The Jean-Jacques Laffont Prize recognizes an internationally renowned economist whose research combines both the theoretical and empirical aspects of economics. Last year’s recipient of the award was Susan Athey, the Economics of Technology Professor at the Stanford Graduate School of Business.
Professor Athey’s research spans over a wide range of topics in microeconomic theory, industrial organization and econometric methods. Her current research focuses on online advertising, the economics of the media and machine learning. She regularly advises governments and businesses on questions regarding the digital economy. Professor Athey visited TSE in December to give a lecture on the future of media platforms. The TSEconomist met her for an interview where she spoke on the news industry, women in economics and current trends in economic research.
- During your lecture yesterday, you spoke on the effects of search engines, news aggregators and social media on the dynamics of the news industry. With regards to political news, how much of an impact do you believe that these different news platforms have on actual political outcomes?
I think they have a pretty big impact. For example, in the most recent US presidential election, on many issues we could see an alignment of people’s social networks with their opinions on the issue. If you think about the presidential election between Mitt Romney and Barack Obama, there were many educated people that supported Mitt Romney and many people in cities who supported Mitt Romney. What we saw in this election was that educated people who also lived in cities were overwhelmingly for Hillary Clinton. More than that, Clinton’s supporters were very upset by the policies, opinions and values that were promoted by Trump.
As a result, people’s social media feeds were very one sided, especially for the people who supported Clinton. The fact that social media feeds were very strong in one direction made them more different than what you might have gotten from just reading three different newspapers. Had their social network been more diverse, it would have been more reflective of a full range of people’s opinions.
For example, Facebook has shown graphs about what people’s social networks look like geographically. So people from California have friends from California and also have friends from Boston. In this particular election, people in Boston were a lot like the people in San Francisco. So I think that these things are very important, and I think probably something similar happened in Brexit, where urban educated people have mostly other urban educated people in their social networks, and they were all mainly in favour of keeping Britain in the European Union, but they were not as connected to the people that opposed it.
- And in reverse, how do you think these political outcomes may in return influence the reputation of these platforms?
We are seeing a lot of discussion right now about what Facebook’s role is. The same goes to other online platforms. I think, traditionally, these social networks and search engines have erred on the side of being very open. For example, if you talked to a representative of YouTube, he would say something like this: “Look, if we find someone uploading a YouTube video of a beheading, we are going to take it down, but if someone makes a YouTube video that is just generally racist or sexist, we are not going to take that down because we are an open platform. We want to make sure that if there is an oppressive government, people can put out a video against it, but we are not going to be in the business of making a community that has a set of predefined values.” So I think it is very tricky when you realise that those kinds of “openness” values are actually changing people’s informativeness. The people in charge of these platforms are worried about it and are thinking hard about how to change the situation, but it still would be a pretty big change for them to start imposing values. Some of my former PhD students at Facebook just did a project where they tried to demote what they call “click-bait”: headlines that make people want to click but the headline does not really tell you what is in the story. They released this algorithm a few months ago to try to reduce the ranking of articles that had misleading headlines. However, what I saw in the presidential election is that the problem is not so much completely “fake news”, but that you have one-sided news, you have poor interpretation of facts. That is not truly fake news. And then if a bunch of people want to share that information with each other, it is very hard for a social media website to control that. Traditional newspapers explicitly had a view: “We are going to decide that even though people don’t like to read about Syria, we are going to tell them about Syria. Because that is what we do.” Newspapers historically made the editorial part separate from the business and advertising part. They also maintained this idea of journalistic integrity, where they gave people stuff that they didn’t really want, and then they bundled it together with the things that they did want so that they read it. However, in today‘s world where people can choose article by article and they can see what their friends share, it is very hard to make people read things.
- What do you think should be the role of regulators with regards to news aggregators (e.g. Google News, etc.)?
First of all, I think it is reasonable for news organisations to collectively bargain. Normally, antitrust law would prevent them from doing so. However, when an aggregator or platform aggregates a large set of users, and then on the other side there is a large set of relatively substitutable service providers that must go through the platform to reach the users, those service providers have basically no bargaining power. The platform can control access to the users—this is an example of the competitive bottleneck analysed in the economics literature on two-sided markets. In many such cases, welfare would be improved if the service providers can collectively bargain to reduce access fees or improve terms. For example, if one newspaper goes away, there is no effect at all on Google News because you can replace all of the stories with stories from other newspapers. Because newspapers are in general very substitutable, one single newspaper has basically no bargaining power. The papers that pull out of Google News then lose all their traffic without hurting Google News. So, Google says: “Well, a newspaper can always opt out” – but that is kind of an empty statement. So having them somehow collectively bargain is very important. It is very tricky to think about exactly what regulation is going to fix the problem; the problem being than if ultimately these newspapers don’t get enough advertising revenue they are not going to be able to stay in business, or they are not going to have the incentives to produce quality news. I don’t think there is an easy answer to this question yet. My research showed that there is a valuable service from reducing search cost from these aggregators and intermediaries, so you do not want to lose that valuable service. In any case, this is the trend; people are getting their news through these intermediaries. I think that it is not really realistic to think that this trend is going to change, so it is more of a question of how can we make sure we still get enough (good) news, and that the firms that invest resources in reporting on that news see a return on their investment. There may still be a wider range of policy options, like subsidising some of the investigative reporting, or helping with some of the infrastructure that is required to do some of the investigative reporting. For example, in the US there is a group called ProPublica, and they created databases of government information that made it easier for newspapers to do their research. Ultimately, I think we are going to see some consolidation because when newspapers are more consolidated, they have more bargaining power, and also save on fixed costs.
