Macro Roundup

Ezra Klein, "David Shor Is Telling Democrats What They Don’t Want to Hear,"New York Times, October 8, 2021, https://www.nytimes.com/2021/10/08/opinion/democrats-david-shor-education-polarization.html

David Shor Is Telling Democrats What They Don’t Want to Hear

President Biden’s agenda is in peril. Democrats hold a bare 50 seats in the Senate, which gives any member of their caucus the power to block anything he or she chooses, at least in the absence of Republican support. And Senators Joe Manchin and Kyrsten Sinema are wielding that leverage ruthlessly.

But here’s the truly frightening thought for frustrated Democrats: This might be the high-water mark of power they’ll have for the next decade.

Democrats are on the precipice of an era without any hope of a governing majority. The coming year, while they still control the House, the Senate and the White House, is their last, best chance to alter course. To pass a package of democracy reforms that makes voting fairer and easier. To offer statehood to Puerto Rico and Washington, D.C. To overhaul how the party talks and acts and thinks to win back the working-class voters — white and nonwhite — who have left them behind the electoral eight ball. If they fail, they will not get another chance. Not anytime soon.

That, at least, is what David Shor thinks. Shor started modeling elections in 2008, when he was a 16-year-old blogger, and he proved good at it. By 2012, he was deep inside President Barack Obama’s re-election campaign, putting together the fabled “Golden Report,” which modeled the election daily. The forecast proved spookily accurate: It ultimately predicted every swing state but Ohio within a percentage point and called the national popular vote within one-tenth of a percentage point. Math-geek data analysts became a hot item for Democratic Party campaigns, and Shor was one of the field’s young stars, pioneering ways to survey huge numbers of Americans and experimentally test their reactions to messages and ads.

But it was a tweet that changed his career. During the protests after the killing of George Floyd, Shor, who had few followers at the time, tweeted, “Post-MLK-assassination race riots reduced Democratic vote share in surrounding counties by 2 percent, which was enough to tip the 1968 election to Nixon.” Nonviolent protests, he noted, tended to help Democrats electorally. The numbers came from Omar Wasow, a political scientist who now teaches at Pomona College. But online activists responded with fury to Shor’s interjection of electoral strategy into a moment of grief and rage, and he was summarily fired by his employer, Civis Analytics, a progressive data science firm.

For Shor, cancellation, traumatic though it was, turned him into a star. His personal story became proof of his political theory: The Democratic Party was trapped in an echo chamber of Twitter activists and woke staff members. It had lost touch with the working-class voters of all races that it needs to win elections, and even progressive institutions dedicated to data analysis were refusing to face the hard facts of public opinion and electoral geography.

Freed from a job that didn’t let him speak his mind, Shor was resurrected as the Democratic data guru who refused to soften an analysis the left often didn’t want to hear. He became ubiquitous on podcasts and Twitter, where Obama posts his analyses and pundits half-jokingly refer to themselves as being “Shor-pilled.” Politico reported that Shor has “an audience in the White House and is one of the most in-demand data analysts in the country,” calling his following “the cult of Shor.” Now he is a co-founder of and the head of data science at Blue Rose Research, a progressive data science operation. “Obviously, in retrospect,” he told me, “it was positive for my career.”

At the heart of Shor’s frenzied work is the fear that Democrats are sleepwalking into catastrophe. Since 2019, he’s been building something he calls “the power simulator.” It’s a model that predicts every House and Senate and presidential race between now and 2032 to try to map out the likeliest future for American politics. He’s been obsessively running and refining these simulations over the past two years. And they keep telling him the same thing.

We’re screwed in the Senate, he said. Only he didn’t say “screwed.”

In 2022, if Senate Democrats buck history and beat Republicans by four percentage points in the midterms, which would be a startling performance, they have about a 50-50 chance of holding the majority. If they win only 51 percent of the vote, they’ll likely lose a seat — and the Senate.

But it’s 2024 when Shor’s projected Senate Götterdämmerung really strikes. To see how bad the map is for Democrats, think back to 2018, when anti-Trump fury drove record turnout and handed the House gavel back to Nancy Pelosi. Senate Democrats saw the same huge surge of voters. Nationally, they won about 18 million more votes than Senate Republicans — and they still lost two seats. If 2024 is simply a normal year, in which Democrats win 51 percent of the two-party vote, Shor’s model projects a seven-seat loss, compared with where they are now.

Sit with that. Senate Democrats could win 51 percent of the two-party vote in the next two elections and end up with only 43 seats in the Senate. You can see Shor’s work below. We’ve built a version of his model, in which you can change the assumptions and see how they affect Democrats’ projected Senate chances in 2022 and 2024.

The “Physics” of Elections

Projection is an uncertain exercise, but that doesn’t make it useless. There is, as Shor puts it, a certain “physics” to elections. How a state votes in presidential elections is largely how it votes in midterm elections. Partisanship and demographics are uncomfortably revealing and don’t change much from year to year. None of this is inevitable or unalterable in the face of campaigns or catastrophe. But it’s somewhat predictable, and attempting a prediction can force a confrontation with reality that would otherwise go ignored until it’s too late.

This is the confrontation Shor is trying to force. The Senate’s design has long disadvantaged Democrats. That’s in part because the Senate overweights rural states and Democrats are a disproportionately urban coalition and in part because Republicans, in a bid for political advantage, added a flurry of states in 1889 and 1890 — North Dakota, South Dakota, Montana, Washington, Idaho and Wyoming — many of which largely vote Republican to this day. But that’s been true for decades, and Democrats have held their own in the Senate. What’s changed the equation, Shor believes, are several interlocking forces.

First, educational polarization has risen sharply in recent years, particularly among white voters. Democrats are winning more college-educated white voters and fewer non-college white voters, as pollster shorthand puts it, and Donald Trump supercharged this trend. There was a time when Democrats told themselves that this was a byproduct of becoming a more diverse party, as non-college white voters tend to be more racially reactionary. Then, in 2020, Democrats lost ground among Black and Latino voters, with the sharpest drops coming among non-college voters.

I want to stop here and say I believe, as does Shor, that educational polarization is serving here as a crude measure of class polarization. We tend to think of class as driven by income, but in terms of how it’s formed and practiced in America right now, education tracks facets that paychecks miss. A high school dropout who owns a successful pest extermination company in the Houston exurbs might have an income that looks a lot like a software engineer’s at Google, while an adjunct professor’s will look more like an apprentice plumber’s. But in terms of class experience — who they know, what they believe, where they’ve lived, what they watch, who they marry and how they vote, act and protest — the software engineer is more like the adjunct professor.

Either way, the sorting that educational polarization is picking up, inexact as the term may be, puts Democrats at a particular disadvantage in the Senate, as college-educated voters cluster in and around cities while non-college voters are heavily rural. This is why Shor believes Trump was good for the Republican Party, despite its losing the popular vote in 2016, the House in 2018 and the Senate and the presidency in 2020. “Sure, maybe he underperforms the generic Republican by whatever,” Shor said. “But he’s engineered a real and perhaps persistent bias in the Electoral College, and then when you get to the Senate, it’s so much worse.” As he put it, “Donald Trump enabled Republicans to win with a minority of the vote.”

The second problem Democrats face is the sharp decline in ticket splitting — a byproduct of the nationalization of politics. As recently as 2008, the correlation between how a state voted for president and how it voted in Senate elections was about 71 percent. Close, but plenty of room for candidates to outperform their party. In 2020, it was 95.6 percent.

The days when, say, North Dakota’s Republicans would cheerfully vote for a Democrat for the Senate are long past. Just ask Heidi Heitkamp, the defeated North Dakota Democrat who’s now lobbying her former colleagues to protect the rich from paying higher taxes on inheritances. There remain exceptions to this rule — Joe Manchin being the most prominent — but they loom so large in politics because they are now so rare. From 1960 to 1990, about half of senators represented a state that voted for the other party’s nominee for president, the political scientist Lee Drutman noted. Today, there are six.

Put it all together, and the problem Democrats face is this: Educational polarization has made the Senate even more biased against Democrats than it was, and the decline in ticket splitting has made it harder for individual Democratic candidates to run ahead of their party.

Atop this analysis, Shor has built an increasingly influential theory of what the Democrats must do to avoid congressional calamity. The chain of logic is this: Democrats are on the edge of an electoral abyss. To avoid it, they need to win states that lean Republican. To do that, they need to internalize that they are not like and do not understand the voters they need to win over. Swing voters in these states are not liberals, are not woke and do not see the world in the way that the people who staff and donate to Democratic campaigns do.

