Friday, September 18, 2020

Every Day is a Bad Day, Say a Rising Share of Americans

The Behavioral Risk Factor Surveillance System (BRFSS) is a standardized phone survey about health-related behaviors, carried out by the Centers for Disease Control and Prevention (CDC). One question asks: “Now thinking about your mental health, whicjh includes stress, depression, and problems with emotions, for how many days during the past 30 days was your mental health not good?” 

David G. Blanchflower and Andrew J. Oswald focus on this question in "Trends in Extreme Distress in the United States, 1993–2019" (American Journal of Public Health, October 2020, pp. 1538-1544).  I particular, they focus on the share of people who answer that their mental health was not good for all 30 of the previous 30 days, who they categorize as in a condition of "extreme distress." Here are some patterns: 

This graph shows the overall and steady rise for men and women from 1993-2019. 

Here's a breakdown for a specific age group of those 35-54 years of age, with a simple breakdown by education and by ethnicity. 
This kind of survey evidence doesn't let a researcher test for causality, but it's possible to look at some correlations. The authors write: "Regression analysis revealed that (1) at the personal level, the strongest statistical predictor of extreme distress was `I am unable to work,' and (2) at the state level, a decline in the share of manufacturing jobs was a predictor of greater distress."

Of course, one doesn't want to overinterpret graphs like this. The measures on the left-hand axis are single-digit percentages, after all. But remember, these people are reporting that their mental health hasn't been good for a single day in the last month. The share has been steadily rising over time, through different economic and political conditions. In those pre-COVID days of 2019, 11% of the white, non-college population--call it one out of every nine in this group--reported this form of extreme distress. The implications for both public health and politics seem worth considering. 

Thursday, September 17, 2020

Stock Buybacks: Leverage vs. Managerial Self-Dealing

Consider a company that has been earning profits, and wants to pay or all of those earnings to its shareholders. There are two practical mechanisms for doing so. Traditionally, the best-known approach was for the firm to pay a dividend to shareholders. But in the last few decades, many US firms instead have used stock buybacks. How substantial has this shift been, and what concerns does it raise? 

Here, I'll draw upon a couple of recent discussions of stock buybacks. Siro Aramonte writes about "Mind the buybacks, beware of the leverage," in the BIS Quarterly Review (September 2020, pp. 49-59). Kathleen Kahle and René M. Stulz tackle the topic from a different angle in "Why are Corporate Payouts So High in the 2000s? (NBER Working Paper 26958, April 2020, subscription required). 

Kahle and Stulz present the evidence both that overall corporate payouts to shareholders are up in the 21st century, and that stock buybacks are the primary vehicle by which this has happened. They calculate that total payouts from corporations to shareholders from 2000-2017 (both dividends and share buybacks) were about $10 trillion. They find that corporate payouts to shareholders have risen substantially post-2000, and that stock buybacks are the main vehicle through which this has happened. They write: 
In the 2000s, annual aggregate real payouts average roughly three times their pre-2000 level. ... Specifically, in the aggregate, higher earnings explain 38% of the increase in real constant dollar payouts and higher payout rates account for 62% of the increase. ...

In our data, the growth in payout rates, defined as the ratio of net payouts to operating income, comes entirely from repurchases. This finding is consistent with the evidence in Skinner (2008) on the growing importance of repurchases. Dividends average 14.4% of operating income from 1971 to 1999 and 14% from 2000 to 2017. In contrast, net repurchases, defined as stock purchases minus stocks issuance, average 4.8% of operating income before 2000 and 18.3% from 2000 to 2017.
The tax code offers obvious reasons for share buybacks, rather than dividends, as economists were already discussing back in the 1980s.  Dividends are subject to the personal income tax, and thus taxed at the progressive rates of the income tax. However, the gains of an investor who sells stock back to the company are taxed at the lower rate for capital gains. In addition, when a company pays a dividend, all shareholders receive it, but when a company announced a share buyback, not all shareholders need to participate, if they do not wish to do so. Thus, share buybacks offer investors more flexibility about when and in what form they wish to receive a payout from the firm. 

