Matthew GentzkowPhoto by Peter Tenzer
Before Matthew Gentzkow entered the field, the economics of media was largely uncharted territory. Today, media economics is flourishing thanks largely to him and his co-authors—particularly Jesse Shapiro, a frequent collaborator. But Gentzkow’s expertise is not confined to media; he’s also a pioneer in methodology, empirical procedure and economic theory with landmark research on communication, social influence and marketing.
With unique insights, innovative technique, methodological rigor and massive databases he often creates for an express purpose, Gentzkow has answered questions about television, newspapers, product branding, competition, persuasion and politics that many scholars had asked but no one had answered convincingly.
Due to this work, we now know that newspaper media slant is driven mostly by the preferences of readers, not newspaper owners. And by examining browser data, he discovered that people don’t largely live in internet “echo chambers”—that is, they don’t exclusively visit sites that align with their political bent. Product brand preferences, he found, are established early in life and endure long after exposure to essentially identical, less expensive alternatives. These and dozens of other economic mysteries have yielded to his curiosity, insight and skill.
Gentzkow received the John Bates Clark Medal in 2014, recognizing him as the best American economist under 40. “He applies frontier methods in empirics and theory to an important set of questions,” noted the Clark citation, and “has established himself as a role model in both substance and execution.” He’s been honored with fellowships in the American Academy of Arts & Sciences and the Econometric Society and was just selected for the 2016 Calvó-Armengol International Prize, a European award for economists 40 and under, comparable in some respects to the Clark award.
In 2015, he moved to Stanford University from the University of Chicago, where he’d taught since getting his Ph.D. from Harvard in 2004. He sustains close UChicago ties through the Becker Friedman Institute’s annual media and communications conferences, the most recent in November 2015.
The following interview covers only a fraction of Gentzkow’s existing research. And neither he nor media is resting. “The rate at which we’re learning about media is very, very fast,” he observes. “And the rates at which technology is advancing and data availability is improving keep accelerating. All of this means that in 10 years, looking back over the work—it’s going to be very exciting.”
Interview conducted March 9, 2016.
TV and children
Region: Let’s begin with two studies you’ve done on the impact of television, starting with your 2008 paper, co-authored by Jesse Shapiro, on the effect of TV viewing on preschool-aged children. People—parents and educators especially—had long worried that TV was “rotting the brains” of the kids who watched it. Your research indicated otherwise.
Could you tell us about that, briefly describing both method and results?
Gentzkow: That paper grew out of earlier work I’d done looking at the effect of TV on voting. For the earlier paper, I developed a strategy of using the rollout of TV across different markets. In the United States, television was introduced to different cities in different years, basically because of idiosyncrasies in the regulatory process.
There is an enormous literature on the effects of TV on kids, and it has essentially been limited to looking at correlations—cross section for the most part—between how much TV kids watch and things like their test scores. But learning anything from those correlations is hard. It’s not only that there are lots of things that are correlated with how much TV kids watch and also might affect their outcomes; it’s hard to think of anything that determines how much TV kids watch that isn’t correlated with their parents, their environment, their intelligence. So, looking at a kid who watches four hours of TV a day and another kid whose parents don’t let him watch any TV—it’s hard to learn much from that.
If you focus on kids who come from more disadvantaged backgrounds, you actually start to see marginally significant positive effects on test scores from watching TV.
This reflects what I think is an important conceptual point: You can’t talk about the effect of TV without thinking about what it’s crowding out.
We wanted to find something that could get closer to causal effects, and this variation across cities in the rollout of TV seemed like a good candidate.
I think it was Jesse who had the insight about using the Coleman Report, a huge study that was done in the U.S., by far the largest survey of its kind up to that time—a half million kids participated. It was designed to look at the educational system—particularly racial disparities—and the database had detailed information on test scores.
The timing of the Coleman Report was perfect for looking at this question about TV’s impact. It was done in the early 1960s, looking at high-school-aged children, among others. Kids who were in high school at the time of the Coleman Report are exactly the kids who, depending on where they grew up, might or might not have had television as preschoolers. So, we were able to put that Coleman outcome data together with this variation across cities in TV rollout and get what we thought were closer to causal estimates of the impact of TV.
We found, first, that the overall point estimate of TV’s impact is basically zero, with a pretty tight confidence interval, so you can rule out even pretty small negative effects of television, on average.
We also found what look like positive effects on certain groups of kids. If you focus on kids who come from more disadvantaged backgrounds—for example, nonwhite kids or kids whose parents don’t speak English as a first language—you actually start to see marginally significant positive effects on test scores from watching TV.