All this comes down to a simple prescription: Democrats should do a lot of polling to figure out which of their views are popular and which are not popular, and then they should talk about the popular stuff and shut up about the unpopular stuff. “Traditional diversity and inclusion is super important, but polling is one of the only tools we have to step outside of ourselves and see what the median voter actually thinks,” Shor said. This theory is often short-handed as “popularism.” It doesn’t sound as if it would be particularly controversial.

It is.

Popularism, Explained and Questioned

Shor’s theory of popularism, at its heart, is a critique of the professional staffers, consultants and organizers who shape the Democratic Party’s message, image and strategic choices.

“I think the core problem with the Democratic Party is that the people who run and staff the Democratic Party are much more educated and ideologically liberal and they live in cities, and ultimately our candidate pool reflects that,” he said.

Nor is Shor’s ire aimed only at the liberal wing of the party. Popularism isn’t mere moderation. One of the highest-polling policies in Shor’s research is letting Medicare negotiate prescription drug prices, but it’s so-called moderates, like Sinema, who are trying to strike that from the reconciliation bill. To Shor, this is lunacy.

Shor believes the party has become too unrepresentative at its elite levels to continue being representative at the mass level. “I don’t think it’s a coincidence that the people we’ve lost are likely to be low-socioeconomic-status people,” he said. “If you look inside the Democratic Party, there are three times more moderate or conservative nonwhite people than very liberal white people, but very liberal white people are infinitely more represented. That’s morally bad, but it also means eventually they’ll leave.” The only way out of this, he said, is to “care more and cater to the preference of our low-socioeconomic-status supporters.”

The Democratic strategists and analysts who Shor said are causing the party’s problems seethe at his criticism and the influence he has commanded over the past few years. Among them, a few counterarguments dominate.

The first is that Shor doesn’t really show his work. There’s no comprehensive paper or experiment in which he has constructed and footnoted a full theory, in which his data can be rerun and his footnotes picked through. He sometimes refers to polling he conducted but doesn’t release the underlying numbers and cross-tabs. To be fair, that’s often because he can’t: He conducts much of his polling on behalf of clients, and they own the results. But it frustrates those trying to assess the arguments he makes publicly.

“In the data world, if you take Shor on, you face intense backlash now,” said Michael Podhorzer, the longtime political director of the A.F.L.-C.I.O., who’s something of a godfather in Democratic data circles. “You’re seen as less rigorous or pleading a woke case. I’m in an unusual space: I’m an older white man with access to a lot of the data, so I can say it. I feel like he’s found this weird sweet spot with the media where he never actually shows anyone the evidence for his claims. He just does interviews with reporters.”

This is somewhat unfair. Shor’s tweets and even his comments are thick with citations to political science papers and regression tables. Compared to most pundits, he is amply footnoted. But it’s true that compared to other data analysts, he’s not. Speaking mainly through tweets and interviews lets him sidestep some of the standards that others in his profession are held to. In their view, Shor has cloaked himself in the aesthetics of data, but he’s not doing the rigorous, reviewable work demanded of others in the field. Some of his most influential theories are plausible, but he has never fully laid out the evidence needed to prove them.

“In the summer, following the emergence of ‘defund the police’ as a nationally salient issue, support for Biden among Hispanic voters declined,” Shor said in a March interview with New York magazine. “So I think you can tell this microstory: We raised the salience of an ideologically charged issue that millions of nonwhite voters disagreed with us on. And then, as a result, these conservative Hispanic voters who’d been voting for us despite their ideological inclinations started voting more like conservative whites.”

It’s a striking argument, and it fits Shor’s broader theory of the case: Liberal Democrats were either backing or cowering before a politically toxic slogan that had taken over Twitter but was alienating them from their working-class supporters. And even though Biden publicly and repeatedly repudiated the idea, it hurt him anyway, because voters don’t distinguish between different Democrats anymore.

In the same interview, Shor said he based this theory on “extensive postelection surveys of 2020 voters” he conducted with partner organizations. He told me he couldn’t release the underlying numbers because they belonged to another group, but he sent me a table that showed the relationships between various issue positions and whether Latinos shifted their vote between 2016 and 2020, and it indicated that views on defunding the police were the strongest driver.

Other analysts, however, came to very different conclusions using more visible data sets. Robert Griffin, a research director at Democracy Fund, and Natalie Jackson, the research director of the Public Religion Research Institute, both tweeted that their polling data didn’t show Latino voters moving to Trump as a result of the Floyd protests. But it’s possible, as Shor noted in the same thread, that those polls could have had the same flaws that biased other polls toward Biden.

More work was done after the election to try to sort this out. EquisLabs produced a huge study of Latinos in the 2020 elections, conducting over 40,000 interviews with voters across 12 states. It found that Democratic policies did alienate working-class voters but that it wasn’t “defund the police” that did it. “For many who had jobs, there was a calculation to not rock the boat, a fear Biden would come in and shut down the economy,” Carlos Odio, EquisLabs’ senior vice president, told me. “That’s the baseline shift.”

EquisLabs’ research found support for other theories, too, including that some Latino voters worried that Democrats would be too soft on border security and that others feared socialism. Odio also believes that because neither campaign emphasized immigration in 2020, conservative Latinos who were repelled by Trump’s xenophobic rhetoric in 2016 felt able to vote for him in 2020. “What doesn’t come through is ‘defund the police,’” he said. “That feels like part of the elite discourse criticizing another part of the elite discourse. That was not part of the conversation happening at kitchen tables, when it mattered.”

There are other data points supporting Shor’s views. He pointed to a regression analysis by Alexander Agadjanian, a political science Ph.D. student, that used public data to show that pro-police views were unusually potent in increasing the probability that a voter would switch to Trump, though somewhat less so for Latinos than for white voters. The problem with all of this regression data, though, is that voters who switched to Trump in 2020 might have adopted his views on policing rather than switched because of his views on policing.

Having spent a lot of time trying to untangle this debate, I’d say it left me sympathetic to those who wish Shor would release more of his data and make these arguments in thicker formats. “I agree with David that ‘defund the police’ is an unfortunate slogan in a number of ways,” said Sean McElwee, a co-founder of Data for Progress and a frequent collaborator of Shor’s. “I’m a little skeptical that it was particularly devastating.” Again, the argument isn’t that Shor is wrong that “defund the police” hurt Democrats but that he hasn’t done the work to prove that he’s right. “There was never a comprehensive David Shor putting out a report showing that ‘defund the police’ cost us,” McElwee told me.

The second level of disagreement is more fundamental: Many in the Democratic data world simply disagree that policy communication holds the power Shor believes it does or that the popularity of a message is as important as he thinks it is.

“There’s no argument that saying unpopular things is better than saying popular things. My argument is it’s not close to being an important enough factor to warrant attention,” Podhorzer told me. “If the object is for Democrats to win, that’s a tertiary, at best, factor.”

The suspicion here is that Shor has come up with a class-polarized way of responding to class polarization. He’s a smart, wonky nerd who thinks about politics in terms of polling and policy, and maybe he’s projecting that onto the electorate, too. According to this line of thinking, even as he’s trying to escape his ideological biases about what voters believe, he’s replicating his biases as to how they think and act.

“It’s almost laughable to me the notion that what people think about Democrats is made out of what Democrats say,” said Anat Shenker-Osorio, the founder of the progressive firm ASO Communications and a principal on the Race-Class Narrative Project. “I wish we lived in that world. I’d probably be on vacation. But that’s not our world.”

Our world, Shenker-Osorio argued, is one in which the voters Democrats most need to reach are the ones paying the least attention. What they hear comes at the end of a long game of telephone, and they’re only half-listening even then, as their kids are yelling and the bill collectors keep calling. If you start with that model of the electorate, you end up with different recommendations. “A message is like a baton. It needs to be handed from person to person to person,” she said. “If it gets dropped, it’s not persuasive. Unless you’re testing for what the base — what I think of as the choir — is willing to sing, then you’re going to be hard-pressed to get the middle to hear that song, to get the congregation to hear that song.”

Shor’s critics argue that he’s too focused on the popularity of what Democrats say, rather than the enthusiasm it can unleash. When pressed, Podhorzer called this theory “viralism” and pointed to Trump as an example of what it can see that popularism cannot. “A lot of things Trump did were grossly unpopular but got him enormous turnout and support from the evangelical community,” Podhorzer said. “Polling is blind to that. Politics isn’t just saying a thing at people who’re evaluating it rationally. It’s about creating energy. Policy positions don’t create energy.”

Podhorzer also pointed to Biden: “He’s done much more than I thought he’d be able to do. All the things he’s doing are popular. And yet he’s underwater.”