In addition, economists have also recognized for some decades that corporations will sometimes find themselves in a position of "free cash flow," where the company has enough money that it can make choices about whether it can find productive internal investments for the funds, or whether it will fiud a way to pay out the money to shareholders, or whether it will use the money to pay bonuses and perquisites to managers. If we agree that lavishing additional benefits on managers is not a socially attractive choice, and if the firm honestly doesn't see  how to use the money productively for internal investments, then paying the funds out to shareholders seems the best choice. 

The public response to firms that pay dividends is often rather different than when a firm does a share buyback--even when the same payout is flowing from the firm to its shareholders. The concern sometimes expressed is that corporate managers have an unspoken additional agenda with stock buybacks, which is to pump up the price of the company's stock--and in that way to increase the stock-based performance bonus for the managers.

Sirio Aramonte also documents the substantial rise in stock buybacks in recent decades. He points out that a primary cause for stock buybacks is for firms to increase their leverage--that is, to increase the proportion of their financing that happens through debt. He writes: "Corporate stock buybacks have roughly tripled in the last decade, often to attain desired leverage, or debt as a share of assets." This pattern especially holds true if the firm finances the stock buyback with borrowed money, rather than out of previously earned profits. He writes: 
In 2019, US firms repurchased own shares worth $800 billion (Graph 1, first panel; all figures are in 2019 US dollars). Net of equity issuance, the 2019 tally reached $600 billion. Net buybacks can turn negative, and they did during the GFC [global financial crisis of 2007-9], as firms issued equity to shore up their balance sheets. ... Underscoring the structural differences between dividends and buybacks, the former were remarkably smooth, while the latter proved procyclical and co-moved with equity valuations ...
Aramonte crisply summarizes the case for share buybacks: 
In a number of cases, repurchases improve a firm’s market value. For instance, if managers perceive equity as undervalued, they can credibly signal their assessment to investors through buybacks. In addition, using repurchases to disburse funds when capital gains are taxed less than dividends increases net distributions, all else equal. Furthermore, by substituting equity with debt, firms can lower funding costs when debt risk premia are relatively low, especially in the presence of search for yield. And, by reducing funds that managers can invest at their discretion, repurchases lessen the risk of wasteful expenditures.
What about the concern that corporate managers are using share buybacks to pump up their stock-based bonuses? Aramonte's discussion suggests that this may have been an issue in the past--say, pre-2005--but that the rules have changed. Companies have been shifting away from bonuses based on short-term stock prices, and toward bonuses based on long-term stock value for executives who stay with the firm. There are increased regulations and disclosure rules to limit this practice. Also, if CEOs were using stock buybacks in a short-term pump-and-dump strategy, then the stock price should first jump after a buyback and then fall back to its earlier level--and we don't see this pattern in the data. Thus, this concern that managers are abusing stock buybacks seems overblown. 

What about the linkages from stock buybacks to rising corporate debt? Aramonte provides some evidence, and also refers to the Kayle/Stulz study: 
[B]uybacks were not the main cause of the post-GFC rise in corporate debt. After 2000, internally generated funds became more important in financing buybacks. For one, economic growth resulted in rising profitability. In addition, firms exhibited a higher propensity to distribute available income. Kahle and Stulz (2020) find that cumulative corporate payouts from 2000 to 2018 were higher than those from 1971 to 1999 and that two thirds of the increase was due to this higher propensity.

In short, the overall level of rising corporate debt in recent years is a legitimate cause for concern (as I've noted here, here, and here). Share buybacks are one of the tools that US firms have used to increase their leverage, but the real issue here is whether the higher levels of debt have made US firms shakier, not the use of share buybacks as part of that strategy. The pandemic recession is likely to provide a harsh test of whether firms with more debt are also more vulnerable. As Aramonte writes: 

There is, however, clear evidence that companies make extensive use of share repurchases to meet leverage targets. The initial phase of the pandemic fallout in March 2020 put the spotlight on leverage: irrespective of past buyback activity, firms with high leverage saw considerably lower returns than their low-leverage peers. Thus, investors and policymakers should be mindful of buybacks as a leverage management tool, but they should particularly beware of leverage, as it ultimately matters for economic activity and financial stability.