This reflects what I think is an important conceptual point—that took a while to really sink in for us—which is that you can’t talk about the effect of TV without thinking about what it’s crowding out. TV viewing is shifting time around. And, really, for any new technology, any change that is shifting the allocation of time, its effect is the effect of that technology relative to whatever you would have been doing otherwise.
That has pretty important implications for this question because if you think about children of different backgrounds and what else they might be doing with their time, it’s easy to imagine that for some kids, watching television is a much richer source of input than a lot of what it might be crowding out. TV has lots of language; it exposes them to lots of different people and ideas.
It’s also easy to imagine kids for whom it could be a lot worse than whatever else they would have been doing. Educated, wealthy parents or parents with a lot of time to invest in their kids might be taking them to museums and doing math problems with them and so forth. I think part of the reason so many people writing about this assume TV is bad is that they themselves are in the latter group.
TV and politics
Region: You found a less promising, or less positive, outcome regarding TV and political participation, voting in particular. Can you elaborate on that?
Gentzkow: This paper uses the same kind of strategy as the kids paper. I worked on it in grad school, which was where I first put together the data on TV’s rollout across different cities. The origin of that paper, actually, was that I had been doing some other work for my dissertation on how the Internet was affecting the market for print newspapers. I was thinking about substitution and how the new technologies affect old technologies. So I initially got all this data on the introduction of TV to think about how its introduction affected newspapers. How much substitution was there: Did people stop reading newspapers? And, indeed, you see that in the data.
Photo by Peter Tenzer
But that didn’t seem so interesting by itself; it’s not surprising that TV crowded out newspaper consumption. Reflecting on what else might have changed, I started thinking about this huge, downward trend that we’ve seen since about the middle of the 20th century in voter turnout and political participation. It’s really around the time that TV was introduced that that trend in the time series changes sharply, so I thought TV could have played a role.
Now, a priori, you could easily imagine it going either way. There’s a lot of evidence before and since that in many contexts, giving people more information has a very robust positive effect on political participation and voting. So, if you think of TV as the new source of information, a new technology for delivering political information, you might expect the effect to be positive. And, indeed, many people at the time predicted that this would be a very good thing for political participation.
On the other hand, TV isn’t just political information; it’s also a lot of entertainment. And in that research, I found that what seemed to be true is that the more important effect of TV is to substitute for—crowd out—a lot of other media like newspapers and radio that on net had more political content. Although there was some political content on TV, it was much smaller, and particularly much smaller for local or state level politics, which obviously the national TV networks are not going to cover.
So, we see that when television is introduced, indeed, voter turnout starts to decline. We can use this variation across different places and see that that sharp drop in voter turnout coincides with the timing of when TV came in.
The more important effect of TV is to substitute for media that on net had more political content. So, we see that when television is introduced, indeed, voter turnout starts to decline. That drop is especially big in local elections. A lot of new technologies … are pushing people toward paying less attention to local politics, local issues, local communities.
That drop is especially big for a turnout in local elections. There’s a smaller and less significant effect on national turnout, but it’s particularly strong on local turnout. Other people have worked on that in other contexts. I think there’s a general sense that as a lot of new technologies, including the Internet as well, have increased particularly in national news and political coverage, these forces are pushing people toward paying less attention to local politics, local issues, local communities, and they’re being pulled toward either more entertainment options or more attention to national-level politics.
Region: So Tip O’Neill’s comment that “all politics is local” is less true than it was in his day.
Gentzkow: Yes, exactly. In looking back at those two papers, the kids and TV paper and the politics and TV paper, I realized that they both have this theme that the effects of TV can be different from what you might expect because of what TV is crowding out. That wasn’t something I thought about before that research, but you see that idea popping up in lots of other contexts.
Region: In the run-up to an election, political ads on TV and elsewhere seek to persuade voters one way or another. With Emir Kamenica, you’ve done several theoretical papers on persuasion, using game theory to investigate the interaction between “sender” and “receiver.” In politics, that would be the candidate and the voter; in product markets, the company and customers; in a courtroom, lawyer and the jury. The work is pretty technical, but could you tell us a bit about it, and how it’s relevant in real-world settings?
Gentzkow: As you mention, we’ve worked on a couple of projects. Our first paper is about a setting with just a single sender and a single receiver. We were trying to understand what versions of persuasion that we see in the real world are consistent with rationality on the part of senders and receivers.
When you think about political campaigns and advertising, or product advertising and salespeople knocking on your door and trying to convince you to buy a vacuum cleaner, these are settings where we have a pretty strong intuition that there may be a lot of important non-rational components to the persuasion.
However, intuitions about where the line falls between those things—what we would expect to see in a world where things are rational versus a world where things are not rational—had not been drawn very clearly; at least we didn’t have a clear understanding of where that line fell.
