. And that does give you drastic effects on growth, where you go from a 2% annual productivity growth

to 3.4%, the first bar there. The second bar is just focusing on the very best. For example, if you focus on the top 1%, say, in mathematics, so those who have the latent ability that's highest for innovation, if you just reduce the barriers for these people, you would also already get a lot of the growth impact because a lot of the innovation impact comes from a small fraction of the population. So you see this.

really important impact of an increase in the growth rate from 2% to 3.4%. So 1.44 percentage points higher growth every year. That's a fraction of the baseline growth rate, so 72% increase in the growth rate of productivity. So is that too large to be, it's a model, so how can we assess how plausible this might look? So it's interesting to, I think, draw a comparison to this

big shock that was studied in the 1930s. So conceptually here, I'm saying we're going to bring into the innovation sector a lot more inventors. The women before didn't have access to these careers because they faced these barriers. And so we don't have examples of a country achieving

a policy like this, bringing an underrepresented group. But we do have examples of large immigration shocks and large entry of inventors in certain areas of innovation. We can use that to assess the plausibility of the results from the model. And so one that has been studied by others is

In the 1930s, you had a wave of immigration from Nazi Germany, so you had Jewish scientists who were fleeing Germany, arrived in the US, and especially in chemistry, a really high profile German chemist who then arrived in the US, and they transformed their fields and they attracted a lot of talents in their fields. And so you have an influx

of about an increase in the number of inventors in these fields where the German chemists arrived, about a 61% increase in the number of inventors in these specific fields. And that led to a 71% increase in the patents in these fields. So you had an increase in the supply of inventors that also led to an increase in the overall production of patents. It's not like you had a game of musical chair where the new inventors just replaced the existing inventors.

And in fact, this aligns well with the numbers of the model where in our model, this kind of factual that we study where you have equal access for women, it leads to a 65% increase in the number of inventors and a 72% increase in this growth rate of productivity. And so I also want to mention that these numbers are quite large compared to--

what's been discussed for the generative AI revolution that's undergoing. So there's an active debate in economics with some people like Doron Acemoglu claiming that there's little to

expect from Journey of AI. The effect on growth might be 0.07 percentage point a year. And then others like Philippe Aguillon and Simon Brunel who have effects that are 10 times as big, 0.7 percentage point a year. That's still half as large as my 1.44 percentage point a year. So I think in the public, most people would be willing to buy that there might be uncertainty, but Journey of AI might be very important.

And that level of centrality of the journey of AI in innovation policy leads to very ambitious policies. So literally hundreds of billions of dollars that have been committed to changing the supply chain of Gen AI in various countries, including the US and now more recently Europe.

And there is not that level of public perception for this simpler idea of democratizing access to innovation careers. And so the first part of the argument today is actually we have lots of status facts saying there's vast under-representation. And then we have models that are very much of the style of what we use to assess other types of impacts of technological change where we say, yes, there could be a huge impact. So then the other question I'd like to--

to raise in front of you is also the types of innovation. So, so far we've talked about the rates of innovation by mobilizing the entire pool of talent, we could increase the overall rates of innovation, but maybe we could also change the types of innovations we see and the types of products, maybe also the types of people more broadly that get involved in startups as employees. So, let me talk about this in steps and the idea is that if you open

these gates to the innovation sector, you might bring in people who have different visions of what technology should be used for, what societal goals innovation should be trying to serve. And so there are historical examples of this, of people trying to solve issues that they've encountered in their own lives.

So let me first mention a French and somewhat tragic example, which is Louis Braille. So you probably know the Braille writing and reading system for the blind. And so Louis Braille actually became blind at the age of three. He had an accident in his father's workshop. And then later on, he went to a specialized school. And then he found a system that someone was trying to develop that he developed further himself. And then that became a world standard, the Braille.

A less tragic example, but telling at different levels, is the example of Josephine Cochrane, who is an American inventor, who is actually widely recognized as an extremely influential inventor. In fact, she is in the US Inventor Hall of Fame. That's close to the US panel office. And she was a rich aristocrat in the 19th century. And she was managing a team of maids and people,

doing things in the house and these people were not always very good. And so in particular they were not being very good with refined china. They kept breaking things and she felt this just could not be tolerated and so we should just move this to a system with high reliability and invent a machine to take care of this.

didn't have any technical knowledge herself, but she knew she could probably put together a team. In fact, she was the granddaughter of someone who had been quite influential in the first industrial revolution that obtained some of the first patterns for the steam engine. And so she had this broad idea that this could be done. So she brought together a team of engineers, and then they came up with a machine that they called the dishwasher, which indeed could handle the fine china with no problem.