Wednesday, September 16, 2020

Why Foreign Direct Investment Was Already Sagging

Foreign direct investment (FDI) involves a management component. In other words, it's not just a financial investment in stocks and bonds ("portfolio investment"), but involves partial or in some cases complete management responsibility. This distinction matters for a couple of reasons. One is that for developing countries in particular, FDI from abroad is a way of gaining local access to management skills, technology, and supply chains that might be quite difficult to do on their own. Another reason is that pure financial investments can come and go, sometimes in waves that bring macroeconomic instability in their wake, but FDI is typically less volatile and more of a commitment. 

FDI seems certain to plummet in 2020, given that so many global ties have weakened during the pandemic recession. But as the World Bank points out in its Global Investment Competitiveness Report 2019/2020: Rebuilding Investor Confidence in Times of Uncertainty, a decline in FDI was already underway. Here, I'll quote from the "Overview" of the report by Christine Zhenwei Qiang and Peter Kusek. They write (footnotes omitted): 

Even before the COVID-19 pandemic upended the global economy, global FDI was sliding to levels even below those last seen in the aftermath of the global financial crisis a decade ago (figure O.1, panel a). The decline was more concentrated in high-income countries, where inflows of FDI fell by nearly 60 percent in recent years. Although FDI to developing countries did not decline as steeply, it nonetheless fell to its lowest levels in decades relative to gross domestic product (GDP).  Compared with the mid-2000s, when FDI reached nearly 4 percent of GDP in developing countries, that share fell to under 2 percent in 2017 and 2018 (figure O.1, panel b).

What were the main drivers of this decline before 2020? Qiang and Kusek write that it's been a combination of economic, business, and political factors. They write: 

More specifically, worsening business fundamentals have driven much of the decline in FDI since 2015, when FDI flows reached their post-crisis peak. The global average rate of return on FDI decreased from 8.0 percent in 2010 to 6.8 percent in 2018 (UNCTAD 2019). While the rates of return have dropped in both developing and developed countries, the declines have been especially large in developing countries.

Furthermore, changing business models resulting from technological advances have driven declines in FDI levels and returns. In particular, increases in labor costs and the rise of advanced manufacturing technologies have eroded or decreased the significance of many developing countries’ labor cost advantages. At the same time, the increasing importance of the digital economy and services is shifting businesses toward more asset-light models of investment (UNCTAD 2019). In addition, commodity price slumps have adversely affected returns on FDI in more commodity-dependent markets (such as many economies in Latin America and the Caribbean, the Middle East and North Africa, and Sub-
Saharan Africa)
Countries around the world, including developing countries, have also become less supportive of FDI in recent years. This figure is based on actions by 55 countries, and whether those countries are changing their rules to be more or less favorable to FDI in a given year.

Much of the rest of the report is made up of case studies of the effects of FDI, including how governments can take full advantage of its potential benefits and cushion any resulting disruptions. But for now, that side of the argument seems to be losing ground.

Tuesday, September 15, 2020

Africa is Not Five Countries

Scholars of the continent of Africa sometimes feel moved to expostulate: "Africa is not a country!" In part, they are reacting against a certain habit of speech and writing where someone discusses, say, the United States, China, Germany, and Africa--although only the first three are countries. More broadly, they are offering a reminder that Africa is a vast place, and that generalizations about "Africa" may apply only to some of the 54 countries in Africa

Economic research on "Africa" apparently runs some risk of falling into this trap. Obie Porteous has published a working paper that looks at published economics research on Africa: "Research Deserts and Oases: Evidence from 27 Thousand Economics Journal Articles" (September 8, 2020). Porteus creates a database of all articles related to African countries published between 2000-2019 in peer-reviewed economics journals. He points out that the number of such articles has been rising sharply: "[T]he number of articles about Africa published in peer-reviewed economics journals in the 2010s was more than double the number in the 2000s, more than five times the number in the 1990s, and more than twenty times the number in the 1970s." His data shows over 19,000 published economics article about Africa from 2010-2019, and another 8,000-plus from 2000-2010. 