We started by thinking about just how much persuasion can happen in a setting where both sides are rational. The receiver knows that the sender is trying to persuade them, and the sender knows that the receiver knows that they’re trying to persuade them.
There’s a sort of naïve intuition that persuasion wouldn’t be possible in that setting because if I exaggerate, if I tell you that my vacuum cleaner works much better than you might have thought it would, you know that I have incentives to distort in that direction. And so you should sort of invert that, or undo whatever exaggeration I’m doing, to get back to the truth. And there are simple cases in which that’s right, where I’m not going to be able to systematically change your behavior if you’re rational.
But what we find in that paper, which pulled together threads that were in previous work by others, is that more generally that’s not true. There are many ways in which, by controlling the way I communicate—controlling information that is made available—I actually can systematically persuade you, systematically influence your behavior, even if we’re all rational.
How does it show up in the real world? Well, I think one of the core intuitions is that if I’m trying to persuade you of something, I can often benefit by either having access to or gathering less-than-complete information about the question at stake.
Photo by Peter Tenzer
Region: Does that mean the sender gathers complete information, but doesn’t communicate all of it? That is, the sender will withhold?
Gentzkow: Well, there are a couple of ways to think about the model. One is that it’s a model of a sender who has acquired knowledge about something exogenously, then choosing what to communicate. And the key assumption in our model then (relative to some past work on strategic communication) is, in that context, it’s a sender who can commit to the rule by which they disclose things.
So, I can commit to reveal something or I can commit not to reveal something. That sets aside a bunch of other really important issues in communication: When am I going to hide things? When am I going to distort things? When are there going to be incentive problems?
The other way to think about the model is that it’s one where I, as the sender, am choosing how much information to gather, or how much information you’ll have access to.
Think about a prosecutor in a courtroom deciding which witnesses to interview, which forensic tests to commission and so forth, in a setting where whatever information I do gather will be made available to the other side.
So, those are both plausible ways to interpret that model. But the core idea that comes out of it is that I can often benefit by not revealing everything. And, in particular, in any situation where you’re going to be positively inclined enough in my direction to take the action I want—to buy the vacuum cleaner, to convict the defendant—at that point, I don’t want to give you any more information. Once you’re already buying, providing you with additional tests or additional information is only going to hurt me because something you learn could possibly tip you back the other way. On the other hand, once you’re not buying the vacuum cleaner, it’s not costly at all to be totally honest and disclose everything.
That means that most of the time we’ll see less than fully clear, less than fully credible positive claims that are just enough to kind of nudge you over the edge. And then occasionally, when it’s very clear that this is not the right product for the person I’m selling to or it’s very clear that the defendant is probably not guilty, we should sort of concede and make that clear.
But at that higher level that I started with, the broader insight is that there is a lot of scope for persuasion even when both agents are rational, but there are also clear limits to when that can be true and the form that it can take.
So, that paper doesn’t say that that model and the form of persuasion it delivers is the right way to understand all the persuasion we see in the real world. It’s more like what those rational agents are going to do feels right for some of the things we see in the real world, but it feels wrong for others, and the patterns of persuasion that don’t really fit within that probably need some ingredient that’s outside the rational model.
Region: People often note that we have strong preferences for a particular brand of a product, even if it’s essentially identical to another brand. In a 2012 American Economic Review paper with Bart Bronnenberg and Jean-Pierre Dubé, you studied how people’s preferences for one brand of a product change over time, and you find a very high degree of persistence.
Could you describe your method for measuring that persistence and
what it implies for competition in product markets?
Gentzkow: Sure. First just to say that the background of that project for me was that brands seem like one of these things that play an enormous role in the real world but have played a very small role in economics. In most of our models, people who run firms are spending a lot of their time thinking about how to set exactly the right price, and maybe some time thinking about investment and financing. But if you talk to people who actually run companies, a huge part of what they’re thinking about—at least if they’re selling products to consumers—is marketing and branding, and all the issues around that.
How is it that in homogenous goods markets, some firms end up having all this market power? If some brand can get a head start and be the preferred brand for 10 years, this model predicts it’s going to be really, really hard for subsequent firms to come in and compete with them. These preferences seem to be so persistent and so durable.
Region: So branding is something that economics has tended to neglect?
Gentzkow: Neglected, yes. It’s something that’s important in the real world that we need to understand better. In that project, we were trying to understand where these preferences come from. That a consumer is willing to pay much, much more to buy one brand rather than another, even where the two things are physically indistinguishable, is a puzzling thing for economics.
It goes back to a lot of older work and discussion in IO [industrial organization economics], thinking about how firms create or end up with market power in markets that we would have thought to be perfectly competitive: the markets for bleach, for example, or for aspirin, baking soda or canned ground coffee. These are all really homogenous products, and yet people have very strong preferences, and firms are able to earn big rents as a result.
