Among the many impossible-to-predict consequences of the Covid-19 pandemic for the 2020 presidential campaign, this may be one of the most surprising: The Trump campaign has taken door-knocking much more seriously than the Biden campaign.
Door-to-door canvassing — where campaign workers knock on doors to either persuade residents to vote for their candidate or remind the already persuaded to turn out — is traditionally a strong suit of Democratic campaigns. “Field,” as it’s called, is where many leading party strategists, from 2008 Obama campaign manager David Plouffe to 2020 Biden campaign manager Jen O’Malley Dillon, came up. Political scientists have written whole books about Obama’s effort to mobilize millions of volunteers for field operations in 2008 and 2012. In 2016, the conventional wisdom that Hillary Clinton was a heavy favorite to defeat Trump derived partly from a sense she had a better “ground game.”
But in 2020, the politics of Covid-19 mean that pattern is reversed. Trump is knocking on doors, and Biden, until very recently, was not.
In late August, the Trump campaign in Michigan boasted that it has “an army of over 43,000 volunteers and staff covering all 83 counties,” whereas the state Democratic Party told reporter Jonathan Oosting it was “not yet comfortable” knocking on doors. On August 28, the Trump campaign bragged about knocking on its 1 millionth door in Florida. Door-knocking numbers like this aren’t really useful (they don’t tell you how many people were actually contacted), but the pattern is illustrative.
The Biden campaign responded to the risk that door-to-door canvassing will spread Covid-19 infection by shutting down its door-knocking efforts. On October 1, O’Malley Dillon finally announced the campaign would start limited door-knocking efforts, but only after months of not doing any door-knocking at all. The Trump campaign, reflecting the president’s efforts to downplay the pandemic and speed up “reopening,” is doing the opposite, and still doing volunteer recruitment for canvassing.
Given the central role canvassing has played in recent Democratic presidential campaigns, you would think this discrepancy would prompt some concern among Democrats who believe in the ground game. But some people in the world of Democratic campaigns are coming around to a different view: Maybe door-knocking really isn’t as important as we all thought.
Most political scientists I talked to affirmed this view. Melissa Michelson, professor of political science and an expert on field experiments for voter turnout, told me, “That the Biden campaign can’t engage in door-to-door canvassing is maybe not that big of a loss because we actually have tons of data about how to effectively and in a cost-efficient way mobilize voters from all different parts of the Democratic coalition — younger voters, low-income voters, Black voters, Latino voters — without going door to door.”
Indeed, there’s a growing body of research suggesting that methods like calling voters and “relational” voter turnout seem to be as effective — if not more effective — than traditional door-knocking.
To be clear, there’s a consensus that field work can juice turnout and even persuade voters in primary elections or local elections where the candidates are less well-known and voters’ opinions are less formed.
But skeptics argue that you can’t just look at whether the effect of a field operation is positive or negative. You have to ask how many votes it pulls in per dollar spent, and compare that to whatever the equivalent figure is on alternative uses of campaign money: TV ads, digital ads, direct mail, and non-knocking fieldwork like phone banking. Given the expense of running a good field team, skeptics argue that the cost-per-vote is too high relative to alternatives and that Covid-19 might serendipitously be pushing campaigns away from inefficient uses of resources and toward more efficient ones.
“There’s not any other kind of information we try to communicate by going door to door,” says David Shor, an independent Democratic data analyst who helped develop Obama’s data analytics operation in 2012. When big companies want to get the word out about their products, they use ads — and Shor and other field skeptics think campaigns should double down on those, too.
Persuading voters in presidential general elections is difficult
It’s important to distinguish between two different purposes of political campaigns. One purpose is persuasion: convincing an undecided voter, or even one who’s decided for the other candidate, to support your candidate. The other purpose is turnout: getting people who are already persuaded to support your candidate to actually vote.
Persuasion obviously happens by some mechanism in elections — swing voters are rare, but real, and a large number of voters switched from supporting Trump in 2016 to supporting Democrats in the 2018 midterms. That’s persuasion of at least some kind taking place.
But when political scientists try to evaluate the effect of specific campaign interventions at persuasion, the results tend to be quite dire. Even though persuasion happens, it’s hard for campaigns to reliably pull it off. UC Berkeley’s David Broockman and Yale’s Joshua Kalla conducted an evidence review in 2017 of studies evaluating how effective interventions aimed at persuading voters are, be they canvassing, phone calls, direct mail, TV, online ads, or anything else under the sun.