But the alert reader will notice how easy, as shown in the previous paragraph, to slip into discussing articles "about Africa." Are economists studying a wide range of countries across the continent, or are they studying relatively few countries. Porteous has some discouraging news here: "45% of all economics journal articles and 65% of articles in the top five economics journals are about five countries accounting for just 16% of the continent's population."

The "frequent 5" five much-studied countries are Kenya, South Africa, Ghana, Uganda, and Malawi. As Porteous points out, it's straightforward to compile the "scarce 7": the seven countries Sudan, D.R. Congo, Angola, Somalia, Guinea, Chad, and South Sudan,with the same population as the frequent 5, but account for only 3.5% of all journal articles and 4.7% of articles in the top 5 journals.

What explains what some countries are common locations for economic research while others are not? Porteous writes: "I show that 91% of the variation in the number of articles across countries can be explained by a peacefulness index, the number of international tourist arrivals, having English as an official language, and population." It's certain easier for many economists to do research in English-speaking countries that are peaceful and popular tourist destinations--and that's what has been happening. There's also evidence that even within the highly-researched countries, some geographic areas are more often researched than others. 

Of course, it's often useful for a research paper to focus on a specific situation. The hope is that as such papers accumulate, broad-based lessons begin to emerge that can apply beyond the context of a specific country (or area within a country). But local and national context is often highly relevant to the findings of an economic study. It seems that a lot of what economic research has learned about "Africa" is actually about a smallish slice of the continent. 

Monday, September 14, 2020

CEO/Worker Pay Ratios: Some Snapshots

Each year, US corporations are required to report the pay for their chief executive officers, and also to report the ratio of CEO pay to the pay of the median worker at the company. Lawrence Mishel and Jori Kandra report the results for 2019 pay in "CEO compensation surged 14% in 2019 to $21.3 million: CEOs now earn 320 times as much as a typical worker" (Economic Policy Institute, August 18, 2020). 

Back in the 1970s and 1980s, it was common for CEOs to be paid something like 30-60 times the wage of a typical worker. In 2019, the ratio was a multiple of 320. 
A result of this shift is that while CEOs used to be paid three times as much as the top 0,1% of the income distribution, now they are paid about six times as much. 
What is driving this higher CEO pay ratio? In an immediate sense, the higher pay seems to reflect changes in the stock market. The left-hand margin shows CEO pay; the right-hand margin shows the stock market as measured by the S&P 500 index. 
This rise in CEO/worker pay ratios has led to a continually simmering argument about the underlying causes. Does the rise reflect the market for talent, in the sense that that running a company in a world of globalization and technological change has gotten harder, and the rewards for those who do it well are necessarily greater? Or does it reflect a greater ability of CEOs to take advantage of their position in large companies to grab a bigger share of the economic pie? One's answer to this question will turn, at least in part, on whether you think CEOs have played a major role in the rise of the stock market since about 1990, or whether you think they have just been riding along on a stock market that has risen for other reasons. For an example of this dispute from a few years ago in the Journal of Economic Perspectives (where I work as Managing Editor), I recommend: 
Without trying to resolve that dispute here, I'd offer this thought: Notice that pretty much all of the increase in CEO/worker pay ratios happened in the 1990s, and the ratio has been at about the same level since then. Thus, if you think that the market for executive talent was rewarding CEOs appropriately, you need an explanation for why the increase happened all at once in about a decade, without much change since then. If you think the reason is that CEOs are grabbing a bigger share of the pie, you need an explanation for why CEOs became so much more able to do that in the 1990s, but then their ability to grab even-larger shares of the pie seemed to halt at that point. To put it another way, when discussing a change that happened in the 1990s, you need an explanation specific to the 1990s. 