You could think of various stories for where brand preferences come from. Maybe all that matters is what happened in my childhood and what kind of mayonnaise my mother gave me in my sandwich. At the other extreme, you could think that it’s all a function of the current environment that I’m in: what ads were on TV yesterday, what products are in the store, what my friends are doing.
So, at one extreme there is some kind of fixed capital stock that is developed early that people just carry around. At the other extreme, all that matters is what’s happening right now. To try to separate those things, it seemed like you could learn a lot from looking at people who move from a place where one brand is popular to a place where another is popular.
My co-authors, Bart and JP, along with Sanjay Dhar, another co-author of theirs, had written a really important paper in the Journal of Political Economy a couple of years earlier that documented huge differences across U.S. cities in which brands are popular. They showed that that actually is correlated with the timing of which brands were introduced first in those cities, even though all of those introductions happened, for the most part, 50 or 100 years ago and few people remember a time when you couldn’t buy both. Say, for example, that we have two brands that have both been in a particular city for 50 years. If one was introduced 70 years ago and the other 50 years ago, you can predict that the one that’s been there for 70 years is going to have a much bigger market share.
In our study, we built on that fact. We gathered some new data from a survey of consumers who were participants in a Nielson panel that was already tracking all of those people’s purchases. We did a survey to ask them where they were born and how much time they spent living in different places.
Region: How big was your sample again?
Gentzkow: About 50,000 people. The question then is, if somebody was born in a place where Coke is popular and moved to a place where Pepsi is popular, where does that person’s consumption fall in between those products, and how does that depend on how long they’ve been in one place or the other? Under some assumptions, that pattern can identify the process by which this brand capital is formed—that is, how quickly it evolves and how much of the preference we see is coming from the current supply-side environment versus the stock of brand capital from a person’s birthplace.
At one extreme, you could see that all that matters is where you were born; if you were born in a Coke place, you drink Coke for the rest of your life. At the other extreme, even people who moved to the Pepsi place just yesterday would drink as much Pepsi as everybody else there because it’s all about current advertising and prices. Where things fall between those extremes is going to tell you something about how sticky those preferences are.
And as you said, we find quite a lot of persistence. We see that on the day people move from one place to the other, their consumption actually shifts discretely. So if you move to the Pepsi place, you immediately start drinking more Pepsi. Why? Because there actually are differences that you get hit with right away: differences in advertising, in prices, in what’s available on the store shelves. But the remaining gap between you and the person who’s lived there forever closes very, very slowly.
What are the implications for firms and for markets? Basically, this research says, if getting somebody to buy my product today creates a stock of preference that means they’re going to be willing to pay more for my product a long time into the future, then being first to a market or making some other investment that gets you a lot of market share early on has big returns.
And this question that people had speculated about for a long time: How is it that in homogenous goods markets, some firms end up having all this market power? It now starts to seem less mysterious. If there’s an otherwise homogenous goods market where some brand can get a head start and be the preferred brand of consumers for 10 years, this model predicts it’s going to be really, really hard for subsequent firms to come in and compete with them. They’re going to have to discount their prices quite a lot for a long time, do tremendous marketing efforts, do something to close that gap, because these preferences seem to be so persistent and so durable.
Photo by Peter Tenzer
Region: It seems almost like parental imprinting. The first thing you see is...
Gentzkow: Yes. Although, importantly, we also can reject that opposite extreme, that all that matters is what your mother did, because even people who move later in life, in their 30s or 40s, eventually converge to look quite a lot like the people in the place they moved to: They eventually start drinking Pepsi—it just takes 30 years for it to happen. In the meantime, those 30 years provide a lot of profit for Coke.
Region: You wrote a more recent paper with Amy Finkelstein and Heidi Williams about geographic variation in health care, using the same technique of comparing movers and nonmovers. As I recall, Amy was quite surprised that there was even more persistence in health care than in goods. Were you equally surprised?
Gentzkow: The thing that was surprising … Well, imagine watching somebody move, first looking at how their brand preferences change; say they move from a Coke place to a Pepsi place and you see how their soft drink preferences change. Then imagine somebody moving from a place where there’s low spending on health care to a place with high spending, and you see how things change. In what way are those patterns different?
The first thing you can look at is how big the jump is when somebody moves. That’s sort of a direct measure of how important is the stuff you are carrying with you relative to the factors that are specific to the places. How important is your brand capital relative to the prices and the advertising? Or in a health care context, how important are the fixed characteristics of people that are different across places, relative to the doctors, the hospitals and the treatment styles across places.
It turns out the jumps are actually very similar. In both cases, you close about half the gap between the place you start and the place you’re going, and so the share due to stuff people carry with them—their preference capital or their individual health—is about the same.