They found that there’s lots of room for persuasion in primaries; a canvass during the 2015 Philadelphia mayoral primary, an open race where voters didn’t know the candidates that well, was quite effective. But in general elections? Nope.
“The best estimate for the persuasive effects of campaign contact and advertising — such as mail, phone calls, and canvassing — on Americans’ candidate choices in general elections is zero,” they write. “Our best guess for online and television advertising is also zero, but there is less evidence on these modes.”
There are unique circumstances where persuasion tactics become more effective for campaigns (see the last section here). But the 2020 presidential election doesn’t feel like one of them. Voters know who Donald Trump is. They know who Joe Biden is. Given everything voters have learned about both of them over the past four years, or past several decades even, it’s unlikely that door-knocking or phone banking is going to be the decisive factor in changing people’s votes.
Turning out existing voters, however, is possible
Even though campaign contact doesn’t seem to persuade people in general elections, it’s not necessarily a bad idea. That’s because campaign contact can increase turnout, too.
The empirical literature here was kicked off by political scientists Alan Gerber and Donald Green two decades ago, and their book Get Out the Vote! is still one of the best sources on the evidence base around the effectiveness of canvassing, phone banking, and direct mail at turning out voters.
In the book’s fourth edition, released in 2019, Gerber and Green estimate that the average per-conversation effect size of canvassing turnout operations is 4.0; the average effect of commercial phone banks is 0.947; the average effect of volunteer phone banks is 2.8. This is across a variety of elections, though, not just presidential ones, where effect sizes might be lower.
That’s a lot of numbers, so let me spell it out: If canvassing has an effect size of 4.0, that means that a door-knocking operation that knocks on 5,000 doors, and gets a response at 1,000 of those doors (a pretty standard or even high response rate) will generate 40 new voters. Similarly, a volunteer phone bank that reaches 1,000 people will produce about 28 new voters, since the effect size is 2.8.
Contacting voters by phone might be just as good as door-knocking anyway
When you put it like that, it makes door-knocking look considerably better than calling voters, which is likely to replace it in a Covid-19 environment.
But you can also talk to more people in an hour through phone banking than through canvassing. You don’t have to walk or drive between addresses. Put it all together and Gerber and Green’s rough estimate is that canvassing can garner campaigns a vote for about $33, while volunteer phone-banking can garner a vote for $36 — not too different, especially when you consider how imprecise these estimates necessarily are.
Michelson, the professor of political science at Menlo College, has evaluated dozens of experiments testing turnout and persuasion, reporting many in her 2012 book with UC Berkeley’s Lisa García Bedolla, Mobilizing Inclusion. They found that calling voters produced more consistently positive results than door-to-door canvassing, in part because it was easier for callers to stick to a script than it was for canvassers. “What we found is that although door-to-door canvassing could generate the largest effects, it was not actually our recommended method,” Michelson told me. “We thought that other things like two-round phone banks were more effective.”
In “two-round” phone banks, voters are first contacted several weeks before the election, and then get a follow-up call to encourage them to vote a day or two before the election.
Door-knocking and phone banking are not the only possible ways to contact voters, of course. Gerber and Green estimate that the average effect of conventional mailers from nonpartisan groups meant to get out the vote is 0.296 per mailer (so a mailer that reaches 1,000 people might produce 3 new voters). Robocalls and explicitly partisan direct mail (fliers advertising a particular candidate, say) don’t seem to have any effect.
But “social pressure” mailers have a bigger effect on average. In that tactic, campaigns or other turnout organizations use mail to remind voters that whether or not they voted in the past is public information, and share information on which of their neighbors voted.
In a study of the June 2012 Wisconsin recall election for Gov. Scott Walker, Green, Harvard’s Todd Rogers, Yale’s John Ternovski, and Carolina Ferrerosa Young (now working for Democratic Sen. Mark Warner) found that social pressure mailers like this sent out by an anti-Walker group boosted turnout of Democratic-leaning voters substantially, working out to a cost per vote of roughly $55. And in a 2006 primary election in Michigan (a much less visible race), Gerber, Green, and Christopher Larimer found that social pressure mailers produced votes for $1.93 each, an astoundingly low price.
Social pressure techniques can generate backlash — as when Ted Cruz used them in the 2016 Iowa caucus but used made-up voter histories and told voters they’d committed a “VOTER VIOLATION” in a font that kind of implied they’d done something illegal — but they seem consistently effective.