Tragic example, less tragic example, in both cases problems that people encountered first-hands. And these are just historical anecdotes. Is this true more broadly? So in more recent work we've looked at various sectors of the economy and looked at systematically the relationship between the types of products that people create and the customers who buy these products.

and the gender, the parent income, even the age of the inventor. And we find systematically this homophily, meaning that people tend to invent things that are relevant for customers like themselves. Just one example here is for phone applications in the US. So this has the advantage of

being something where you can record individual usage. And so here you have two histograms that are overlaid. The blue is when you have a female entrepreneur, and then the white is just male entrepreneurs. And then the x-axis is the fraction of female usage. So the blue is an histogram that's shifted to the right compared to the white, meaning that you have more female customers.

And one number is you have 30% more usage from female when you are yourself a female entrepreneur. And so this then can be seen in many other sectors. One important sector is health, where the differences are even starker. And so then this can be important for consumers in a broad sense. So here if I go back to my model, which earlier was used to make statements about growth,

This is used to think about inequality in purchasing power. And so here I'm just going to state the results and give you a comparison point. The comparison point is to the left. Here I'm reporting the gender gap for the median worker in the US economy. So the median female worker is paid 20% less than the median male worker.

But then if you actually look at things like product variety and the diversity of products that's offered, and then through the lens of a model, you can see how much this is worth in terms of effective purchasing power. And so this actually is shown in the first line here, and it's 18%. So it means that you have effectively reduced purchasing power through reduced product variety, and that's actually worth a lot to these customers. And that's something that you would get rid of if you were to

break these barriers to innovation for women because you would equalize the number of female and male investors. This was for gender. You have similar things for income. So the graph here relates parent income to the income of the customer on the y-axis. And so if you're from a rich background, you tend to work more often in industries like

financial services, introduce new products there. If you're from a low-income background, on average, you're more likely to enter necessity industries like, say, food processing, food manufacturing. And so finally, we can also think about the collaborators of these inventors. And we see there, again, a form of strong homophily.

So the graph here shows that there's a relationship between the parent income of the entrepreneur and the parent income of the employees of that entrepreneur. So people tend to work with people like themselves also in terms of social class. So this is actually evidence from Finland. As I mentioned, we have this type of results in a variety of countries.

So together, the evidence here suggests that you have an effect on the rate of innovation, but also on the types of people that are going to benefit from innovation downstream, customers, and also employees who join these startups. So if we agree on these facts and that there's potentially a lot at stake, then we have to think about

what could be done. And so here I want to be relatively brief, but give you one example of something that I think could be done at a large scale and currently isn't done at a large scale. And mention that there's probably many more things that we could do, but currently we know less about that from an academic perspective. And so I think it would be great, in fact, to encourage more work on this given the stakes. So what I'll emphasize is role models for career choice. So the idea that some people can be inspired by

different careers just by interacting with people in particular interacting with people that they can identify with so it's not just about information it's also about someone embodying a career

And then there could be many other things. You could think about dedicated funding scheme to help people, perhaps from low income backgrounds. You could think about various policies to try to prevent discrimination. And this also relates to a broad literature in public economics. What I'll do is emphasize the first one because we have great evidence from RCTs. And the effects seem to be massive. So this is the graph from a paper by colleagues that came out last year.

So this is an experiment that was conducted in France. And so they go to a number of schools, and they look at students in the last year of high school, at the time when they have to choose where to go to university. And they're going to look at enrollment in selective STEM fields, where then you're disproportionately likely to then contribute to innovation in various ways by creating technologies or adopting technologies. And so there are these stark differences by gender.

And they're going to try to see if a simple intervention where someone goes to a school, presents their career as a scientist at a large firm. So look at L'Oreal, so it's like a firm that has mostly beauty products, but that has very different types of jobs, and some data analytics jobs. And so they have different people doing various types of performing scientific tasks for the organization. And that will present their careers and their curriculum.

So let's first look at the baseline. So the control group is classes that were not visited. And then the classes that are visited by the scientists are going to be picked at random. So at baseline, you see that if you rank the students by how good they are at math, at the top 25%, women, you have about 24% women who go to selective STEM fields. For men, it's much lower. You have 45% or so. So there's this huge gap in who goes to selective STEM.

But then you have a huge treatment effect, especially for women. So this is the effect after the classes have been visited. And this is a very short intervention. It's about two hours where you have this presentation of the career and the role of education, what educational path they took. And so for these girls in the last year of high school, they go from 24% going to selective STEM field to now

38%. So they almost catch up with the control group of boys without the intervention. The intervention also has some effect on the boys, but it's much smaller. In fact, it's up to the noise, it's just noise or not. For the boys, is also female? Yes, so great clarification.

So another thing they randomized is the gender of the scientists. And so here, this result is for everyone, and it's mostly female scientists. But they also have some male scientists. And so in the paper, they also have a contrast. And they show that this big effect here for women is entirely driven by the female scientists. And so I think, indeed, here this effect mainly depends more on the male scientists.

homophily effect here as well. You need people who have, that you can identify with here in terms of gender, and that might well be true in terms of race, in terms of parent income, in terms of a personal history. So it's not about just information, but more about the type of people who carry this information and share it with others.