What’s very different and was a huge surprise to me, not what I would have guessed, is that with brands, you see a slow-but-steady convergence after people move; so, movers steadily buy more and more Pepsi the longer they live there. But in a health care context, we don’t see that at all; your health care consumption changes a discrete amount when you first move, but the trend is totally flat thereafter—it doesn’t converge at all.
I might have thought that if you move to a place that treats people really intensely and the doctors like to prescribe lots of tests and lots of treatments, your consumption would gradually shift over time as you’re exposed to that more and more. But instead it hits you when you get there and then it stays pretty constant.
What does that mean? What it points to for me is that what people are carrying with them is different in the two contexts. In the brand context, it’s really something like habit, the capital stock of habit that is a function of my past experience, but then it’s going to shift slowly over time. In the health context, it instead points you at least toward thinking that what people are carrying around with them is more like a fixed individual characteristic.
What might that be? Well, differences in health status are going to look like that. Suppose the reason people in Minneapolis seem to have really low spending is, in part, that they are just healthy—they have better diets, better exercise, lower smoking rates or other, less-observable reasons why their health status is better. If they move to Texas, they’re going to carry that with them, and it’s not going to diminish or deteriorate as they live in Texas. It’s something innate or pretty fixed.
Now, it’s important to say that in that paper, we’re looking at people in Medicare, so all of the movers are moving after age 65. So when we say things looked fixed, they look fixed for people in that population, meaning they evolved endogenously based on your health behaviors when you were younger, but by the time you’re 65, they don’t change.
Newspapers and politics
Region: You’ve done a great deal of research on newspapers and particularly their relationship to politics. There certainly isn’t time to get to it all. But in a key, early paper with Shapiro, you build a model in which media bias emerges because firms slant their coverage toward their audience to build a reputation for quality. I’m curious to know how that holds up empirically.
I’d also like to ask about your findings on the role of newspaper owners in driving media slant, and what that implies for competition policy for media.
Gentzkow: Let me start with the second and think a little bit more about the first.
One of the things we found in looking at newspapers is that their political slant or political content seems to be driven very strongly by demand from their readers, the fact that people in conservative places want to hear conservative stuff and the converse for liberals. And that slant is basically uncorrelated with anything about their owners, the ownership of the paper.
So if you look at two newspapers that are owned by the same company, or the same individual, they are no more similar to each other than two unconnected newspapers in those same places. For instance, the New York Times Company at one time owned a bunch of newspapers all over the country. Those newspapers did not all look like the New York Times; they looked like other newspapers in the places where they were located.
That speaks directly to regulation because much of the way we regulate media is by regulating media ownership. The premise of a lot of that regulation has explicitly been that having diverse viewpoints, diverse ideas and independent reporting is important for democracy. And that in order to guarantee, we need to have diverse ownership because owners are going to put their own imprint on the content of their media.
Region: The concern that if Rupert Murdoch, for instance, or William Hearst in an earlier era, controls newspapers, radio, et cetera, he’ll use that platform to push his political agenda.
Gentzkow: Right, if Murdoch takes over everything, we’re not going to like it because everything will look like Fox News.
Our research results push back on that and say that, at least in this particular context, ownership is not really the key driver of slant and, in fact, a lot of the driver is actually coming from consumer demand. Not only does that say that you might not need to be as worried about ownership, but it also says that the welfare implications of this are a little more complicated because now consumers are getting what they want.
We might think from a political, democratic point of view that it would be better if the public got different, more diverse information. But there’s going to be a welfare trade-off because we would be giving them content they would prefer less. If we want to give people diverse content that we think is good for democracy, then we have to get them to actually read, watch or consume it. And, you know, giving a bunch of people in conservative places some liberal newspaper—well, our results would suggest they’re not going to read it. So, that seems to have important implications for policy.
But it comes with a really important caveat. The finding that ownership doesn’t matter in terms of a newspaper’s political slant is not a universal result. It doesn’t apply everywhere. It’s a statement about newspaper markets in the United States—a highly commercialized, relatively competitive setting, and a place where the political returns to manipulating the average content of a newspaper might not be all that big.
It could be entirely consistent with those results that in other countries, in other contexts, it may well differ. Does Silvio Berlusconi influence the media in Italy? A lot of evidence suggests yes. Does control by the government of Russia affect the content of the media in Russia? Or even if we were to look at national cable outlets in the U.S., would we be confident that ownership doesn’t matter? I think that’s a pretty big leap; you need to be careful.
My own view would be that, probably more than is often assumed, the fact that Fox News has conservative content in the U.S. is related more to the fact that that’s a very profitable business strategy than to any personal political agenda of Rupert Murdoch. But our results don’t settle that question, and it’s an important question.