I don't have a complete explanation to offer, but one obvious possible cause was in 1993, when  Congress and the Clinton administration enacted a bill with the goal of holding down the rise in executive pay (visible in the first graph above). Up into the 1980s, most top executives had been paid on via annual salary-plus-a-bonus. However, the new law put a $1 million cap on salaries for top executive, and instead required that other pay be linked to performance--which in practice meant giving stock options to executives. Although this law was intended to hold down executive pay, the was The stock market more-or-less tripled in value from late 1994 to late 1999, and so those who had stock options did very well indeed. My own belief is that combination of events reset the common expectations for what top executives would be paid, and how they would be paid, in a way that is a primary driver of the overall rise in inequality of incomes in recent decades. 

Friday, September 11, 2020

100 Million Traffic Stops: Evidence on Racial Discrimination

 A primary challenge in doing research on racial discrimination is that you need to answer the "what if" questions. For example, it's not enough for research to show that blacks are pulled over by police for traffic stops more often than whites. What if more blacks were driving in a way that caused them to be pulled over more often? A researcher can't just dismiss that possibility. Instead, you need to find a way to think about the available data in a way that addresses these kinds of  "what if" questions. 

When it comes to traffic stops, for example, one approach is to look at such stops in the shifting time window between daytime and darkness. For example, compare the rate at which blacks and whites are pulled over for traffic stops in a certain city during a time of year when it's light outside at 7 pm and at a time of year when it's dark outside at 7 pm. One key difference here is that when it's light outside, it's a lot easier for the police to see the race of the driver. If the black-white difference in traffic stops around 7 in the evening is a lot larger when it's light at that hour than when it's dark at that hour, then racial discrimination is a plausible answer.  Taking this idea a step further, a researcher can look at the time period just before and after the Daylight Savings Time time shifts.

A team of authors use this approach and others in "A large-scale analysis of racial disparities in police stops across the United States," published in Nature Human Behavior (July 2020, pp. 736-745, authors are  Emma Pierson, Camelia Simoiu, Jan Overgoor , Sam Corbett-Davies, Daniel Jenson, Amy Shoemaker , Vignesh Ramachandran, Phoebe Barghouty, Cheryl Phillips, Ravi Shroff and Sharad Goel ). The authors make public records request in all 50 states, but (so far) have ended up with "a dataset detailing nearly 100 million traffic stops carried out by 21 state patrol agencies and 35 municipal police departments over almost a decade." Their analysis sounds like this: 

In particular, among state patrol stops, the annual per-capita stop rate for black drivers was 0.10 compared to 0.07 for white drivers; and among municipal police stops, the annual per-capita stop rate for black drivers was 0.20 compared to 0.14 for white drivers. For Hispanic drivers, however, we found that stop rates were lower than for white drivers: 0.05 for stops conducted by state patrol (compared to 0.07 for white drivers) and 0.09 for those conducted by municipal police departments (compared to 0.14 for white drivers). ... 

These numbers are a starting point for understanding racial disparities in traffic stops, but they do not, per se, provide strong evidence of racially disparate treatment. In particular, per-capita stop rates do not account for possible race-specific differences in driving behaviour, including amount of time spent on the road and adherence to traffic laws. For example, if black drivers, hypothetically, spend more time on the road than white drivers, that could explain the higher stop rates we see for the former, even in the absence of discrimination. Moreover, drivers may not live in the jurisdictions where they were stopped, further complicating the interpretation of population benchmarks.

But here's some data from the Texas State Patrol on the share of blacks stopped in different evening time windows: 7:00-7:15, 7:15-7:30, and 7:30-7:45. A vertical line shows "dusk," considered the time when it is dark. The researchers ignore the 30 minutes before dusk, when the light is fading, and focus on when the period before and after that window. You can see that the share of black drivers stopped is higher in the daylight, and then lower after dark.