What does create energy, Podhorzer thinks, is fear of the other side. His view is that Democrats’ best chance, even now, is to mobilize their base against Trump and everything he represents. “The challenge in 2022 is to convince people that they’re again voting on whether or not the country is going in a Trumpist direction,” he said.

This is an argument Shor is happy to have. “I think the conventional wisdom has swung too far toward believing policy isn’t important,” he said. He agrees that enthusiasm matters, but it has to be enthusiasm for a message that doesn’t alienate the undecided. “A lot of politics is about what you talk about,” he told me. You should sort your ideas, he said, by popularity. “Start at the top, and work your way down to find something that excites people. But I think that what actually happens is people sort by excitement first. And the problem is the things that are most exciting to activists and journalists are politically toxic.”

Shor showed me, as an example, a set of environmental talking points he’d tested, in which the ones that mentioned climate change performed worst. “Very liberal white people care way more about climate change than anyone else,” he said. “So when you talk about climate change, you sound like a weird, very liberal white person. This is why policy issues matter more than people realize. It’s not that voters have these very specific policy preferences. It’s that the policies you choose to talk about paints a picture of what kind of person you are.”

I should say that the polling differences here struck me as modest: The best environmental message on Shor’s list increased Biden’s approval rating by 1.7 percentage points, while the worst-performing message cut it by 0.4 points. On the other hand, a percentage point here, a percentage point there can be the difference between winning the White House and losing it.

Shor’s example speaks to the hardest questions raised by popularism. “Talk about your most popular, most energizing ideas” isn’t controversial advice. The real disagreements come on the ideas that don’t poll so well. There are a lot of issues that Democrats want to talk about that Shor thinks they’d be better off not talking about.

Hillary Clinton “lost because she raised the salience of immigration, when lots of voters in the Midwest disagreed with us on immigration,” Shor said. This is where popularism poses its most bitter choices: He and those who agree with him argue that Democrats need to try to avoid talking about race and immigration. He often brandishes a table showing that among voters who supported universal health care but opposed amnesty for unauthorized immigrants, 60 percent voted for Obama in 2012 but 41 percent voted for Clinton in 2016. That difference, he noted, was more than enough to cost her the election.

This can read as an affront to those who want to use politics to change Americans’ positions on those issues. “The job of a good message isn’t to say what’s popular but to make popular what needs to be said,” Shenker-Osorio told me.

Shor’s rejoinder to this is that the best way to make progress on race and immigration policy is for Democrats to win elections. Obama’s twin victories loom large in his thinking here, since he watched Obama’s brain trust carefully decide what to avoid and the result was the election and re-election of the country’s first Black president, to say nothing of all the policies he passed.

Shor is right about how the Obama campaign understood the electorate. David Simas, the director of opinion research on Obama’s 2012 campaign, recalled a focus group of non-college, undecided white women on immigration. It was a 90-minute discussion, and the Obama campaign made all its best arguments. Then they went around the table. Just hearing about the issue pushed the women toward Mitt Romney. The same process then played out in reverse with shipping jobs overseas. Even when all of Romney’s best arguments were made, the issue itself pushed the women toward Obama. The lesson the Obama team took from that was simple: Don’t talk about immigration.

“You don’t have the luxury of just sending one mobilization message that isn’t going to be heard by a whole bunch of persuadable voters,” Simas told me. “So if we make immigration the central part of a message in Wisconsin, what’s that going to do to the massive amount of non-college whites who’re much more concerned about bread-and-butter economic issues?”

This is the kind of thinking Shor thinks Democrats have largely lost. “Obama and his messaging team were very calculated and measured about that,” he said. “That’s the piece we dropped. I think it’s great to push the envelope and be ahead of history. But you want to be five years ahead of history, not 15 years.”

But one difference between 2016 and 2012 is that Romney was complicit in making economics the center of the campaign. Like Obama, he preferred to argue over tax policy and spending cuts and was plainly uncomfortable talking about immigration or race. He ran, self-consciously, as a former management consultant who would govern on behalf of America’s makers rather than its takers. Trump descended a golden escalator to call Mexican immigrants criminals and rapists. What was Clinton supposed to do?

The implication of popularism is that Clinton shouldn’t have heavily engaged Trump on immigration and race, no matter the provocations. Instead, she should have stuck to a higher-polling economic message. Shor’s critics think that theory is, to put it gently, impractical. The media focuses on the points of controversy between the candidates, and Trump relentlessly weaponized the energy contained in America’s deepest divisions. Clinton talked far more about jobs and the economy than about anything else on the campaign trail, but the comments that generated the most media attention and popular energy were the ones that engaged Trump’s attacks.

But even if Clinton could have sustained Shor-level message discipline, would it have worked, or would the perception that Clinton wasn’t standing up for her voters or their ideals have left large swaths of the Democrats’ base demoralized?

“Look, he’s right about a class and cultural divide,” Odio said. “He’s right about a liberal establishment that’s out of touch with working-class voters. He’s right that Latino and Black voters used to be insulated from polarization and now aren’t. But where he falls short is in investigating why that is. He’s really missing a race and ethnicity lens. If you fail to incorporate group identity into the analysis, you really miss why Black voters have been voting at astronomically high levels for Democrats. Why have Latinos, who are more moderate and even conservative in his analysis, been voting for Democrats? There’s a group threat that factors into their analysis. If you only talk to Latinos about immigration, you lose voters on the table. If you only talk to them about economics, you’ll arguably leave more votes on the table.”

But if there’s a narrowness to Shor’s focus, there can be a dissonance in the arguments of his critics. On the one hand, they frame this moment in politics as existential, an era in which democracy itself is teetering on the edge of calamity. And in the next breath, they treat message discipline, of any sort, as an impossible and perhaps even useless ask to make of the Democratic Party. At times, their arguments carry an air of resignation.

“I don’t think there’s a short-term solution to the predicament we’re in,” Podhorzer said. “There’s not a set of things Democrats can say that will make them popular to the extent they can start winning the Senate. I don’t think it exists.”

In a way, this is where Shor and his critics converge: They are both deeply pessimistic about the near-term chances for Democrats and thus for democracy.

What Democrats Need to Understand

Models can mislead. The demographic triumphalism that Democrats felt a decade ago has vanished, as reality proved more complicated than regressions. The same may be true here, too. McElwee, for one, thinks these disasters are being projected “with more certainty than is warranted.” He noted that the Democrats’ new coalition may put them at a disadvantage in the Senate but college-educated voters are more likely to turn out in 2022. “Educational polarization could be a stabilizing force for Democrats in midterms,” he said. “I think there’s reason to believe, looking at Georgia and Nevada and California, that we now have a coalition that’s much more robust in midterms.”

Trump may also prove unique in his ability to polarize the electorate along class lines. If he doesn’t run again in 2024, will a Ron DeSantis or a Mike Pence really be able to generate the fury and fervor that Trump did in 2016 and 2020? His successors might polarize the electorate somewhat differently, just as Romney and John McCain did before him.

But no matter who Republicans nominate for president, Democrats face a terribly uphill battle in the Senate, and they don’t seem to have a plan for what to do about it. If the stakes are as dire as they appear to be and Republicans are as dangerous as Democrats say, Shor is right that they need one. Now. And any such plan will require compromises and discipline that many Democrats will loathe.

“When I first started working on the Obama campaign in 2012, I hated all the last remnants of the Clinton era,” Shor said. “When I go back now and think about the fights between the analytics team and the consultants, about 80 percent of the time, they were right. There was an old conventional wisdom to politics in the ’90s and 2000s that we all forget. We collectively unlearned those lessons over the past 12 years. We’ve told ourselves very ideologically convenient stories about how those lessons weren’t relevant — that tax phobia isn’t real or we didn’t need to worry about what conservative white people thought. And it turned out that wasn’t true. I see what I’m doing as rediscovering the ancient political wisdom of the past.”

Sometimes, when I report on a debate, I emerge with a strong view on who’s wrong. In this case, I think both sides are right. Democrats are often trapped in an echo chamber of their own making — a problem Twitter has made immeasurably worse — and they are too quick to dismiss evidence that their ideas and messages are alienating voters. The political system is stacked against them, and unless they are going to change it by adding states and reforming election laws, they need to campaign with the constant recognition that the pivotal voter is well to their right and skeptical of everything they say. On all of that, Shor is offering a warning Democrats should heed.

At the same time, I think he overstates the power of policy communication and the control Democrats have over the debates that will dominate politics. There is little Biden can do to stop Sinema from making a hash of his agenda and muddling his message, and Democrats can’t, in reality, avoid talking about race and immigration and climate change, for reasons both practical and moral. Politics is also about changing what’s possible tomorrow.