People in different places and different backgrounds can have such persistently different beliefs about even factual issues, beliefs that never seem to converge. How do people end up with such different beliefs? Why don’t they converge? The sharp empirical test that’s going to pin [this] down—that, I think, we’re still looking for.
Region: And your earlier paper with Shapiro in which you developed a model about newspapers and political slant: Could you tell us a bit about it and how it fares empirically?
Gentzkow: The theory paper that Jesse and I wrote makes the point that, first of all, there’s a mechanism by which even rational consumers, even consumers who really care about getting the truth, are nevertheless going to demand news in a way that matches what we see empirically. That is, they’re going to demand news that matches their own ideology.
Why is that true? Well, suppose you live in a world where you don’t know ahead of time which sources have accurate information and can be trusted, and which don’t have accurate information and can’t be trusted. In that world, a correct, rational, Bayesian inference is that if you say a bunch of stuff that I think a priori is incredibly unlikely to be true, I’m going to trust you less. And if you say stuff that sounds to me like it’s probably right, that is consistent with my prior beliefs about what’s most likely to be true, I’ll trust you more.
It’s obvious in the extremes. If we go into a supermarket and see a tabloid newspaper reporting that Elvis Presley was spotted in New York or that aliens came down from outer space, and you have a strong prior belief that that’s not true, then even if we’ve never seen that newspaper before, we’ll infer that it’s probably not a very accurate newspaper. It’s a totally reasonable judgment that nobody would take issue with.
That same kind of judgment leads to things such as, if I’m somebody who believes very strongly that global warming is a hoax or that the evidence for it is weak, and a lot of people I’ve talked to believe that global warming has been exaggerated, then if I see news outlets that are arguing otherwise, I’m going to trust them less. And if I see news outlets that are skeptical about global warming, I’m going to trust them more.
You can see that playing out on lots and lots of political issues. So, on net, if I’m conservative, I will sincerely believe that Fox News is a more trustworthy source of information. I’m not simply watching Fox News because, “Well, I know that it’s distorted, but it makes me feel better.” That is, I’m not watching it because it confirms my biases and makes me feel good by telling me that I’m right even though I sort of know that it’s less accurate. Rather, I’m watching it because I genuinely think it’s the most accurate source of information there is.
So, that’s what’s true in the world of that model but, as you asked, does that match the facts? Has that been confirmed? I think it resonates very strongly with my casual, anecdotal impression of how people feel about the media choices they make, and it resonates with a lot of survey evidence.
Surveys show that people who, for instance, happen to be liberal and are also consistent readers of the New York Times (including many of our friends in academia) sincerely believe that the New York Times is a trustworthy and accurate newspaper. True, it happens to agree with their political point of view, but if you ask them to bet on a factual question, they would put money on the New York Times being accurate. And surveys also show that people who are conservatives believe exactly the same thing about Fox News.
But that’s all anecdotal, survey-type evidence. We haven’t figured out a way to confirm it empirically, so it remains kind of an open question. We need sharper empirical tests that separate how much of the demand for like-minded information comes from this kind of mechanism versus a variety of other psychological mechanisms that we know are operating.
For example, there’s good evidence that we remember things better when they’re consistent with our prior point of view. There is also a sense of enjoyment at hearing confirmatory information. It’s really fun if you’re a conservative, for instance, to listen to Rush Limbaugh, and it’s really fun if you’re a liberal to listen to Jon Stewart or the “Daily Show”—people making fun of the people you disagree with is really enjoyable. So, there’s definitely that element; the pleasure of hearing somebody reinforce your beliefs is a very real thing.
That tendency of people to seek out like-minded information is so pervasive. You see it in absolutely every context that anyone has ever looked at; you see it for well-educated people and those who are far less-educated. You see it when the stakes are low as well as when they’re high. You see it everywhere and all the time. And some big component of that fact that it’s so robust and pervasive is related to this underlying fact that it’s also rationally what you would do if you were genuinely trying to figure out what is true.
There’s a broader question that is part of what first got me interested in all of this, which is, how is it that people in different places and different backgrounds can have such persistently different beliefs about even factual issues, beliefs that never seem to converge.
We’ve talked about liberals and conservatives in the U.S., but there are also huge differences across countries. Jesse and I looked at some Gallup data where following 9/11, 75 percent of people across nine Islamic countries believed that the World Trade Center was not destroyed by an airplane that was hijacked by Arab terrorists. They had a variety of other explanations: The CIA did it, Mossad did it or something else. In stark contrast, basically 100 percent of people in the U.S would agree that 9/11 was a terrorist attack.