Another test for racial discrimination looks at the rate in which cars are searched, and then looks at the success rate of those searches. Interpreting the result of this kind of test can be mildly complex, and it's useful to go through two steps to understand the analysis. The the authors explain the first step in this way: 
In these jurisdictions, stopped black and Hispanic drivers were searched about twice as often as stopped white drivers. To assess whether this gap resulted from biased decision-making, we apply the outcome test, originally proposed by Becker, to circumvent omitted variable bias in traditional tests of discrimination. The outcome test is based not on the search rate but on the ‘hit rate’: the proportion of searches that successfully turn up contraband. Becker argued that even if minority drivers are more likely to carry contraband, in the absence of discrimination, searched minorities should still be found to have contraband at the same rate as searched whites. If searches of minorities are successful less often than searches of whites, it suggests that officers are applying a double standard, searching minorities on the basis of less evidence. ... 

Across jurisdictions, we consistently found that searches of Hispanic drivers were less successful than those of white drivers. However, searches of white and black drivers had more comparable hit rates. The outcome test thus indicates that search decisions may be biased against Hispanic drivers, but the evidence is more ambiguous for black drivers.

This approach sounds plausible, but if you think about it a little more deeply, it's straightforward to come up with examples where might not work so well. Here's an example: 

[S]uppose that there are two, easily distinguishable, types of white driver: those who have a 5% chance of carrying contraband and those who have a 75% chance of carrying contraband. Likewise assume that black drivers have either a 5 or 50% chance of carrying contraband. If officers search drivers who are at least 10% likely to be carrying contraband, then searches of white drivers will be successful 75% of the time whereas searches of black drivers will be successful only 50% of the time. Thus, although the search criterion is applied in a race-neutral manner, the hit rate for black drivers is lower than that for white drivers and the outcome test would (incorrectly) conclude that searches are biased against black drivers. The outcome test can similarly fail to detect discrimination when it is present.
To put it another way, the decision to search a vehicle is binary: you do it or you don't do it. Thus, the key issue is the threshold that a police officer applies in deciding to search. As in this example, you can think of the threshold in this way: if the percentage chance of finding something is above the threshold level, a search happens; if it's below that level, a search doesn't happen. The next step is to estimate these threshold probabilities: 
In aggregate across cities, the inferred threshold for white drivers is 10.0% compared to 5.0 and 4.6% for black and Hispanic drivers, respectively. ... Compared to by-location hit rates, the threshold test more strongly suggests discrimination against black drivers, particularly for municipal stops. Consistent with past work, this difference appears to be driven by a small but disproportionate number of black drivers who have a high inferred likelihood of carrying contraband. Thus, even though the threshold test finds that the bar for searching black drivers is lower than that for white drivers, these groups have more similar hit rates.
A short takeaway from this research is that when blacks complain about being stopped more often by police, there is solid research evidence backing up this claim. The evidence on blacks being searched more often in a traffic stop is real, but probably best-viewed as a little weaker, because it doesn't show up in the basic "success rate of searches" data and instead requires the more complex threshold analysis. 

For other discussions of how social scientists try to pin down evidence the extent to which racial discrimination underlies racial disparities, see: 

Wednesday, September 9, 2020

Misperceptions and Misinformation in Elections Campaigns

It's an election season, so many people are widely concerned about  how all those other voters are going to be misinformed into voting for the wrong candidate. Brendan Nyhan provides an overview of some research in this area in "Facts and Myths about Misperceptions" (Journal of Economic Perspectives, Summer 2020, 34:3, pp. 220-36). 

To be clear, Nyhan describes misperceptions as "belief in claims that can be shown to be false (for example, that Osama bin Laden is still alive) or unsupported by convincing and systematic evidence (for example, that vaccines cause autism)." Thus, he isn't talking about issues of shading or emphasis. Nyhan writes: "Misperceptions present a serious problem, but claims that we live in a `post-truth' society with widespread consumption of `fake news' are not empirically supported and should not be used to support interventions that threaten democratic values." 