In fact, the last layer of randomization here is that they also randomized the content of what the female scientists would say. So one group was told to be completely upfront about all the challenges they might have faced as women in this environment. And another group was told to not dwell on that.

And so the effect, as you might expect, is bigger when the speaker does not emphasize the difficulties they encountered as well. So there might be this trade-off between transparency and inspiration. So this closes the main sets of stylized facts I wanted to share with you.

i wanted also here to uh share results that we obtained recently with fiddish in some work where we uh we use ai to connect interviews with people at a large scale and so in particular we asked people who ended up in stem fields what they think mattered for them and role models will be mentioned but you'll see it's not the the first thing that they mentioned so we asked people why you chose this particular educational trajectory

And people really emphasize their personal interests and passions. So 70% of people almost said this was another key reason. A key reason was that they were just really passionate about mathematics or science. 33% mentioned that they knew that there was big financial rewards to going into that type of field.

Something maybe we weren't expecting as much is exposure to STEM through hobbies. For example, a lot of men mentioned that they were doing, you know, they were just playing with computers, video games, got interested in coding through that, and that led them into STEM. And then a bit further down below here you have the people mentioned family mentors role model, but 20% of them do mention that, but certainly not the...

the only factor that people think mattered for them. So it's a broad set of factors, of course, that result in these decisions. But we can probably move that quite a lot through light touch intervention about career choice. And you could think of massifying this, bringing this to the large scale in any country. You should be careful about finding the right type of profile to do this well. You need enough female scientists, enough scientists of the right background.

And you could think of many other related things like internships, maybe online interventions could work as well. But in this, we have less causal evidence. But this could be learned at scale, I think, as we deploy policies if we were to say, this is a national priority, let's do this at a large scale. And you could do this at a fraction of the cost of what we do for generative AI and probably at growth impacts that could be just as large in maybe a relatively similar time frame.

Hi, I'm interrupting this event to tell you about another awesome LSE podcast that we think you'd enjoy. LSE IQ asks social scientists and other experts to answer one intelligent question. Like, why do people believe in conspiracy theories? Or, can we afford the super rich? Come check us out. Just search for LSE IQ wherever you get your podcasts. Now, back to the event. So this was sort of my main...

message. And I never do know how much time I have left. What time is it? You have 15 minutes. 15 minutes, great. We can also have more questions. 15 minutes is great. So I want to highlight the last mercury agenda, the status facts, the macro impacts, the policy levers you might want to use.

Then I also wanted to take a step back and say that with this motivation of both increasing innovation rates, reducing inequality, there's a simple idea of just pushing education policies and realizing that this can have a great impact on

on innovation and also inequality because with a more educated workforce you will have faster adoption of technologies, more productivity growth. If you have less inequality in access to education, you will also have less inequality in ability to adopt these innovations and so then less inequality created by the diffusion of technologies. And so I view this in a much broader area as complementary to the last Merit Three's agenda where on the one hand

For the diffusion of innovation, you want to educate the general population sufficiently that people feel empowered to adopt innovations. That's certainly the case with GenAI, which is an innovation that might have the potential to be adopted more rapidly than other types of technologies in past decades.

So that's for the diffusion part, involving everyone in the diffusion. And for the discovery process, that's the focus on really the top students who then go into science, innovation, entrepreneurship, the last Merit Curie's agenda. And I think for a long time, we actually didn't have great

causal evidence of this. I think more recently there has been very interesting work showing this link between education and diffusion of technology with micro data approaches. And so I wanted to show one graph here from who is now at Cornell, where he's studying the massification of

access to universities in France in the 1990s. So engineering schools, new colleges were built across France in the 90s as part of a big push for higher education. And so here the author compares areas that where the new engineering school was created early in the 1990s to other areas where it was created late in the 1990s. And so the zero indicates that data of creation, opening of the engineering school.

And then he's looking here at the creation of new firms. And so you see that there is this increase of 20% increase in the creation of new firms locally, shortly after these engineering schools open, about five to six years after. And labor productivity also goes up. And so that's another way of seeing that these education policies can have a pretty large impact. And we often see in innovation policy that we

tend to really invest in the discovery process, huge investments in certain areas like Gen-AI and much less so in areas like education where in fact for productivity you want the technologies to diffuse, you want the population to be able to be empowered to then adopt these innovations.

And so before concluding, I want to highlight two ideas that are related to what I said earlier because they're about innovation and inequality, but I want to highlight them more as a way to invite questions as they depart a bit from the main focus of tonight's talk.

When you think about innovation and equality, often people talk about the labor market effects of inequality. And so my view is probably we should worry less than we typically do about these labor market impacts. And perhaps we should be most worried

not about being replaced by AI, but being replaced by someone else in another country or in another firm that's adopting AI faster than we do. So I'll show you some evidence for that relatively quickly. And then on the other hand, I said probably we should worry more about who benefits from the gains from innovation from the point of view of consumers for reasons that I mentioned in a couple of slides. So on the labor market effects of trade, there is...