How do people end up with such different beliefs? Why don’t they converge? Is it all because people are deceiving themselves, or they’re biased, or they want to be told that they’re right? Maybe that’s part of it, but at the root of it, I think, is that figuring out who you can trust is a really, really hard thing. We all start out surrounded by information coming from all these different sources—from our friends, our parents, different media outlets, from the government. We need some point of reference to judge who we’re going to listen to.
So it seems obvious—if you’re sitting in America—that, “Well, of course, the World Trade Center was destroyed by terrorists. Every single news organization that we’ve ever seen agrees with that; every single expert we’ve ever heard from agrees with that.”
But if you’re sitting in Pakistan, it’s not crazy to say, “Well, those are all Western news organizations, they share the same bias, they’re part of the same conspiracy, they’re controlled by the same people. Telling me that a hundred of them say this or that isn’t so different from telling me that one of them says it. And I’ve heard from a bunch of other people and seen some videos on YouTube. There’s actually a lot of evidence on the other side, and so I’m going to make a different judgment.”
So that core question of trust has seemed to me for a long time to be really important, but the sharp empirical test that’s going to pin down how much is due to a particular thing— that, I think, we’re still looking for.
Technology and media
Region: In one of your earliest papers, with Claudia Golden and Ed Glaeser, you examined the rise of the “informative” newspaper, the transformation that occurred between the 1870s and early 1900s as newspapers became less partisan.
You say that it resulted because technological progress increased scale and competition, and caused a dramatic reduction in bias and rhetoric, and that “it seems a reasonable hypothesis that” this “was one of the reasons why the corruption of the Gilded Age was sharply reduced.”
I’m curious to know if you see any parallels with today’s technologies like the Internet, smartphones, the rise of social media. These too are technologies that dramatically increase scale, creating intense competition from the massive proliferation of online sources of news and opinion. Is there a parallel today, relative to levels of bias, rhetoric and corruption?
Gentzkow: A really interesting question. A lot of people would say the opposite: Newspapers in that earlier period might have gotten better and more objective and helped the political system work better, but technological changes today are making media less objective, less reliable and making the political process work worse.
I think to understand how those two things fit together, it’s important to think about a distinction between bias that comes from the supply side of the market—an owner, a political party—versus bias that’s driven by catering to consumer demand.
The way we thought about that historical period in newspapers was that it’s very well documented historically that up until the late 19th century, newspapers were to a substantial degree funded by and explicitly controlled by political parties in this country. Newspapers had explicit affiliations with political parties. Newspaper directories from that period list “This newspaper is Democratic” or “This newspaper is Republican.” Nobody was shy about that. When those newspapers started publishing, they said, for example, “We are going to represent the views of the Republican Party in this state,” and their content reflected that to a significant degree.
As costs of printing newspapers fell dramatically, as the potential profits you could earn from selling newspapers increased, the incentive to instead focus on what consumers want got really big. So you saw a shift. If you started from a situation where the political parties control all the newspapers, the driving force of consumer demand is going to be a demand for more independence.
New technology has made much more [media] diversity possible, much more competition possible. Is this a more competitive marketplace of ideas that does a better job in disciplining government and getting out the truth? Or is it a bunch of echo chambers where … we are going to increase partisan divisions and nobody’s going to learn the truth? The jury is still out.
And that was the buzzword: independence; that was the key. Newspapers were not advertising that they were going to be neutral. They weren’t advertising that they weren’t going to take political positions or weigh in on issues or have a strong point of view. They were advertising that they were going to be independent; they were not going to be subject to explicit control by some outside force; they were going to look at the facts and report the facts as they saw fit.
The evidence that we’ve looked at from the content analysis we can do suggests that when newspapers switched from calling themselves Republican to calling themselves Independent, the left-right slant of their content actually didn’t change at all or very little.
But as we show in the paper with Claudia and Ed, the explicit partisan rhetoric shifts from sounding like an angry opinion page to a more independent, objective coverage of the issues that happens to end up, possibly, in a similar place. Before, we would have said these Republican guys—or these Democrat guys—are crooks. Now, we’re still going to say that they’re crooks, but we’re going to do it in a way that is not because we’re controlled by a party, but because we’ve looked at the facts and we think this is what’s true.
And today? What is happening today is consistent with the idea that the main driver of content in the media remains catering to what consumers demand. It’s just that where we’re starting from is different now than in the late 1800s and, therefore, the direction in which things are changing is different.
New technology has made much more diversity possible, much more competition possible. When I was a kid, when probably all of us grew up, three broadcast networks dominated the news that the vast majority of Americans saw, and those three networks were kind of indistinguishable from each other politically.
That was a very unusual period historically in this country and a very unusual situation across countries. That was really the anomaly, not the norm, and it existed for a bunch of reasons: technological constraints that meant you could only have only three networks; regulatory constraints that I think played a big role in why they weren’t too different from each other.