- We would like to know a bit more about your academic and research career. You spoke yesterday on how Jean Jacques Laffont inspired you to delve into empirical research in economics, where before you had worked solely on theory. How was this transition for you? Why do you find it interesting to do both?
I was really motivated to go into economics by policy problems, but during my PhD I focused on theory, because you have to focus while you are doing your PhD. Then as a junior faculty my mentors encouraged me to keep my focus, because that is the advice that everybody gets.
Jean-Jacques Laffont had these really lovely theoretical papers about how you do empirical work. He showed me that if you think clearly theoretically, then you can make a different kind of contribution to empirical work.
There are really two parts to empirical work: one is just the theory of what is possible in empirical work. We call that econometric identification. In other words, you are trying to answer the question: if you had a very large data set, what is it that it is possible to learn from it? That was what I saw Jean-Jacques Laffont do that was so inspiring to me, because I realised that I could do that kind of theory. And there were a lot of open questions like that, especially in the space of problems where you were bringing theory to data. So auctions were a perfect application of this type of work: I was already doing theory of auctions, but now I could do theory of how to use auction data to answer questions. That connexion then made it easier to transition to just doing the empirical work, for example testing theory with data. So Jean-Jacques showed me a clear path from theory to empirical work. In my research today I am having kind of a similar type of focus, but this time around I’m thinking about these questions in the areas of big data and machine learning. I am trying to determine the methodologies that allow you to draw inferences from big data sets, in particular causal inference. Machine learning is all about prediction: how do you use some things you observe about individuals to predict an outcome that you observe. Economics and social sciences are mostly about causal inference: what is the effect of Google News on news outlets? What is the effect of trade policy on employment? Machine learning hasn’t focused on that as much, and so now I am working on statistical theory for how you use big data to answer this type of causal questions. So I have come back to theory. Basically, I started with economic theory, then did theory of empirical work, then empirical work, and now I have come back to statistical theory again. Overall, I think that if you have good training in theory and you apply it to empirical problems, you can really bring a new set of insights to empirical problems.
- You were the first woman economist to be awarded the John Bates Clark Medal in 2007. That same year, you wrote for the Committee for the Status of Women in the Economics Profession giving advice to women economists in negotiating senior job offers. Do you think that the challenges facing women in economics are the same today as they were a decade ago?
I think that the challenges are very similar. A few things have changed and I have also gained more experience. Ten years ago, I would have said that things are hardest at the beginning and then get easier as you get older. Early on in your career, when you don’t know whether you are good at things and nobody else does either, there’s a lot of learning. Stereotypes can then matter a lot, given that there is not a lot of information. In my case, I was trying to do very technical theory but I somehow didn’t look like what people expected someone who does economic theory to look like. I didn’t even talk the way someone who you expected to be very good at math to talk. I had to counter stereotypes. People would ask: “Is she serious? Can she really be serious? She smiles too much” [laughs]. And in a sense they were right that I was more interested in connection to the world than your typical theorist. So in some ways the way that people sized me up was accurate. But on the other hand, I was perfectly capable of doing hard theory. I was good at math, I just cared about how theory could be used to change the world.
I felt that as I got more experienced and people got to know me as a person, they thought less about my gender and more about my work. So when I met somebody, they would say: “Oh, you wrote this and this paper”. And then that makes everything about your work and all stereotypes –about your ethnicity, your gender, your age– become more or less irrelevant. People just want to talk about your paper. The more papers you have, the easier it is – you could be purple if you have enough great papers.
So I thought that all my problems would be solved once I wrote enough papers because that is all people would care about. But then I realized – in the last 10 or 15 years – that different problems come when you become more senior because your actual job changes. When you are young, your job is just to write papers and all you worry about is whether people will accept your papers or whether they will stereotype how smart you are. When you are older, you are supposed to be a leader in a lot of different ways. You have to make decisions about hiring; you have to advise students; people listen to you in terms of leadership about where the field is going; you have evaluations. You have power dynamics; one group wants one thing, another group wants something else. I think gender plays a role there too, and in some ways, it’s even a harder role to overcome because you can’t just write more papers or win more prizes and make people feel comfortable with you being a strong woman. So that is kind of depressing.