The limits of our existing data
There are a few factors, however, that complicate this public data. For one thing, only a small share of all studies conducted on the effects of canvassing and phone banking on turnout are public. The majority are conducted through firms like the Analyst Institute (for Democrats) or Evolving Strategies (for Republicans), and the results are typically proprietary. Democrats don’t want Republicans to benefit from studies they spent millions of dollars conducting and vice versa (though everyone acknowledges that Democrats invest vastly more in these kinds of experiments than Republicans do).
Don Green, the Columbia political scientist and doyen of campaign field experiments, told me, “When [Gerber and I] do the meta-analysis, we are super careful to get our hands around everything we can find,” including unpublished working papers and even individual regressions. He believes that their public summary of the evidence is representative of what groups like the Analyst Institute know privately. But there’s a limit on how much journalists like me or the general public can know about the evidence base, given how much of it is secret.
The other complication is that phone banking has diminished in effectiveness as people have moved to cellphones, which are harder to match to voter records, and which they’re less likely to use as phones and less likely to pick up calls from unfamiliar numbers. Parties are getting better at reaching cellphones — the Democratic National Committee announced a massive purchase of tens of millions of cellphone numbers in January — but it does complicate matters.
Advertising matters — potentially a lot
Digital advertising is another arrow in a campaign’s quiver. In recent months, it has become a highly public topic of controversy in light of Facebook and Twitter’s role in spreading false or misleading ads.
Most evidence to date, though, is dismissive or murky as to digital advertising’s effectiveness. An early randomized experiment by Broockman and Green using Facebook found that ads weren’t effective at boosting favorability or name recognition for a Republican state legislative candidate or a Democratic congressional candidate. The nonpartisan group Rock the Vote conducted experiments in 2012 and 2013 using Facebook ads, garnering millions of impressions, meant to boost turnout, and found no difference between the treatment group and the control group.
Kalla evaluated a similar experiment conducted by NextGen Climate, an advocacy group founded and funded by billionaire Tom Steyer. The experiment assigned over 1 million voters in New Hampshire, Nevada, and Pennsylvania to receive online ads urging them to turn out and vote in the 2016 general election. Two-thirds of the ads were on Facebook. Kalla found a very small effect on turnout, with an overall cost of $474 per vote, which is very high compared to phone banking or door-knocking.
Political scientists tend to be similarly dismissive of television ads. “The turnout effects of TV and digital ads are basically zero,” Green told me, with some exceptions like Rock the Vote that he himself has tested. Green and UCLA’s Lynn Vavreck found that an advertisement specifically designed to boost youth turnout did just that — but most campaign ads aren’t as laser-focused on turnout.
“I don’t know that there’s a study that shows that the turnout effects of a regular TV campaign, not the old ones that Lynn Vavreck and I crafted 16 years ago that were specifically about get-out-the-vote. It’s basically a null set. There are some non-experimental papers that purport to find effects, though there are plenty of non-experimental findings that show no effect,” he said.
But those TV ads are generally trying to do persuasion, not turnout — and some non-experimental papers suggest they help with persuasion. There’s a long history of observational studies (looking at ads after the fact rather than deploying them in a randomized way as part of a formal experiment) evaluating the effectiveness of TV ads, and analysts like Shor find them compelling. Moreover, they think the weight of the non-experimental evidence is more in favor of TV than against it.
In 1999, University of Texas at Austin’s Daron Shaw found substantial statewide effects of TV advertising in statewide races from 1988 to 1996; a book by a trio of political scientists found that TV advertising was effective in the 2000 race. Studies analyzing the 2004 and 2006 elections came to similar conclusions.
Most recently, a study under the unassuming title of “Political Advertising and Election Results,” from economists Jörg Spenkuch and David Toniatti (un-paywalled copy here) pooled results across three elections and used a particularly compelling design. Spenkuch and Toniatti exploit the fact that TV ad-buying is done on the “media market” level, and that neighboring and similar counties are often in different “media markets” due to FCC regulations, to measure the effect of TV ads on US presidential elections from 2004 to 2012.
They find substantial effects: “Showing someone 100 ads in the month before the election in a presidential race increases their odds of voting for you by 1 percent,” Shor summarizes. If a TV ad costs only, say, $30 per 1,000 impressions, then that’s a much more favorable cost-per-vote than phone banking or canvassing.