As the technology has changed, we’re moving back to something that looks more like newspapers both in 1920 and in 1890, where there was much more diversity, much more competition and a much wider array of viewpoints catering to what people want.
I think the jury is still out on is that a good thing or a bad thing. If the rise of independent newspapers helped the reform of corruption and so on, is this increase in competition and diversity today, is that a more competitive marketplace of ideas that does a better job in disciplining government and getting out the truth? Or is it a bunch of echo chambers where everybody is just going to hear what they want and we are going to increase partisan divisions and nobody’s going to learn the truth? There’s still a lot of debate, I think, about how all that’s going to stack up on net what the facts will be.
International media and social media
Region: Your research has made key contributions to methodology and theory, and has spanned a wide range of media from many perspectives, but there are two areas that seem to beg for your attention still. With a few exceptions, one of which you discussed today, your analysis of attitudes across nine predominantly Muslim countries, almost all of your work studies the United States, and I wonder how it would play out in other countries. And relatively little of your research looks at social media, a rapidly growing media phenomenon.
I’m guessing that the lack of large, reliable databases might be one factor, but what would you say are the reasons?
Gentzkow: Honestly, the main reason is just that I haven’t been quick enough to write those papers. Fortunately, other people have. People have done really extraordinary work on both those topics, particularly the first—media outside the U.S. There have been at least a dozen papers written in the last five, six, seven years that I wish I had written because I think they’re incredibly important. There’s been great work on media in Russia, China, Italy, in Europe broadly. And whatever the policy stakes are in the U.S. for how we regulate media, how you think about political effects, those stakes are, if anything, higher if we’re talking about what’s happening in China or Russia.
So, I think that work is just incredibly important. Other researchers have dug up extremely rich data, so I don’t actually think data is a fundamental impediment. The data available to study media are typically very good almost everywhere. Media are funded by advertising, and advertisers need information about who’s watching and reading and what people are doing. So, no matter where you go in the world, people are keeping track of what are the TV stations and where are they available, the newspapers and their circulations, and just generally, what are people watching and how many people are watching different things.
Social media is obviously a more recent phenomenon—also incredibly interesting, both as a media phenomenon and a social phenomenon. The issues we talked about earlier regarding media’s effect on children and the way they’re growing up, for example. Social media is obviously a pretty dramatic change in how kids are growing up today.
People are complaining about social media now. But think back to what they were saying when kids were all watching TV: “It’s this passive thing and they’re not thinking, they’re alone, they’re not communicating!” Now kids are writing text messages and creating pictures and editing them on Instagram. It’s certainly not passive or not solitary. It has its own risks perhaps, but not the risks that worried people about TV.
A lot of people are complaining about social media now. But think back to what they were saying back when kids were all watching TV: “It’s this passive thing where kids sit there and zone out, and they’re not thinking, they’re alone, they’re not communicating!” Now, suddenly, a thing that kids are spending lots of their time doing is interacting with other kids. They’re writing text messages and posts and creating pictures and editing them on Instagram. It’s certainly not passive; it’s certainly not solitary. It has its own risks perhaps, but not the risks that worried people about TV. I think there’s a tendency, no matter what the new technology is, to wring our hands about its terrible implications. Kind of amazing how people have turned on a dime from worrying about one thing to worrying about its exact opposite.
But that’s just one among many interesting questions. On the political side of it, for example, how is social media changing the sort of things that we were talking about before? For example, the extent to which people are segregated in their use of media, and what they’re exposed to. Is it an echo chamber? Are people always hearing like-minded information?
Jesse and I wrote a paper looking at that question circa 2008—just before the real takeoff in social media. At that point, the evidence suggested people are not in an echo chamber nearly as much as you might think. Has social media changed that? I think the best evidence we have so far would say, not so much.
Why? It’s true that social media is significantly segregated; what people click on and see and share tends to be close to their own ideology. We know that social networks are very segregated ideologically, and social media reflects that. It just remains true that the share of news that people are getting through social media is still, on average, a small share of the total.
This is less true of opinion content, but if you look at just news content—reading about the election, say, or the war in Syria—most people are not getting that news through social media. If they did, if we shifted to a situation where actually the way most people get news is from a Facebook feed, that could have pretty dramatic implications. So, I think that’s a really interesting thing to watch going forward.
So, back to your initial question about research on other countries and on social media, I think that there’s just a huge number of interesting and important questions in this area. Maybe I’ll be able to look at a few more of them. Luckily, there are a lot of other really talented scholars digging in and looking at these things.
I think the rate at which we’re learning about media is very, very fast. And the rates at which technology is advancing and data availability is improving keep accelerating. All of this means that in 10 years, looking back over the work—it’s going to be very exciting.