A long tradition of being fearful of innovation creating a lot of unemployment, going back to the Luddites who were breaking the textile machine in 19th century England. And many people have, for example, proposed to tax robots and might think it's better to slow down the pace of technology adoption to make it easier for labor markets to adapt. But you might also have the opposite view. We have economies that remain very open.

so if you tax robots slow down technology adoption you might compete with robots compete with ai coming from elsewhere and that might be worse because if you were to adopt ai you would be more productive you would be able to pass some of these productivity gains to consumers you would be able to expand your market reach expand your sales and then also expand your labor force so you have these two effects the labor displacement effects and then the

market size, scale effects. It's an empirical question of which one dominates. And in recent work, we find clear evidence that market size effect seems to dominate and others have also then looked at this in other contexts. This is a paper looking at France, looking at investment in automation technologies, and we see that firms that invest in automation right after this investment, they expand their total workforce.

This is true for total workforce, it's also true for sales. So that's this idea that because you've invested, you can expand your sales because you can attract more consumers with better product or lower prices.

And something that we found surprising but found consistently is that this increase in employment was widely distributed across types of jobs. So we have an increase in the number of high-skilled jobs, but we also have an increase in the number of low-skilled jobs. And it's not the case that you just hire more engineers. You also hire more lower-skilled groups. You have more churn. So you have some winners, some losers. But the winners are not systematically the highly paid workers, and the losers are not systematically the low-paid workers. So what does it mean that those people work

So you expand your sales so they can help with different parts of the production process. There might also be room for the low-skilled workers to... It's still a part of the production process. And in some cases, some papers were able to show that thanks to these new technologies, this paper here by a team in Finland,

He's looking at a subsidy program, and they show that by adopting new robots and new types of numerical control machines, these firms tend to introduce new products. So they don't just automate the existing supply chains for the existing products. They also expand the types of products that are offered by the firm. And it turns out that the new products, the new supply chain, also involves some lower-skilled workers.

So we see this with different types of technologies, robots, different types of automation. In a more recent paper, we see it with AI, the AI that was introduced in the late 2010s, not yet generative AI. And so we can talk about this more through the questions, but I do think that the main risk might be for us to be replaced by AI that would be adopted faster in other countries or by other firms.

Now to get close to wrapping up, I wanted to highlight a question of who benefits from innovations on the consumer side. So often we don't ask this question because we think about the iPhone or the TV or the airplane which started with a select few and then diffused to everyone, which is sort of the destiny of all the major innovations.

But there are also many different types of innovations that can take substantial investment and remain very limited to a particular market. So here's one extreme example, which is Ubercopter. Ubercopter you can take in JFK. You arrive in New York at the airport. You can take an Ubercopter to go to the center of Manhattan. Initially Uber claimed that this would soon be cheaper than the taxi. The taxi is already very expensive.

It turns out it did not become cheaper than the taxi. This does not diffuse at all. The market remains segmented. And so is this the exception, or is the iPhone the exception? And so in some work I've done that I will not dwell on too much in the interest of time here, I think that for important parts of modern economies, this is an important case because products cater to particular tastes, particular needs.

In most countries you have some economic growth. In many countries you also have rising inequality, for example in the United States. So in general you have faster growth of demand for the more premium segments of the markets.

And the basic logic of innovation is that if you have an expanding market, you will also have more innovation. Financial incentives to enter, pay the fixed cost of innovation, try to out-compete other firms. In fact, other firms may not have yet entered. And so growth of market size leads to more product introduction, more competition. So you have more new products, more competitive pressure on the existing ones, lower prices, lower markups on these existing products. And so that could then

amplify the level of inequality because the initial rise in income inequality then changes the types of goods that get introduced. And if we think about

generative AI say, here the question is who is going to adopt Gen AI faster to make their product more affordable, more interesting, and there as well we might think market size, growth of market size will play a key role. This is something I've studied in certain specific contexts like consumer packaged goods in the U.S. where you do see that you have more new products for

for parts of the park space that cater to richer households. And in those parts of the park space, you also have lower inflation. And these patterns of lower inflation are, in fact, true much more broadly, not just in these consumer packaged goods, but more broadly across the United States.

If you take, try to use exactly the same methodology as the Bureau of Labor Statistics of the US, you find that in the long run, you have lower inflation rates for higher income groups, which I think can be largely tied to these innovation dynamics. And this is meaningful because

If you measure real income inequality with the standard metrics versus adjusting these metrics by differences in inflation rates, you find a much faster increase in inequality in the US, which here is depicted. And just to give you one summary number, this shows by quintiles of the income distribution. And so during the period studied here,

The top income quintile became 15% richer than the bottom income quintile with the basic metrics. If you include these differences in inflation rates, the gap goes from 15% to 22%. So faster increase in inequality. So that was the second topic that I wanted to introduce as a teaser. But let me wrap up with a summary of the takeaways.