So why is the belief that everyone on the other side of the political fence is subject to dramatic misperceptions so widespread. One reason is that both academic research and examples of that research in the media tend to focus on examples with partisan distinctions. 
Public beliefs in such claims are frequently associated with people’s candidate preferences and partisanship. One December 2016 poll found that 62 percent of Trump supporters endorsed the baseless claim that millions of illegal votes were cast in the 2016 election, compared to 25 percent of supporters of Hillary Clinton (Frankovic 2016). Conversely, 50 percent of Clinton voters endorsed the false claim that Russia tampered with vote tallies to help Trump, compared to only 9 percent of Trump voters. But not all political misperceptions have a clear partisan valence: for example, 17 percent of Clinton supporters and 15 percent of Trump supporters in the same poll said the US government helped plan the terrorist attacks of September 11, 2001.

One of my favorite examples is a study which showed respondents pictures of the Inauguration Day crowds for  President Obama in 2009 and President Trump in 2017.: "When the pictures were unlabeled, there was broad agreement that the Obama crowd was larger, but when the pictures were labelled, many Trump supporters looked at the pictures and indicated that Trump’ crowd was larger, an obviously false claim that the authors refer to as `expressive responding.'” (I love the term "expressive responding.")

Sometimes that people are aware of slanting their answers in this way. When people give these kinds of answers to poll questions, they often know (and will say when asked) that some of their answers are based on less evidence than others. One study offered small financial incentives (like $1) for accurate answers, and found that the partisan divide was reduce by more than 50%.  

But other times, people make meaningful real-world decisions based on these kinds of partisan feelings. as one example with particular relevance just now, evidence from the George W. Bush and Barack Obama administrations suggests that when the president you supported is in office, people "express more trust in vaccine safety and greater intention to vaccinate themselves and their children than opposition partisans," which shows up in actual patterns of school vaccinations. 

An underlying pattern that comes up in this research is that if people are exposed to an concept many times (an example is the false statement “The Atlantic Ocean is the largest ocean on Earth”), they become more likely to rate it as true. The underlying psychology here seems to be that when a claim seems familiar to people, because of repeated prior exposure, they become more likely to view it as true. An implication here is that while those who marinate themselves in social media discussions of news may be more likely to think of themselves as well-informed, they are also probably more likely to have severe misperceptions. Indeed, people who are more knowledgeable are also the same people who have become aware of how to deploy counterarguments so that they believe their misperceptions even more strongly. 

Nyhan's paper mentions many intriguing studies along these lines. But do we need public action to fight misperceptions? It's not clear that we do. A common finding in these studies is that if someone discovers and admits that they have a misperception on a certain issue, it doesn't actually change their partisan beliefs.  "Fact-checking" websites have some use, but they can also be another way of expressing partisanship--and those who hold misperceptions most strongly are not likely to be reading fact-checking sites, anyway. Even general warnings about "fake news" can backfire. Some research suggests that when people are warned about fake news, they become skeptical of all news, not just part of it. One interesting study warned a random selection of candidates in nine states who were running for office in 2012 that the reputational effects of being called out by fact-checkers could be severe, and found that candidates who received the warnings were less likely to have their accuracy publicly challenged. 

Nyhan concludes with this response to suggestions for more severe and perhaps government-based interventions against misperceptions: 

Calls for such draconian interventions are commonly fueled by a moral panic over claims that “fake news” has created a supposedly “post-truth” era. These claims falsely suggest an earlier fictitious golden age in which political debate was based on facts and truth. In reality, false information, misperceptions, and conspiracy theories are general features of human society. For instance, belief that John F. Kennedy was killed in a conspiracy were already widespread by the late 1960s and 1970s (Bowman and Rugg 2013). Hofstadter (1964) goes further, showing that a “paranoid style” of conspiratorial thinking recurs in American political culture going back to the country’s founding. Moreover, exposure to the sorts of untrustworthy websites that are often called “fake news” was actually quite limited for most Americans during the 2016 campaign—far less than media accounts suggest (Guess, Nyhan, and Reifler 2020). In general, no systematic evidence exists to demonstrate that the prevalence of misperceptions today (while worrisome) is worse than in the past.
Or as I sometimes say, perhaps the reason for disagreement isn't that the other side has been gulled and deceived, and if they just learned the real true facts then they would agree with you. Maybe the most common reason for disagreement is that people actually disagree.