I think Shor overreads the experience he had on the Obama campaign: It’s precisely because Obama was a thrilling, historic figure that he could tailor his message so carefully. Unless Democrats can conjure up a generational political talent for every election, they’ll often have to mobilize their base in ways that might unnerve the uncertain or fight on ground that the other side has chosen. But that’s precisely when a bit more of a Shor-esque obsession with polls and skeptical voters might help them most.

To a debate full of inelegant coinages — “popularism,” “viralism” — let me, with apologies, add one more: partyism. The core problem Democrats face is that almost all politics is now national. They are one party facing electoral disaster, and they will rise or fall together. Democrats cannot escape one another, no matter how they might try.

This, to me, is the most important part of Shor’s argument: He is right to insist that the Democratic Party is an institution that is composed, at the top, of a narrow group of people and that is afflicted by many of their blind spots. Whether he is right about what those blind spots are or his critics are right that he is adding some of his own is a secondary concern. For the Democratic Party to chart any course out of the peril it faces, it must first accept that in the minds of most Americans, it is a party, a singular entity. And before that party can shape what voters think, it must find a way to see itself clearly and act collectively.

Scott Winship on Bruce Meyer's Research (attached), "...But in an explosive new working paper from the University of Chicago’s Becker Friedman Institute, Kevin Corinth, Bruce Meyer, Matthew Stadnicki, and Derek Wu (henceforth, CMSW) demonstrate that the behavioral effects of the expanded CTC are a much bigger issue than previously documented. They do so using new, improved data and correcting inconsistencies in the modeling conducted by the NAS committee. CMSW find that the work disincentives created by the expanded CTC are larger than previous researchers have claimed — perhaps sizable enough to reverse the employment gains caused by welfare reform and the EITC in the 1990s — and primarily impact single parent families.As a result, existing studies have overstated the short-term poverty-reduction impact of the policy by a third. Notably, the NAS committee failed to model the primary behavioral response to the new CTC that would be expected to reduce employment. In contrast, the committee did include such a response in its modeling of an Earned Income Tax Credit (EITC) expansion, when the financial incentives involved would be expected to increase employment....But by demonstrating the importance of short-term work disincentives (and the blind spot that many researchers have regarding behavioral effects), CMSW have strengthened the case that child allowances might have precisely the unintended consequences that conservative critics fear...."

Scott Winship, "New evidence on the benefits and costs of an expanded child tax credit," American Enterprise Institute, October 7, 2021, https://www.aei.org/poverty-studies/new-evidence-on-the-benefits-and-costs-of-an-expanded-child-tax-credit/

New evidence on the benefits and costs of an expanded child tax credit

Good intentions cannot be the measure of a policy’s merit. The case for expanding the child tax credit (CTC) has always been disarmingly simple: Giving parents money will reduce hardship among children. The American Rescue Plan Act (ARPA) temporarily made the CTC more generous for all but the richest families — especially to those with little to no earnings. Unlike the old CTC, the newly expanded version — which Democrats are trying to extend via the budget reconciliation bill — provides the full credit amount to families even if they owe no income tax or have no earnings. Estimates of the effect of the expansion have ranged from a drop in poverty of one-third to 46 percent. These estimates simply add the new CTC benefits to families’ incomes and re-calculate whether they are poor. Unsurprisingly, when poverty is defined as having too little income, giving more income to more families means that fewer fall below the poverty threshold.

Critics of the CTC expansion have offered a number of objections, including its $1.6 trillion cost over a decade.

An important criticism stems from concerns that the larger CTC will lead some parents to behave in ways that will counter the immediate (mechanical) poverty-reducing effect of the benefits. For example, if some parents react to having more purchasing power by working fewer hours, leaving the workforce, or not entering it in the first place, that will negate (at least partly) the short-term effects on poverty of the more generous CTC.

Advocates of the CTC expansion have downplayed the importance of such behavioral effects, pointing to (among other studies) the work of a recent National Academies of Sciences, Engineering, and Medicine (NAS) committee examining strategies to reduce child poverty. But in an explosive new working paper from the University of Chicago’s Becker Friedman Institute, Kevin Corinth, Bruce Meyer, Matthew Stadnicki, and Derek Wu (henceforth, CMSW) demonstrate that the behavioral effects of the expanded CTC are a much bigger issue than previously documented. They do so using new, improved data and correcting inconsistencies in the modeling conducted by the NAS committee.

CMSW find that the work disincentives created by the expanded CTC are larger than previous researchers have claimed — perhaps sizable enough to reverse the employment gains caused by welfare reform and the EITC in the 1990s — and primarily impact single parent families. As a result, existing studies have overstated the short-term poverty-reduction impact of the policy by a third. Notably, the NAS committee failed to model the primary behavioral response to the new CTC that would be expected to reduce employment. In contrast, the committee did include such a response in its modeling of an Earned Income Tax Credit (EITC) expansion, when the financial incentives involved would be expected to increase employment.

The CMSW paper does not model all of the potential short-term work disincentives embedded in the new CTC, nor does it model short-term incentives that would be expected to increase the share of children living with single parents, nor any long-term incentives on work, living arrangements, marriage, or fertility that might be expected to work against poverty reduction even more. It does not examine the potentially negative impact of the expanded CTC on other outcomes, such as intergenerational mobility. But by demonstrating the importance of short-term work disincentives (and the blind spot that many researchers have regarding behavioral effects), CMSW have strengthened the case that child allowances might have precisely the unintended consequences that conservative critics fear.

Policy Incentives, Work, and Poverty

CMSW rely on a restricted-use dataset that combines data from the widely used Current Population Survey with administrative data on earnings, retirement income, federal benefit programs, and taxes. The merger of the two datasets allows the authors to correct for substantial underreporting of income.

The Chicago team first analyzes the effect of the expanded CTC using the same conventions as previous researchers, ignoring potential behavioral responses. It finds that the expanded CTC reduces child poverty by 34 percent. In fact, the expansion of the CTC — the new part alone — reduces child poverty more than any other existing antipoverty program.

CMSW then consider how the work-reducing incentives embedded in the CTC expansion might change that conclusion. Specifically, they consider how those incentives might cause some adults who would have been employed without the expanded CTC to decide not to work. CMSW find that with an expanded CTC, the return to working falls by about 7 percent among families earning under $10,000, by about 10 percent among families earning between $10,000 and $40,000, and by between 7 and 9 percent among those earning between $40,000 and $60,000.

To illustrate how sizable this reduction in the return to work is, CMSW compare it with what would happen if the EITC were eliminated. The EITC is one of the largest antipoverty programs, and it is explicitly designed to incentivize work. For families with under $10,000, eliminating the EITC would reduce the payoff to working by nearly 30 percent. Thus, the work disincentive from expanding the CTC is about one-fourth as large for these earners as the work disincentive that would be created by eliminating the EITC.

However, there are nearly twice as many earners in families making $30,000 to $40,000 as there are in families making less than $10,000. For them, the expanded CTC reduces the incentive to work by more than eliminating the EITC would.

CMSW take the percent changes in the return to work for families in their data and multiply them by labor supply elasticities drawn from the literature, which give the percentage decline in employment for a 1 percent change in the return to work. They also multiply the percent changes in income represented by the expanded CTC for families in their data by income elasticities from the literature, which provide the percentage decline in employment for a 1 percent increase in income. (Additional income affords more people the ability not to work.) Together, and applied across families, these calculations yield an estimate of the decline in employment due to the CTC expansion, through its reduction of the return to work and its increase in income. Finally, CMSW model the change in the child poverty rate resulting from not only the more generous CTC benefits enacted in ARPA, but the reduced earnings that those benefits produce.

Incorporating the work-reducing incentives embedded in the CTC expansion, CMSW find that child poverty is reduced not by 34 percent, but by just 22 percent. More dramatically, while conventional estimates suggest the expanded CTC would lower deep child poverty — being under half the poverty line — by 39 percent, with behavioral effects modeled there is no impact whatsoever on deep child poverty. (The authors suggest that this probably understates somewhat the true impact on deep child poverty due to the way they implement their modeling.)

Revisiting the National Academies Study

The impact on poverty that CMSW report is lower than that found by the widely-cited NAS committee, which modeled the impact on child poverty of an expanded CTC similar to the one that was included in ARPA. The committee included behavioral effects on work in their modeling and found that an expanded CTC would lower child poverty by 41 percent.

In large part, the NAS estimate was higher than CMSW’s because the committee found that just 149,000 people would be induced to stop working by an expanded CTC. CMSW find that the work-reducing effects of the expanded CTC are ten times as large — 1.5 million would leave the workforce. What accounts for the difference?