So the main takeaway is that increasing equality is not a necessary byproduct of innovation. It depends on the types of innovation policies that we want to put in place. And I think we tend to have, at least in policy circles, narratives around innovation policies that are very much top down.

So either giving the central role to entrepreneurs, call this the market-led view, less affair, trickle down. Or another view which is also top-down, which is state-led, where the government will set targets, will define sectors, and then will start the process.

And society at large is left with a relatively limited role. But in fact, anyone who has studied innovation knows that the innovation process is a much broader process. It's this collective, gradual, iterative process which is really a diffusion process. In fact, I think we also know this as researchers. We know that most of the research we do is of that sort, collective, gradual, iterative. And so there is a role for society at large which should also

take a center stage in the way we design innovation policies. And so one term to capture this idea is the "Lost Mercury's Agenda." And there's something also broader, which is education policies that I've also alluded to. And so this type of approach can have really a large potential, I think, to both expand the innovation frontier and reduce inequality of different sort, gender, intergenerational, geographic. And that should become a central part of innovation policy.

So looking forward to some of your questions. And I leave you with our great motto of knowing the cards of things. Thank you, Xavier. That was fascinating. I'm going to start us off with a question, and then we'll turn to the audience's questions. Also, if those of you online, please send your questions through. So my question is about positive selection, because what we care about is these outliers, right? The Marie Curies and the role model effect. So I was struck by the...

the degree of lower citations of women inventors. And you might think that there would be positive selection into entry into inventors. In other words, those who would have been able to overcome the barriers to entry that are much higher for potentially for women to enter might have been more positively selected. And therefore, you might think that they would have even more inventions.

So either they're not positively selected or the discrimination they face when they're in it is so large that it overcomes that.

I'm wondering which one you think it is, and in particular I think it's related to the role models paper in France, and maybe the way we can partially answer that question is the girls who were motivated in that top decile, the quartile, to get into STEM by role models, did they do better once they got there? How were their grades when they actually entered university? Because if they're much, much higher than we might think,

this is actually getting us through. Otherwise, we might think that it's actually getting the more negatively selected people in by doing role model effects. Great. So yeah, thanks so much. The question is sort of clear to everyone that we use in the graph just to restate. You asked it very clearly, but we see that actually the women who do become invanderers, they tend to do worse in terms of citations.

There are very few of them. We might have thought if only the very best ones come in, actually on average they should be much better. And so as Navas said, there's two potential stories. One is their latent ability is actually a little bit lower than men's, which is some kind of negative selection. And all that could be also discrimination on the job. And probably both are at play. But I think we have at least some evidence that

we don't get huge positive selection as we should in the standard models that we often use in economics. And so one way to see this is looking at these predictors of innovation potential like the math test goals. So I don't think there is studies that look at

or academic results once they've handled, say, a STEM degree, as you suggested. That would be great to see. I don't think we have that. But we can look at their math test scores before. And so you see that here you do draw in some of those that had the best math results. That was the case in the French study. There's another even more spectacular study like this by a Stanford job market candidate this year, Ian Callaway. He was looking at students who go to the math Olympiad,

the equivalent of that, in the US, and looking at the role of then certain mentors that talk to this crowd, and then what people do after. And he finds huge impacts on girls, including those that are some of the very best grades in these extremely selective competition. And so if you view that as a measure of latent ability, which seems reasonable, it means that we are missing on some of the highest potential. It's also true that there probably is a lot of discrimination.

So I think both are true, and the rule model approach is fruitful to bring in some of these really potential superstars into this type of career. Yeah. Okay. Great. So yeah, let's have – questions. Go ahead. Just related to that – thank you. Thank you very much. It was a great talk. Just related to that, one aspect you were saying that it might be, let's say, women

work on topics that are more related to women. And because there's less women, that might imply less citation. So in a sense, maybe citation is not the best measure to look at. That's interesting. Yeah, that's a great point. There is also homophily in citations indeed. And so mechanically, that could reflect that. In fact, in some of the macro model I showed you, another effect we have, which is a bit in that spirit,

So women are more likely to produce products for other women. There is discrimination in the labor market. So you also have a lower market size. So in terms of purchasing power, you have less purchasing power for the female population. So as a result, the female-led firms on average are a little bit less productive just because they face a lower market. So there is also some of that. Yes. Yeah, we'll just move it down the first row, and then we'll go past the first row for the next round of questions.

You gave us a very rosy view of how expanded access would happen in terms of its macro consequences. Let me give a very dark alternative, not because I believe it, but so that I understand how you are able to distinguish the two. Imagine, and I'll give it in two scenarios. One scenario is there's only so many patents to be discovered there, there's only so many inventions. If I have to

fewer, say, men having them and more women having them, they'll be wonderful for inequality. But it will have no impact on aggregate whatsoever because it will just displace the men. And that's very good in an equality sense, but not in an aggregate sense. Whereas you seem to assume that no, no, no, more is just going to be very good.