Out of that experience, I have somewhat depressing advice as well [laughs], which is that it’s actually – I hope it changes one day – but it’s actually very difficult for women to be involved in conflicts and power struggles. I think there’s a big disadvantage there that is hard to overcome. So my kind of depressing advice is just to stay out of it.
But the good thing about this advice is that there are so many other interesting things to do, even outside research or academia. For example, I have found that when a government or company comes to me for my advice, once again, they don’t care if I’m purple as long as I’m giving them good advice. They come to me because they want an answer to a question. So that is a way you can have a huge impact. You can have an enormous amount of power through your ideas.
For example, when I worked as consulting chief economist at Microsoft, I changed a lot of things that Microsoft did and that is really cool. Now I am advising start-up companies, among them a company that is trying to reinvent finance. I change the way money moves around the world. I am advising the US government on how to use big data in statistics–that has a huge impact on public policy. That is probably a better use of my time than worrying about university politics or things like that. Since another stereotype is that women get overloaded with administrative jobs, perhaps my advice can serve as a balance against that.
So my advice to women is that there are some environments where gender is an impediment and there are some environments where it is less so. Being an expert and using your expertise to change the world is something incredibly powerful. If you are the world expert on something, you have all the power. When you have unique skills, when you are the best person to solve a problem, people will come to you, and they will do what it takes to make the situation work for you. On the other hand, if the battle is just about power, about whose opinion or feelings matter more, or who should get to speak in a meeting, then even if you are the expert, gender can get in the way. Rather than fight those battles, when given the choice, change the world through channels where you don’t have to waste a lot of time working against gender barriers.
- So we are going to ask you for advice for students in general. Right now, there are more and more jobs in technology. How do you think the current teaching in economics helps students to find this kind of jobs?
First of all, I think Europe is behind the US in terms of educating people for those jobs. So just to think about what the standard is: my ten-year old daughter has been coding in summer camp for four years. Most of upper-middle class kids in the United States will have exposure to coding in elementary school. So when those people get to college, which is only five to eight years away, the standard will be that everybody knows how to code.
At Stanford, we have about a thousand people a year taking our course on machine learning. That is basically saying everybody is going to know this. If you look at business people today, most took some introductory economics. In the near future, everybody will have introductory coding and introductory machine learning. So basically, there are going to be declining industries and growing industries, and people who don’t have these skills are in a declining industry. So, don’t be in the declining industry [laughs].
You need to have the basic skills. If you are a salesperson at Google and you can’t use SQL code to pull your own data and produce a data-driven presentation to your customers, you won’t last more than six months. So, you can’t even be a salesperson without being able to use data. You have to have these skills. But the good news is that it is easy! If a thousand people can take a class, it can’t be that hard [laughs]. And if an eight-year-old can learn how to code, you can, too. You can use Khan Academy, you can use Coursera… It is all available to you; you just take the time to do it.
More specifically, I would say that I am very confident about the fact that economists will have a huge role to play in our society in the future because we know how to use data and we know how to think about equilibrium, feedback effects and incentives. Those thousand people taking the machine learning class aren’t thinking about incentives. They also aren’t thinking about causal inference. They don’t even necessarily think about how to use data to answer a question beyond prediction. Using data to answer questions and evaluate policies is what economics is all about. Most of our research papers are either modelling equilibrium and incentives or they’re measuring the causal effect on something.
And it turns out that is what we need businesses to do. If we have all of these robots and algorithms to do things in the future, we need people to understand them, to manage them, to think about them, to understand how to measure them. We need people in the future to be able to manage and to evaluate algorithms and to put goals on them. Just as an example, take robo-advising: I am going to have an algorithm to tell you how to invest. Well, how do I know if it’s a good algorithm? How do I measure if it’s doing a good job? You can imagine that consumers spend more time with it if you show them lots of cool pictures about how great their retirement will be. They might spend less time with the app if you show them that they probably won’t be able to retire until they’re seventy-five because they haven’t saved enough. If you just take an engineering perspective and measure user experience, you can end up with an algorithm that does not serve your consumers well. This is just an example where thinking about things more economically (about the trade-offs, what the goals and objectives are) could lead you to a different answer. In the future our world is going to be run by these algorithms. If we tell them to do the wrong thing, they will do the wrong thing very well and very fast. So it’s important to look at the long term, the short term, how you measure it, what the objectives are and what the economic context is. But economists will not be able to have a seat at the table unless they have enough technical skills.
by Philip Hanspach and María Paula Caldas