The NAS committee modeled two “income effects” on labor supply — two ways that the increased income provided by an expanded CTC might cause people to work less. It modeled an income effect on the participation margin (the decision to work or not work) and an income effect on the hours margin (whether to work more or less, conditional on working at all). Its modeling assumed relatively small income effects. The Chicago researchers model effects very similar to NAS’s on the participation margin, as their assumptions are explicitly informed by the NAS study.

However, the NAS committee did not model any “substitution effects” on labor supply — changes in work due to the return to work declining. In a report from earlier in the year, I noted this issue, but I highlighted the less important hours margin. Since the old CTC phased in, it incentivized additional work by increasing the CTC amount as earnings rose until it reached the maximum of $2,000 per child. The expanded CTC is available to even non-workers, so it eliminates the pro-work phase-in, and this would be expected to reduce labor supply by lowering the return to working additional hours. CMSW do not model this effect either.

But as the Chicago researchers note, substitution effects are not confined to workers whose earnings put them on the old CTC phase-in, nor to the hours margin. The expanded CTC reduces the return to work across nearly the entire income distribution and therefore affects participation decisions across the distribution. This effect is the focus of the CMSW study. The NAS committee’s omission of any substitution effect on the participation margin in its CTC modeling has more serious consequences than ignoring substitution along the hours margin.

The omission is also curious because the committee’s modeling of an EITC expansion did include such substitution effects. The primary difference between the two policy modeling cases is that the substitution response when expanding the EITC would be expected to increase employment and thereby reduce poverty, while the response in moving from the old phased-in CTC to a child allowance would be expected to reduce employment and thereby increase poverty.

One worries that the NAS committee — even if subconsciously — might have been “rooting” for a more favorable CTC modeling result. At least three of the 15 members of the committee wrote in support of expanding the CTC to resemble a child allowance prior to the NAS report being published. Another committee member recently co-led the development and circulation of an open letter to policymakers for economists to sign supporting a permanent extension of the new CTC. Of the six economists on the NAS committee, five signed the letter. Ironically, that letter cited the NAS committee’s finding that the expansion would not reduce work significantly, which may have contributed to their subsequent advocacy. The modeling oversight uncovered by CMSW should remind us of the importance of including diverse perspectives on government panels examining important policy questions.

Reversing Welfare Reform

The CMSW study should give policymakers pause as they contemplate extending the newly expanded CTC. It identifies significant short-term effects on employment, even without looking at how the expanded CTC might cause some workers to reduce their hours while still remaining employed. Nor does it look at the effect on non-workers’ likelihood of choosing to enter work. So it only partly gets at the labor supply response to an expanded CTC.

The negative effects on work that the study does estimate would be heavily concentrated among low-earning families, who are disproportionately headed by a single parent. Roughly half of the families leaving employment in the CMSW modeling would come from families earning under $30,000. Over 80 percent of families leaving employment in their model switch from one worker to no workers (rather than from two workers to no workers — their model does not allow for two-worker-to-one-worker switches). In other words, the paper’s results are not driven by married mothers (or fathers) leaving the workforce to have a single-breadwinner family.

Neither does the paper look at the response to the expanded CTC in terms of marriage, living arrangements, and fertility. In the same way that safety net benefits make it possible to work less, additional income from the CTC makes it more feasible to raise children outside of marriage. In response to more income and declining returns to having earnings, some parents (disproportionately mothers) will choose to leave their marriage and take their children with them, while some (disproportionately fathers) will choose to leave and to leave the parenting to their ex. Other couples will find out-of-wedlock childbearing marginally easier: Either the mother, the father, or both may feel the mother has less need of the father’s support. More couples will find themselves with nonmarital, often unplanned, pregnancies and choose not to marry in the first place.

Finally, the CMSW paper is unable to consider whether the long-term effects of the policy might be understated by these short-term effects. Most narrowly, the longer-term effects on work might be understated. In the negative income tax experiments of the 1970s, one group of single mothers was assigned to receive a UBI-like benefit (but taxed away as earnings rose) for 20 years. After receiving benefits for three years, the decline in hours worked was twice as large as it was after receiving benefits for one year. (The experiment prematurely ended soon thereafter.) Thus, the long-term impact on child poverty of an expanded CTC might be significantly smaller than CMSW estimate.

More broadly, an expanded CTC might have other negative consequences over the long-run. As I wrote in my March report: "Child allowances run a very real risk of encouraging more single parenthood and more no-worker families, both of which could worsen entrenched poverty in the long run — an overreliance on government transfers, poverty over longer stretches of childhood, intergenerational poverty, and geographically concentrated poverty. And the concern is about not only material poverty but also the social poverty that comes from growing up in non-intact families or communities with limited social capital and a dearth of meaningful roles for members to fill."

Advocates of the CTC expansion point to research suggesting that safety net expansions have improved various long-term outcomes. This research is often non-experimental and ill-equipped to rigorously answer the policy questions of interest. It often comes from older expansions of a relatively threadbare safety net, calling into question its relevance for new safety net expansions when child poverty is at an all-time low. It is also often ambiguous in another way; if transfer income improves outcomes, it stands to reason that income from employment or from marrying someone with earnings might as well. Indeed, many results cited by advocates come from EITC research, which involves not just additional federal benefits but additional work.

To the extent that working increases parents’ skills and social capital, it would be expected to raise their future earning potential, and thus have a bigger impact on lifetime income. Working may also provide beneficial role-modeling for children. While greater work might mean less time for parents to invest in their children, providing better child care opportunities while parents work might improve the cognitive and socioemotional environments of many children, especially those whose parents struggle to provide healthy environments. Growing up with both biological parents is also associated with a variety of improved outcomes. Perhaps it is for these and other reasons that, in the past three months alone, two studies found that welfare reform improved living standards among single-mother families and reduced the adulthood poverty of children affected by it. (Meyer was a coauthor of the former.)

In truth, almost everyone involved in anti-poverty debates has good intentions. No one favors increasing child poverty as a policy goal. We should all rely on evidence as best we can to guide our policy positions, but evidence is almost always more ambiguous than the staunchest advocates of safety net expansions believe. Ambiguity calls for caution and for the kind of experimentation that informed welfare reform. Jumping hastily into a dramatic transformation of the safety net without worrying about unintended consequences may be soft-hearted, but researchers and policymakers must take care to be hard-headed as well — because we are trying to help today’s and tomorrow’s children.

Ruy Teixeira finds that education polarization is breaking the same way in terms of Hispanic voters as white voters, "...Now Pew has released the microdata from their survey allowing researchers to dig into their data to get more detail on Hispanics and other groups. Here are some of the things I found from looking at the Hispanic validated voter data. 1. Trump’s support was higher among Hispanic working class (noncollege) voters than among the college-educated. Biden carried Hispanic college voters by a whopping 39 points (69-30) compared to just 14 points (55-41) among the Hispanic working class. 2. Hispanic Trump voters were 81 percent working class and just 19 percent college-educated.3. Within the working class, the less education Hispanic voters had, the more they supported Trump. Those with some college gave Trump 39 percent of their vote, high school graduates gave him 42 percent and high school dropouts gave him 53 percent. 4. Pew breaks income into three broad groups: lower income, middle income and upper income. Trump’s worst group by far here was upper income Hispanics where he received just 28 percent of the vote. But he got 41 percent support among middle income Hispanics and 40 percent support among lower income Hispanics. 5. Just under a third of Hispanic voters described themselves as conservative. These voters supported Trump by a lopsided 73-26. 6. Over half of Hispanic voters (53 percent) were very or somewhat confident in Trump’s ability to make good decisions about economic policy. Those who were very confident supported Trump 77-18; those who were somewhat confident supported him 56-40. 7. Trump support was highest among young Hispanic voters. Those under 30 gave him 41 percent support, those in the 30-49 year old age group gave him 38 percent; those 50-64 gave him 37 percent and those 65 and over the least at 35 percent. What lies behind these unsatisfying results for the Democrats? One possibility, as I have previously argued, is that Democrats fundamentally misunderstood the nature of this voter group and what they really care about. Hispanics were lumped in with “people of color” and were assumed to embrace the activism around racial issues that dominated so much of the political scene in 2020, particularly in the summer. This was a flawed assumption. The reality of the Hispanic population is that they are, broadly speaking, an overwhelmingly working class, economically progressive, socially moderate constituency that cares above all, about jobs, the economy and health care...."