And likewise, in the same perspective of, imagine that the extra patent or innovation or firm is not a pretty good one. There's diminishing returns to what they are. And therefore, expanding the number of patents or innovations will actually-- the extra patent will become a little bit worse and a little bit worse, to the point where maybe expanding the access actually

There's an opportunity cost, these people could be doing other things, and we're actually pushing them to something else. Now, I don't believe any of these two stories, but it seems important to say why do we think they're not there, as opposed to your rosier view that these extra-padded innovations are going to be as productive as the ones we've seen so far and so on. So can you tell us what kind of evidence allows us to go to the rosier one as opposed to the bleak one? Yeah, great.

Let me just start with the second part of your question about the diminishing returns. I think this relates to the earlier question about are we missing out on the marginal inventors? We're not so good. Are we also missing out on the superstars? So if we think that we miss out on the superstars, which we have some evidence on based on this latent ability, then what you do is that you replace some of the mediocre existing inventors with the superstars that were lost and now brought into the innovation pipeline.

one channel where I think we have some evidence and then you have this broader question of how does the idea innovation productive, how does the idea production function work? Is it mostly replacement of musical chairs or in fact is this recombination people build on each other's ideas and so you can write different models with different results.

And so the two types of evidence we have, one I've alluded to is the paper in the 1930s in Germany, where we have this big supply shock in certain fields of German

chemists and then they then bring a lot more inventors in those fields and you see a ton more innovation in these fields. So, granted it's not macro level, but it's still sector level and there's good causal identification. So I think this is interesting also because it turned out that the magnitudes align well with our model at the macro level. And another thing we have that's more descriptive is just looking at the past 50 years in the US. If you look at the

technology classes where women became relatively more prevalent because overall there has been an increase from about 5% female in there to about 12% which is very heterogeneous across fields. So you can just run simple descriptive regressions of flow of patterns on how the delta of personal representation of women and you have a very strong upward relationship. So that's what I take as the evidence to

and believe in the expanding innovation potential here. Thanks so much for a great talk. Xavier, I'm curious about your thoughts on the rise of corporate innovation.

The view of the world that you've presented fits very well with the idea that growth and innovation comes from loan investors, inventors. But if you look at the United States today, actually a lot of innovation is happening within firms. Now usually we think that that is potentially a worrisome trend also because we see that inventors when they move to large corporations become less productive.

but your results suggest that maybe corporate innovation is a good thing because it might be less subject to the biases that people are subject to if they just innovate in whatever they are personally interested in. So should we take your results as painting quite a benign picture of corporate innovation?

Yeah, interesting. So, I think the current set of my thoughts on this is perhaps limited, but I can tell you what we have on this. So, we looked at founders of firms, which I showed you many of these graphs, right? And showing you that when, for example, women or people from low-income backgrounds start firms, they go after different markets. They also work with different people.

And we've seen that, at least in the time span we have in the data, which is about 15 years, these things persist. So even as the firm grows, they do remain in these different markets. Maybe as if they had set a certain type of corporate culture, and that's then what the company keeps doing, even as the company becomes larger. So that's one type of...

one data point that speaks to your question. Another is we looked at very large firms like Compustat firms, so very large firms in the US, and we see that inventors within these firms, looking at patent data, still exhibit homophily. So within Parker and Gamble,

women are more likely to work on the type of health product and beauty product for women than men for male customers. So even within the companies you might have all these different product managers and they do come up with different ideas and they still have this process of coming up with ideas based on their own personal experience. So we haven't found much evidence of this becoming weaker with the firm size. The same thing also for when you employ different kinds of people, it's still true after 15, 20 years.

But I think that's an interesting area maybe to dig more into. I'm going to turn to online questions and then we'll go back to the audience. This is from Derek Oskal. The loss Murray-Currie suggests that innovation is highly path dependent and constrained by early exposure and opportunity, but in a world increasingly shaped by AI and automation where traditional pathways to innovation might be disrupted, do you see new mechanisms emerging that could democratize access to inventive careers?

Or do these technologies risk reinforcing existing inequalities by automating away the very apprenticeships, tacit knowledge transfers that can drive breakthrough innovation?