Shor also points to the 2020 primary for a particularly illustrative example:
[Elizabeth] Warren spent the entire cycle building up this massive army of organizers in Iowa who knocked on a bunch of doors. This is a low-salience race: only 7.6 percent of Iowans voted in the [Democratic] caucuses. But what [Pete] Buttigieg did is take roughly the same amount of money she spent on field, and spent it on TV in the summer [of 2019]. That made him go up in the polls to the point that the media started covering him. That helped him raise more money so he could buy more TV, and he almost won.
Shor’s conclusion: The Warren organizer-based model is a waste of money. The Buttigieg strategy, based on TV and earned media, got much closer to succeeding.
“A solid 95 percent of people who work in data for politics are coordinating field programs, which is wild,” Shor says. “What we actually need are people with a different skillset, who are good at making ads, are good at ad tech, or work in video production. I have a friend who’s a professional video producer who was asking me where she should go knock on doors. That’s a massive misallocation of resources.”
Others, like Green and Kalla, consider the Spenkuch/Toniatti study a rather thin reed to build a pro-TV, anti-field case upon. Kalla acknowledged that the study “changed my prior a bit” in favor of TV being effective.
But he noted that it identified “a very weird mechanism” for its effects. TV ads tend to be geared more toward persuasion than turnout. But TV, Spenkuch and Toniatti find, changes the turnout of the electorate. When one party has an advantage on TV, its supporters turn out more in general elections than supporters for the candidate who fell behind on TV ads. Green calls the study “basically an anomaly. What is the mechanism that would cause turnout to be the driving force behind a vote shift? That really struck me as strange.”
This debate might seem esoteric, but the stakes are high. If Shor (and Spenkuch/Toniatti and their precursors in the literature) are right, then campaigns should be investing far less in the ground game and much more in TV ads with messages refined and perfected through iterative field experiments. If Green is right, then phone banking and canvassing for turnout are still crucial activities for campaigns to be engaged in.
The future is canvassing your friends
Beyond the disagreement over TV, there’s widespread agreement and optimism across the political scientist/data consultant spectrum about the possibilities of “relational voter turnout” that exploits people’s friendships and social attachments.
Case in point: In 2010, Facebook conducted a 61 million-person experiment testing whether a banner urging US members to vote in the midterm elections could juice turnout. Simply putting up an informational banner didn’t work at all. But including faces of friends who’d clicked an “I Voted” button was effective at increasing turnout.
More recently, campaigns have turned to “relational voter turnout,” where instead of phone banking or canvassing strangers, volunteers try to turn out people close to them, like friends and family. Persuading friends and family can be hard, as anyone who’s gotten into a Facebook argument with an uncle can attest, but encouraging turnout is somewhat easier.
“Everybody, even a relatively high-probability voter … would nonetheless know people in the same family, in the same congregation, on the same street, in the same workplace, you name it, who are low-propensity voters,” Green says. “That’s what you’re looking for in 2020: Find 10 people who haven’t voted in a while, are 19, etc.”
Recently, Green and Columbia University’s Oliver McClellan conducted an experiment for the nonpartisan group Turnout Nation in which 43 organizing “captains” in four states each put together lists of 20 “friends or relatives who would be eligible to vote” in upcoming municipal elections. Half of the names were assigned to be treated, meaning captains were encouraged to contact them and ask them to vote. The other half the captains were told not to contact. The effect was 13.2 percentage points, which the authors call “extraordinary, exceeding estimates from any other randomized trial on voter turnout.” In Ohio, where the program was more regimented, the effects were even greater.
Shor disagrees with Green on TV ads, but he’s absolutely on board regarding relational organizing. One informal way for people to engage in it is to, well, post about who they’re voting for and remind people to vote on social media networks like Instagram and Facebook. “The most effective thing that anyone can do is go out and shit-post and talk to their friends and tell people what they believe and what they care about,” Shor says. “That’s what politics is about.”
2020 in general has been a year in which people have struggled to find digital, or at least socially distant, replacements for in-person experiences: eating at restaurants, connecting with friends, celebrating big events. While the particulars are controversial, one unanimous message from political scientists and election analysts is that electioneering translates quite well to a post-pandemic world. Campaign volunteers don’t necessarily need to knock on strangers’ doors as much as they need to be texting their friends to get them to vote.