Ruy Teixeira, "The Democrats’ Hispanic Voter Problem: More Evidence from the 2020 Pew Validated Voter Survey,"The Liberal Patriot, October 7, 2021, https://theliberalpatriot.substack.com/p/the-democrats-hispanic-voter-problem

The Democrats’ Hispanic Voter Problem: More Evidence from the 2020 Pew Validated Voter Survey

Joe Biden in 2020 characterized Donald Trump as, among other things, an unapologetic racist who particularly detested immigrants. This strand of Biden’s campaign was supposed to have special appeal to Hispanics and juice their Democratic support.

But that didn’t happen. Instead Hispanic voters went in the other direction, giving Trump after four years substantially more support than they did in 2016. According to Catalist, in 2020 Latinos had an amazingly large 16 point margin shift toward Trump. Among Latinos, Cubans did have the largest shifts toward Trump (26 points), but those of Mexican origin also had a 12 point shift and even Puerto Ricans moved toward Trump by 18 points. Moreover, Latino shifts toward Trump were widely dispersed geographically. Hispanic shifts toward Trump were not confined to Florida (28 points) and Texas (18 points) but also included states like Nevada (16 points), Pennsylvania (12 points), Arizona (10 points) and Georgia (8 points).

The national Pew 2020 validated voter data were consistent with Catalist national data on the size of the shift toward Trump, as revealed by Pew’s recent report on these data. Now Pew has released the microdata from their survey allowing researchers to dig into their data to get more detail on Hispanics and other groups. Here are some of the things I found from looking at the Hispanic validated voter data.

1. Trump’s support was higher among Hispanic working class (noncollege) voters than among the college-educated. Biden carried Hispanic college voters by a whopping 39 points (69-30) compared to just 14 points (55-41) among the Hispanic working class.

2. Hispanic Trump voters were 81 percent working class and just 19 percent college-educated.

3. Within the working class, the less education Hispanic voters had, the more they supported Trump. Those with some college gave Trump 39 percent of their vote, high school graduates gave him 42 percent and high school dropouts gave him 53 percent.

4. Pew breaks income into three broad groups: lower income, middle income and upper income. Trump’s worst group by far here was upper income Hispanics where he received just 28 percent of the vote. But he got 41 percent support among middle income Hispanics and 40 percent support among lower income Hispanics.

5. Just under a third of Hispanic voters described themselves as conservative. These voters supported Trump by a lopsided 73-26.

6. Over half of Hispanic voters (53 percent) were very or somewhat confident in Trump’s ability to make good decisions about economic policy. Those who were very confident supported Trump 77-18; those who were somewhat confident supported him 56-40.

7. Trump support was highest among young Hispanic voters. Those under 30 gave him 41 percent support, those in the 30-49 year old age group gave him 38 percent; those 50-64 gave him 37 percent and those 65 and over the least at 35 percent.

What lies behind these unsatisfying results for the Democrats? One possibility, as I have previously argued, is that Democrats fundamentally misunderstood the nature of this voter group and what they really care about. Hispanics were lumped in with “people of color” and were assumed to embrace the activism around racial issues that dominated so much of the political scene in 2020, particularly in the summer. This was a flawed assumption. The reality of the Hispanic population is that they are, broadly speaking, an overwhelmingly working class, economically progressive, socially moderate constituency that cares above all, about jobs, the economy and health care.

For example, in the post-election wave of the Democracy Fund Voter Study Group (VSG) panel survey, well over 70 percent of Hispanic voters rated jobs, the economy, health care and the coronavirus as issues that were “very important” to them. No other issues even came close to this level. Crime as an issue rated higher with these voters than immigration or racial equality, two issues that Democrats assumed would clear the path to big gains among Hispanic voters.

In this context, it is interesting to note that the Black Lives Matter (BLM) movement did not rate very highly among Hispanics. In the national exit poll, Hispanic voters were split close to evenly about BLM, 47 percent unfavorable to 49 percent favorable. This significantly trails not just black voters, but also white college graduates, who rated BLM 61 percent favorable to 35 percent unfavorable.

Consistent with this, Latino voters evinced little sympathy with the more radical demands that came to be associated with BLM. In VSG data, despite showing support for some specific policing reforms, Hispanics opposed defunding the police, decreasing the size of police forces and the scope of their work and reparations for the descendants of slaves by 2:1 or more.

An important thing to remember about the Hispanic population is that they are heavily oriented toward upward mobility and see themselves as being able to benefit from available opportunities to attain that. Three-fifths of Latinos in the national exit poll said they believed life would be better for the next generation of Americans. In the VSG data, these voters agreed, by 9 points, that racial minorities have mostly fair opportunities to advance in America, by 11 points agreed that America is a fair society where everyone has a chance to get ahead and by 20 points agreed that “Irish, Italian, Jewish, and many other minorities overcame prejudice and worked their way up. Blacks should do the same without any special favors.”

They are also patriotic. By well over 3:1, Hispanics in the VSG survey said they would rather be a citizen of the United States than any other country in the world and by 35 points said they were proud of the way American democracy works. Clearly, this constituency does not harbor particularly radical views on the nature of American society and its supposed intrinsic racism and white supremacy.

It is probable that Democrats will continue to have problems with this voter group until they base their appeals to this group on what these voters care about the most rather than what Democrats believe they should care about.

Adam Posen draws a parallel between Ohio and North Rhine-Westphalia a comparable region in Germany, "...manufacturing employment as a share of total employment has been steadily declining in all high-income economies, even those with strong export sectors and trade surpluses, like Germany. State-level data provide a more complete picture. Germany persistently recorded trade surpluses in goods between 1992 and 2020, but manufacturing employment as a share of total employment in its industrial heartland has fallen at a similar rate to that in the United States, which recorded a trade deficit in goods in that time frame. Manufacturing employment in North Rhine-Westphalia, an industrial hub in western Germany that has recently been reinventing itself as a high-tech and services center, fell as a share of total employment by 11.7 percentage points across the period, losing 658 thousand jobs. In Ohio, a Midwestern state in the United States with a similar manufacturing history, manufacturing employment as a share of total employment dropped by 8.1 percentage points, or 341 thousand jobs. Both states are heavily populated. Ohio's population is 11.7 million and North Rhine-Westphalia's is 17.9 million. The parallel trends observed in both regions suggest the decline is driven by forces other than trade balances or policy. Common factors, including increased use of labor-saving technologies and improvements in productivity, may be responsible as more efficient technologies reduce the need for manufacturing workers...."

Adam Posen, "Despite Germany's trade surplus, manufacturing employment share of total employment has fallen in its industrial hub at a similar rate as in Ohio," Peterson Institute For International Economics, October 6, 2021, https://www.piie.com/research/piie-charts/despite-germanys-trade-surplus-manufacturing-employment-share-total-employment

Despite Germany's trade surplus, manufacturing employment share of total employment has fallen in its industrial hub at a similar rate as in Ohio

State-level data provide a more complete picture. Germany persistently recorded trade surpluses in goods between 1992 and 2020, but manufacturing employment as a share of total employment in its industrial heartland has fallen at a similar rate to that in the United States, which recorded a trade deficit in goods in that time frame. Manufacturing employment in North Rhine-Westphalia, an industrial hub in western Germany that has recently been reinventing itself as a high-tech and services center, fell as a share of total employment by 11.7 percentage points across the period, losing 658 thousand jobs. In Ohio, a Midwestern state in the United States with a similar manufacturing history, manufacturing employment as a share of total employment dropped by 8.1 percentage points, or 341 thousand jobs. Both states are heavily populated. Ohio's population is 11.7 million and North Rhine-Westphalia's is 17.9 million.

The parallel trends observed in both regions suggest the decline is driven by forces other than trade balances or policy. Common factors, including increased use of labor-saving technologies and improvements in productivity, may be responsible as more efficient technologies reduce the need for manufacturing workers. The regional histories demonstrate the futility of fixating on trade balances to protect manufacturing jobs, a policy course that is doomed to fail while ignoring long-standing injustices.

Research from Matthew Smith, Owen Zidar and Eric Zwick uses administrative tax to estimate wealth concentration/composition for the US. Their bottom line, “…the top 0.1% share of wealth has increased from 12.9% to 15% from 2001 to 2016. While this increase is lower than some prior estimates, wealth is very concentrated – the top 1% holds nearly as much wealth as either the bottom 90% or the P90-99″ class. We find that pass-through business and public equity wealth are the primary sources of wealth at the top, and pension and housing wealth account for almost all wealth for the bottom 90%….”