Interesting question that I haven't considered before. So in terms of new mechanism, one thing about generative AI is that there's a lot of user innovation. So you can adopt these innovations as an individual. And so it would be interesting to see if that helps perhaps with faster discovery of relevant usage. Another thing I thought a bit more about is more about the point about market size. So when you have really...

general purpose technology, it might be that you break the link between growth of market size and innovation. It's more about the level of market size and innovation. So the usual logic is, if the market is large, you already have a lot of competitors. And so it doesn't really make sense to pursue a larger market. What you pursue is a growing market to outcompete others. And people are going to fight for this expanding market size.

but with a more transformative technology you might have gains in really going after established market segments because you can do things differently and help compete even the established players and so there maybe the middle class would become the main beneficiary not the rich because the middle class remains the largest market even though it's not the fastest growing market so i thought more about

the implications along those lines of the relationship between market size and types of products. And I think it's interesting to also think about the identity of... But then, yeah, I don't see much, but I think it's interesting to see more about. It could also be really related to female entrepreneurs who often don't learn as much from networks and the tacit knowledge that we have, that in fact learn from other things. So this provides access to...

learning your craft. Perhaps, yeah. Okay, more questions from, we'll take that one there. Hi, sorry, I'm almost without my voice, but I still want to ask a question. My name is Karina and I'm pursuing my second master's degree here at LSE.

So this is a very positive agenda as in bringing more women into the innovation. But I'm just wondering what takes place within the fertility rate as in the longer women stay in education in pursuit of their careers, the fertility rate is depressed.

And at the same time, from the policy-making perspective, how can these cohort of women be supported to not only attract them into all these areas but also retain them there? So in terms of child support and the likes. Thanks.

Yeah, you're right that there's some evidence of what people have called the child penalty in a range of occupations and also science and entrepreneurship. So at the moment of childbirth, the men don't decline relative to peers without children, but you have a decline of productivity for women. And we know that this varies a lot across societies. So there are certainly a set of policies and also different types of norms that can help change that.

So I think it's both broad cultural change, and then each institution can also play a role by establishing policies that help people in those times. Other thoughts? I mean, that's probably why Google and Amazon, they subsidize egg freezing for their employees. They do all those kinds of things to sort of change the clock for women. I don't know how effective those are going to be, but precisely for those reasons.

Francesco, and we have another question here, yeah. - Hello, Francesco Caseli from the Economics Department. I wanted to ask you about the focus on innovation in particular. So the birth of minority agreements is many other leadership roles in society, in government, corporate governance.

academia, there are many areas where we see minorities and women underrepresented and we think we may be missing out on very important talent there. My hunch is that a lot of the diagnostics you presented in terms of the importance of family background, geographical clustering,

and the like would also be present in these other areas. So I guess one question is, would you care to speculate if there are specificities or specificities within innovation in these other areas where minorities and women are underrepresented? And as a follow-up to that,

And here I'm going to come out as a low-dive. But is it obvious that the priority for trying to increase representation of minorities and women has to be in innovation? After all, an argument can be made that innovation over the last 20, 25 years has made us much more unhappy, lonely, produced a lot of mental health issues.

has led to a massive deterioration of the political discourse, possibly reduce our freedom, and so on. So this actually leads a little bit to your last discussion about market-driven versus state-driven. I mean, one narrative that I--

think is plausible is that we have transition from state driven to market driven to some extent in the direction of innovation. And state driven, we don't default to the state, has limits to the damage that innovation could make. And now that it's much more market driven, we have all this incredibly, in my opinion,

very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very

Good, I like the countervailing view. I guess it's a bit of maybe a matter of broad take on how we think, where we see most potential. I do think that innovation holds a lot of potential and I mentioned this early on, right? We do need growth for sustainability of public debt, public services, tackling challenges like aging, like the green transition.

And perhaps democratizing access and changing the set of people who engage in this type of activity will also change the types of products that we would see. For example, for women, we do see that women are much more likely to work on the green transition, something I haven't mentioned, but they are three times as likely to work on green transition-related innovations as men. And so if you think about

We have a lot of subsidies to try to reorient innovation away from dirty innovations towards clean innovation. But another way that seems actually very powerful and cost effective is by changing the composition of inventors. So that's one, I think, first set of ideas. At the same time, I agree that representation matters in many other fields, many other activities. And so I don't really see any specificities with innovation.

I don't think there is a case to make other than saying innovation has huge externalities, but political engagement also has externalities. And so I think broad discovery of many fields of activities is really legitimate. Here, the point of view I took was if you want to increase innovation, then this is a great thing you could do. And I don't think it has to come at the expense of representation in other fields. We can do all of this

at once, and in particular for the, as you mentioned, the importance when you have small state-driven innovation. I think also here at the moment, there is very poor representation. The decision makers here are also this small elite from very specific backgrounds. And I think opening that up as well is one interesting way of maybe changing the priorities that we set. Hi. Thanks, David. It's super interesting.

And I have a question, sort of followed up on question on why we observe this for women who made it, but they still didn't do great on the top. And-- is it on? OK.

So yeah, I was thinking maybe one sort of mechanism will be like homophily in terms of teamwork. I guess we know as academics or innovators, they're working in teams. And usually guys like to collaborate with their same gender co-authors. And do we have evidence on that, how they're attributes to this?

pattern you observed? And the second question also relates to that. So you emphasized maybe one of the policy solution is to have a female role model for female students. And have you thought about peer effect?