Core of paper, “…This paper uses administrative tax data to estimate top wealth in the United States. We assemble new data that links people to their sources of capital income and develop new methods to estimate the degree of return heterogeneity within asset classes. We combine this Data on fixed income and pass-through business returns with refined estimates of C corporation equity, housing, and pension wealth to deliver new capitalized wealth estimates. Our approach which builds on SZ and Piketty, Saez and Zucman (2018) (PSZ) as well as Bricker, Henriques and Hansen (2018) (BHH)|reduces bias because wealth and rates of return are correlated. We provide new wealth estimates, new evidence on the rates of return, and a systematic analysis of the issues most consequential for capitalization. We find less wealth concentration relative to the equal-returns, individual-level approach in PSZ, especially at the very top. Figure 1A shows that the top 0.1% wealth share in 2016 is 15% under our approach, and around 20% in PSZ. Top 1% and 0.01% shares fall by 24 percent and 36 percent, respectively, leaving the recent wealth estimates above the estate tax series and closer to the SCF. The growth in top wealth shares is also less dramatic, especially in the tail. For example, our approach reduces the growth in top 0.01% shares since 1989 by 45%. Nevertheless, wealth is very concentrated:the top 1% holds nearly as much wealth as either the bottom 90% or the P90-99″ class. In terms of top portfolios, we find a larger role for pass-through business wealth and a much smaller role for fixed income wealth than PSZ, consistent with the composition of top wealth in the SCF, estate tax data, and surveys of family offices of the ultrarich. Passthrough business and C-corporation equity wealth are the primary sources of wealth at the top. At the very top, C-corporation equity is the largest component, accounting for 40% of top 0.01% wealth, but pass-through business looms large at 29%. In contrast, pension and housing wealth account for almost all wealth of the bottom 90%…”

The Level of Top Wealth, “… Table 2 shows the number of individuals in each wealth group and the wealth thresholds defining each group. We then report average wealth and the share of total wealth for these groups when applying the equal-returns approach and ranking of PSZ. Panel A focuses on top wealth groups. The full population includes 239 million individuals whose average wealth is $364K in 2016. The top 1% includes 2.4 million individuals with wealth of at least $3.7M and average wealth equal to 32 times average wealth in the full population. In terms of shares, this group’s share of total wealth is 31.5% under our preferred approach, compared to 36.6% under PSZ equal returns. Similarly, for the top 0.1%, who have wealth exceeding $17.8M, our estimates reduce their share from 18.6% under equal returns to 15.0% in our preferred specification. Thus, the combined effect of accounting for estimated heterogeneity, updating Financial Accounts aggregates, estimating private business values, adding Forbes 400 data, and including unfunded pension wealth materially affects the estimated concentration of top wealth. These adjustments are increasingly important within the very top group, as the top 1% share falls by 14% (5:1=36:6), the top 0.1% share falls by 19% (3:6=18:6), and the top 0.01% share falls by 26% (2:5=9:5). Panel B focuses on intermediate wealth groups. A key result is that the bottom 90%, who collectively hold 34.3% of wealth, are allocated 5.6 p.p. more wealth than in PSZ. The P90-99″ class, a group with more than $717K but less than $3.73M in preferred wealth, hold 34.2% of total wealth, on par with the bottom 90% and more than the top 1%. Figure 11 compares Forbes 400 wealth to aggregate wealth according to our preferred specification for telescoping subgroups of the top 1%: P99-99.9, P99.9-99.99, and the top 0.01%. We report totals and counts of individuals in each group as well as results using tax units. The wealth threshold to be in the top 0.01% in 2016 is $84M for individuals and $124M for tax units. The figure provides perspective on the relative importance of accounting for Forbes wealth if capitalization alone misses their unrealized stock wealth in non-dividend-paying companies. The Forbes 400 have considerable wealth ($2.4T in 2016), but the total wealth of the P99-99.9 and P99.9-99.99 tax unit groups exceeds this amount by factors of 6.2 and 3.0, respectively. Of course, Forbes members are much wealthier on average: these groups respectively contain 1.5 million and 150 thousand tax units, whereas Forbes represents only 400. Our top 0.01% group contains $6.6T of wealth, which includes the impact of blending Forbes into our data…”

Matthew Smith, Owen Zidar and Eric Zwick, “Top Wealth in America: New Estimates and Implications for Taxing the Rich,” National Bureau Of Economic Research, October 2021, https://www.nber.org/papers/w29374

Individual-Level Rates of Return, “… Figure 4B presents fixed income rates of return for 2016.We calculate rates of return as the group-level ratio of total interest income divided by total interest-generating fixed income assets. We plot these returns ranking individuals by our estimate of total wealth. Rates of return increase from 0.80% for P0-90 to 0.77% for P90-99 to 0.89% for P99-99.9 to 1.65% for P99.9-99.99 to 3.43% for P99.99-100. Rates of return that rank by AGI or by non-interest wealth display moderately greater heterogeneity in absolute terms though the differences are similar in relative terms. Overall, these data reveal a striking amount of return heterogeneity with the P99.99-100 wealth groups receiving returns that are 3.3 times average returns. At the same time, top rates of return are considerably below the top boutique rates, which reflects the mix of high- and low-yielding fixed income assets held by
those at the top of the wealth distribution…”

“… Figure 4C shows the time series of top 0.01%, top 0.1%, top 1%, and bottom 99% rates of return ranked by our preferred wealth estimates. We compare these rates to the equal returns rate and to various capital market rates: the deposit rate from Savov, the 10-year US Treasury rate, and the Moody’s Aaa and Baa corporate bond rates. All interest rates reached a peak in the 1980s during the Volcker tightening and have been falling since then. In the years since 2000, the bottom-99% rate tracks the deposit rate closely, exceeding it by approximately 0.8% in the low-interest-rate period. The equal-returns yield, which fell from 9.5% in 1982 to 1.1% in 2016, exceeds the bottom-99% but is below the top-1% and top 0.1% rates. The top-1% rate tracks the 10-year US Treasury rate although is slightly lower since the Great Recession. The top-0.1% rate hovers between the 10-year US Treasury and the Aaa rate, moving toward the 10-year rate in the last few years of the sample. In our series, top-0.01% rate is below the riskier Baa corporate bond rate in almost all years and is slightly below the Aaa rate in 2016….”
Note they ~ mirror Sarin’s findings on the impact of social security“…For pension wealth, we capitalize an age-group specific combination of wages and pension distributions. This approach allows us to incorporate the life-cycle patterns in pension wealth and associated income flows. While less important for top wealth,pension wealth accounts for 63% of wealth for the bottom 90% and 36% for the P90-99 group. Although we do not account for the value of Social Security in our main specification, we show that doing so would further increase the role of this category of wealth and flatten the trend in measured wealth concentration….The life-cycle of pension wealth accumulation further complicates the capitalization approach. Figure 9A uses the SCF to plot average wages, pension income, and pension wealth in 2016 dollars, averaging across cohorts from 1989 to 2016. Wage income grows over the life cycle and then declines starting around age 55 to near zero by age 75. In contrast, pension income is nearly zero until age 60. Pension wealth has an inverse-U shape that reflects the accumulation and decumulation of savings. These life cycle dynamics result in flow-to-stock ratios that vary by age. Figure 9B summarizes this heterogeneity by plotting the ratio of wage and pension income to total pension wealth, respectively. The blue bars depict the population average and the red bars show the ratios for four age groups: below 45, 45 to 59, 60 to 74, and above 75. Wage income of adults younger than 45 amounts to 108% of their pension wealth on average, whereas average wages for those above age 75 are only 3% of their pension wealth. The patterns for pension income are reversed. The ratios for those between 45 and 74 are closer to the population averages in blue, with the 45 to 59 aged group having a wage to pension wealth ratio that is similar to the overall average, while those aged 60 to 74 have smaller wage to pension wealth ratios, reflecting larger retirement rates. Overall, the heterogeneity in pension wealth and ow-to-stock ratios across age groups means that an age-group-invariant approach will induce large errors…”

Level and composition of aggregate Wealth. “… Figure 2A decomposes aggregate wealth and plots the evolution of key components relative to national income.Other than pass-through business, each component is from the Financial Accounts. In 2016, national wealth amounts to 540% of national income. The largest component is pensions, which equals 203% of national income (of which 40 p.p. are unfunded defined benefit pensions).8 Housing net of mortgages is the next largest (117%), followed by fixed income assets (94%), pass-through business which includes proprietorship, partnership, and S-corporation equity (71%)|and C-corporation equity (67%). Combined C-corporation and pass-through business wealth gives 138%, fifty percent more than the amount of fixed income wealth and commensurate with funded pension wealth. Non-mortgage debt, which includes credit-card balances, debt secured by durable goods, student loans, and other loans, amounts to -16% of national wealth. Aggregate wealth is 77 percentage points of national income higher than in PSZ, of which 40 p.p., 15 p.p., 10 p.p., and 12 p.p. are from unfunded defined benefit pensions, our bottom-up pass-through estimates, adjustments to non-mortgage debt, and residual updates. At the aggregate level, wealth has increased from 346% in 1966 to 540% of national income. Of that increase, 124 percentage points are from pensions, 38 are from net housing, 22 from pass-through business, 18 are from fixed income, and -7 from C-corporation equity.

Level and composition of observed capital income.. “…. Figure 2B plots six types of capital income relative to national income from 1966 to 2016. Aggregate interest income of U.S. individuals increased in the late 1970s and boomed in the early 1980s. It then fell in the 1990s back to its initial share of national income. Since 2000, aggregate interest income has been falling and amounted to 0.6% of national income or $102 billion in 2016. Pension and pass-through income are now the largest sources of capital income. Pension income has risen tenfold from 0.7% to 6% of national income from 1966 to 2016.Pass-through income was 6.8% in 1966, fell to 4% in the early 1980s, and then recovered following the Tax Reform Act of 1986 to 7.3% in 2016. Aggregate dividend income of U.S. individuals amounts to 1.6% and has fluctuated mildly around that level over this period. In contrast, aggregate capital gains of U.S. individuals is much more volatile and ranges from 2% to over 8%. Aggregate property tax payments, which are capitalized to estimate housing assets, amount to approximately 1.2% and grew modestly during the 2000s housing cycle….”
Good factoid, “…Therefore, a dollar of interest income for a wealthy person corresponds to a different level of assets than for a poorer person. Figure 3A uses the 2016 SCF to decompose fixed income holdings into two broad categories: liquid assets, including currency, deposits, and money market funds; and less liquid assets, including bonds, non-money-market fixed income mutual funds, and other fixed income assets. Among fixed income assets, high net worth households have more of their fixed income assets in bonds and other securities. The top 0.1% hold less than 20% of their fixed income portfolio in liquid assets. Bonds and xed income mutual funds account for over 80%. In contrast, the bottom 90% hold more than 80% of their fixed income assets in liquid assets…”

Top wealth overtime, “…. Figure 1 plots our preferred estimates from 1966 to 2016 for the top 0.01%, top 0.1%, and the top 1%. For the top 0.1%, top wealth falls from 10% in the late 1960s to a low of 5.7% in 1978, then steadily rises to around 15% in recent years. Relative to the PSZ series, our preferred series not only shows a lower level in recent years but less growth since 1980. The PSZ top 0.1% series grew from 6.3% in 1978 to 18.6% in 2016; our preferred series grew from 5.7% to 15.0%. Focusing on the 1989-2016 period during which the SCF is available, the top 0.1% share grew 5.1% in our series, 4.3% in the SCF, and 8.1% in PSZ. The gap between the PSZ and preferred series is even larger in recent years for the top 0.01%. Our series and the PSZ series track each other closely before 2000, but they diverge in 2000, especially since 2007. In our series, the top-0.01% shares increase from 5.8% in 2001 to 6.2% in 2006 to 7.0% in 2016; in the PSZ series, the increase from 2001 to 2006 is similar but the increase from 2006 to 2016 is three times larger. For both of these top groups, our series closely tracks the harmonized SCF with Forbes in recent years. For the top 1%, we find a similar trend to the harmonized SCF with Forbes but a lower level. Since 2000, the SCF top 1% share is between the equal-returns series and our preferred series, though shows a sharper increase between 2013 and 2016 that appears to have partly reversed in the 2019 survey. Figure 13 plots time series versions of Figure 12 for the major asset classes for the top 0.01%, top 0.1%, and top 1% in our series, the PSZ equal-returns series, and the harmonized SCF. The figure helps provide a more systematic presentation of the composition of top wealth over time relative to the equal-returns approach. The figure also displays when different updates occur (1980s for pension and housing, 2001 for pass-through and fixed income with information-returns) and the corresponding effects, and how policy and macroeconomic conditions affect the concentration and composition of wealth. For the top 1%, we include estimates from the DFA for comparison…”

Composition of top wealth, “…Tables 3A and 3B show the wealth composition in 2016 for each wealth group in our preferred approach. Pass-through business, C-corporation equity, and fixed income account for 26%, 32%, and 23% of top 0.1% wealth, respectively, with the rest in housing and pensions. At the very top, C-corporation equity is the largest component, accounting for 40% of top 0.01% wealth, but pass-through business looms large at 29%. In contrast, the wealth composition for the bottom 90% is 63% pensions and 23% in housing. The portfolios of the P90-99 are more balanced, with almost equal shares from fixed income (18%), C-corporation plus pass-through equity (21%), housing (25%), and a larger role for pensions (36%). Figure 12 plots the level and allocation of wealth across asset classes among the top 10%. We group individuals into percentile bins and further divide the top 1% into P99- 99.9, P99.9-99.99, and the top 0.01%. Each plot shows the share of total household wealth accruing to that group in a particular asset class. We compare our preferred estimates to the PSZ equal-returns approach and the harmonized SCF with Forbes. The figure displays where in the distribution and across assets differences in approach lead to differences in top wealth shares. Overall, the top 0.01% has 7.0% of total household wealth in our series, of which 1.3 p.p., 2.0 p.p., 2.8 p.p., and 0.9 p.p. are due to fixed income, pass-through business, public equity, and other categories, respectively. The largest difference between our series and the PSZ series is fixed income, for which the PSZ approach estimates fixed income assets of the top 0.01% account for 4.1% of total US household wealth. This difference is partially o set by our pass-through business estimate, which exceeds PSZ’s estimate of 1.1% of total household wealth by 0.9 percentage points. The estimates for the other asset classes are similar for the top 0.01%. In our series, those in the P99-99.9 hold a substantial amount of wealth that exceeds that held by the top 0.1% in terms of fixed income and have considerably more wealth in pensions and housing. For pass-through wealth, the P99-99.9 hold 2.9% of total household wealth, whereas the top 0.1% holds 3.9%. C-corporation equity is more concentrated, as the top 0.01% holds more wealth than the P99-99.9 and P99.9-99.99 groups despite representing 1/100th and 1/10th the number of individuals, respectively…”

“… Figure 14A plots top 1%, P90-99, and P0-90 wealth shares over this time period under both our preferred and the equal-return approaches. The difference in growth between the PSZ and preferred approaches is less pronounced for the top 1% than for the top 0.1% and top 0.01%, with the growth of the top 1% share from 2001 to 2016 falling from 5.4 to 4.2. percentage points. Overall, wealth is still concentrated: the top 1% holds nearly as much wealth as either the bottom 90% or the P90-99″ class. The evolution of the P0-90 versus P90-99 shares from 1965 to 2000 reflects the evolution of pensions, housing, and public equity and relative exposures for different groups. Aggregate pension wealth rises secularly over this time, which is most important for the bottom group. Housing wealth rises and falls in the 1980s, affecting the bottom group and the P90-99 groups significantly. Public equity wealth falls in the 1970s, remains low, and then resurges in the mid-1990s, which drives the time series for the top 1%. In more recent years, the bottom 90 group loses ground relative to both the top 1% and the P90-99. These results are consistent with findings from other data sets (e.g., Kuhn, Schularick, and Steins (2020)). Saez and Zucman (2016) also highlight the decline of P0-90 wealth driven by housing and an increase in debt. Our series shows a less dramatic decline due to the increased role for pensions, including unfunded defined benefit plans, smaller aggregate non-mortgage debt, as well as the more concentrated nature of housing wealth in our unequal-property-tax-rate series. Average wealth of the bottom 90 increased modestly by 17% from 2001 to 2016 (from $120K to $140K in 2016 dollars), whereas average wealth for P90-99 and the top 1% rose by 40% and 49% (from $1.0M to $1.4M and from $7.7M to $11.5M), respectively…”

The rich are able to get better returns, “…We introduce two innovations to estimate fixed income wealth. First, we construct a novel data set on the universe of taxable interest sources linked to owners using de-identified ed data from income tax records spanning 2001-2016. These 3.2 billion source-owner observations allow us to disaggregate taxable interest income into subcomponents. This disaggregation reveals that rich individuals earn a much larger share of their interest income in the tax data in higher-yielding forms (such as boutique investment partnerships of distressed debt or mezzanine funds). Disaggregation also allows us to estimate interest rates more accurately than prior work. These data reveal a striking amount of return heterogeneity across wealth groups, with the top 0.01% group receiving returns that are 3.3 times average returns. In 2016, our estimates increase from nearly 1% within the bottom 99.9 to 1.6% for P99.9-99.99 to 3.4% for the top 0.01%…“