A lot of times we learn from our peers, not necessarily this one. This person has been super successful, but they're doing great. They're my age, my demographic, and we learn from them. Do we have any evidence on that?

Great. Yes, there is a lot of this homophily in teamwork. So by gender, you can measure gender clustering, and you do see a lot of clustering. So there's an over-representation of all male teens and all female teens, and that doesn't change very much over time. So that might maybe slow down the process of representation, perhaps.

So that's one clear trend in the data. And we also have evidence of peer effects from different types of papers. Gustave Kennedy, who is sitting in the background over there, has very interesting work in France showing these peer effects in contexts that are broader than just innovation. But there are reasons to think that they apply very broadly. And one related idea is that if-- well, a set of related ideas is that if you think people want to join programs where there are enough

people of their own type. So enough women, for example, in chemical engineering, that it makes it much easier to go into that type of engineering. And that's then much harder to break because you have these vicious circles where people don't want to switch to a different field and it's harder to expand in those fields. So I think that those are important additional barriers to keep it. Thanks. Alois, just a minute.

Let's say you're a policymaker looking to capture these high growth, high innovation benefits. In a world which right now, led by the US, is moving away from these sort of pro-DI culture initiatives, how do you implement policies like this which are so cohort specific? Yeah, so the place I would start is these policies about discovering potential careers in high school, but also in middle school, and then also throughout university.

with this type of role model program where you just inform people, but you try to organize things such that you have a broad diversity of entrepreneurs, scientists, innovators that come and talk about their careers. A version of this is mentorship programs where you have a more long-lasting interaction between the students and the role model.

And so to be specific, I think the one way to do this at scale would be you would choose five ways of doing this. It could be these information sessions. I show you the RCT in France about that. Mentorship programs that are longer lasting, something online, something about internships at innovative companies. Let's say we start with these four.

And then I would introduce them at random in different parts of the country and then measure what has the most impact. And then scale up what did work the best. And then think first about maybe high school and then move this to university and then middle school. I think what's interesting is that you can fairly easily measure outcomes by gauging people's level of interest in these careers and seeing actual choices that they make a few years down the line.

So that would be one way of starting and then doing this at scale, you know, to give you an order of magnitude for a country like the UK, you probably need 100 million a year to do this well. To some extent in the UK, Tony Blair had done related things with a set of programs that was called Aim Higher.

in the 2000s, where people were from certain more deprived areas were brought to discover universities. So you could do this at a much broader scale with a focus on innovation, maybe even without a focus on innovation. It would already be quite useful. I think Gabriel--

Hi, I'm Gabriel, I'm a PhD student at the LSE, Xavier's former research assistant. In fact, on one of these projects. My question relates to the income gradient of innovation that you showed us at the very beginning. So a couple of years ago, I was working as a teaching assistant in a development economics course, and the classroom was an online course, and the class was all made up of students who were based in Ghana.

And then we were discussing your paper, the paper of the innovation gradient. And then I showed them, the students, this graph. And then one student un-muted themselves and said, I think that in Ghana, the relationship between income and innovation is exactly the opposite of that. Because kids from rich households, they all want to take safe, cushy government jobs. And innovation comes mostly out of necessity, not out of sort of possibilities. So I wonder if you have any thoughts on the--

or if you have any evidence on like how the sort of like the income gradient plays out in settings that are much more sort of like low-income countries or also just more like more innovation poor. Like is the, if you go to one of those counties in the US where there's very little innovation overall, is the income gradient still true or do other things play out in this role? Do we know much about that?

So I think there's not much evidence on that, but there's one paper by Patrick Gole, who was at University of Bath. And he looked at India, and he specifically looked at patents in India. And there you also have a relationship that's similar to what I showed you for the US. He's using names as a way to infer sociodemographic backgrounds.

which has some limitations which should capture the bulk of the effect. And so they're typically similar to the US. I suspect that if you looked at other things, for example, entrepreneurship that's not necessarily high growth entrepreneurship, maybe that would reverse. And that hasn't been looked at as far as I know. There's some other work that doesn't draw a link to--

Parent income just shows that the same author, Patrick Gold and Roshi Agarwal, showing that people who do very well in math olympiads, when they come from low-income countries, they are much less likely to do PhDs and to continue in fields of science compared to peers with similar test scores or the math results at these olympiads from high-income countries.

I wonder if it has to do with the size of the public sector and the relative value of the public sector job versus private sector jobs. Yes, it would be interesting to check that as well. Okay. If there are no other questions, I have the honor of thanking you for an incredible presentation. Thank you so much for spending your time with us. Thank you.

Thank you for listening. You can subscribe to the LSE Events podcast on your favourite podcast app and help other listeners discover us by leaving a review. Visit lse.ac.uk forward slash events to find out what's on next. We hope you join us at another LSE Events soon.