David Broockman and Joshua Kalla | Vox

Bernie Sanders looks electable in surveys — but it could be a mirage

2020-03-08T16:42:39-04:00

Sen. Bernie Sanders at the South Carolina Democratic Party’s “First in the South” dinner on February 24, 2020 in Charleston, South Carolina. | Drew Angerer/Getty Images

The most important factor for Democratic voters in the 2020 primary is electability: A majority of Democrats say they would rather nominate a candidate who can beat President Trump than a candidate who agrees with them on the issues.

So which candidate is most likely to beat Trump? Decades of evidence from academic studies suggests that more moderate nominees tend to perform better in general elections than more ideologically extreme nominees. For example, Democratic US House candidates who supported Medicare-for-all fared approximately 2.2 percentage points worse in the 2018 midterms than candidates in similar districts who did not.

But early polling testing how Democratic nominees would fare against Trump suggests a different conclusion: Bernie Sanders, the most left-wing candidate in the Democratic primary, polls as well against Trump as his more moderate competitors in surveys. Democratic voters have appeared to take these polls to heart, as a recent Washington Post-ABC News poll finds that Democrats believe Sanders has the best chance of beating Trump.

Why does Sanders look similarly electable to leading moderates in polls against Trump? We fielded a 40,000-person survey in early 2020 that helps us look into this question with more precision. We asked Americans to choose between Trump and one of the leading Democratic candidates: Sanders, Elizabeth Warren, Pete Buttigieg, Joe Biden, and Mike Bloomberg.

So that respondents would not strategically claim to only support their chosen candidate against Trump, we only asked each respondent about one Democratic candidate. The surveys were fielded by Lucid, an online market research company that provides nationally representative samples of Americans.

Our data (laid out in an academic working paper here) also found what polls show: that Sanders is similarly electable to more moderate candidates. But, on closer inspection, it shows that this finding relies on some remarkable assumptions about youth turnout that past elections suggest are questionable.

We found that nominating Sanders would drive many Americans who would otherwise vote for a moderate Democrat to vote for Trump, especially otherwise Trump-skeptical Republicans.

Republicans are more likely to say they would vote for Trump if Sanders is nominated: Approximately 2 percent of Republicans choose Trump over Sanders but desert Trump when we pit him against a more moderate Democrat like Buttigieg, Biden, or Bloomberg.

Democrats and independents are also slightly more likely to say they would vote for Trump if Sanders is nominated. Swing voters may be rare — but their choices between candidates often determine elections, and many appear to favor Trump over Sanders but not over other Democrats.

Broockman / Kalla 2020" data-portal-copyright="Broockman / Kalla 2020" />

Despite losing these voters to Trump, Sanders appears in our survey data to be similarly electable to the moderates, at least at first blush. Why? Mainly because 11 percent of left-leaning young people say they are undecided, would support a third-party candidate, or, most often, just would not vote if a moderate were nominated — but say they would turn out and vote for Sanders if he were nominated.

The large number of young people who say they will only vote if Sanders is nominated is just enough to offset the voters Sanders loses to Trump in the rest of the electorate. (Warren appears to lose at least as many Republicans as Sanders but does not seem to benefit from any compensating enthusiasm from young voters.)

Sanders himself has been clear that his strategy for beating Trump is to massively boost turnout, especially among young people — and young people in our data indeed say they would turn out at much higher rates for him.

But for Sanders to do as well as a moderate Democrat against Trump in November by stimulating youth turnout, his nomination would need to boost turnout of young left-leaning voters enormously — according to our data, one in six left-leaning young people who otherwise wouldn’t vote would need to turn out because Sanders was nominated. There are good reasons to doubt that Sanders’s nomination would produce a youth turnout surge this large.

The reason Sanders appears equally electable

These “Bernie or bust” voters that come off the sidelines for Sanders in our survey are almost entirely limited to one group: Democrats and independents under age 35. These voters are about 11 percentage points more likely to say they would vote for Democrats if Sanders is nominated — and almost all of them say they would not vote at all or vote third party if he’s not on the ballot.

However, the “Bernie or bust” phenomenon appears almost entirely limited to left-leaning young people, who are usually a small share of the overall electorate. This stands in contrast to many theories of Sanders’s electoral appeal: For example, whites without a college degree — a demographic some speculate Sanders could win over — are actually more likely to say they will vote for Trump against Sanders than against the other Democrats. The same is true of the rest of the electorate, except left-leaning young people.

Broockman/Kalla 2020" data-portal-copyright="Broockman/Kalla 2020" />

This finding in our data mirrors many other surveys: Morning Consult finds dramatic increases in young Americans’ stated turnout intentions when asked how they would vote in matchups between Sanders and Trump.

How huge of a turnout surge does Sanders need to be as electable as a moderate?

The case that Bernie Sanders is just as electable as the more moderate candidates thus appears to rest on a leap of faith: that youth voter turnout would surge in the general election by double digits if and only if Bernie Sanders is nominated, compensating for the voters his nomination pushes to Trump among the rest of the electorate.

There are reasons to doubt a Sanders-driven youth turnout surge of this size would materialize. First, people who promise in surveys they will vote often don’t, meaning the turnout estimates that Sanders’s electability case rests upon are probably extremely inaccurate. Second, such a turnout surge is large in comparison to other effects on turnout. For example, Sanders would need to stimulate a youth turnout boost much larger than the turnout boost Barack Obama’s presence on the ballot stimulated among black voters in 2008.

Third, Sanders’s electability case requires this 11 percentage point turnout increase among young voters in 2020 to occur on top of any turnout increase that would otherwise occur if another Democrat were nominated.

If the turnout of all age groups increases from 2016 to 2020 (as happened from 2014 to 2018), then the turnout among young people must increase by 11 percentage points above and beyond this broader trend, and must do so solely due to Sanders’s presence on the ticket. Finally, youth voter turnout doesn’t usually go up or down by nearly as much as 11 percentage points from election to election; the Sanders boost would have to be truly unprecedented.

And this enormous 11 percentage point turnout boost is only enough to make Sanders as electable as the more moderate candidates, given the other votes he loses to Trump. For him to be the most electable Democratic candidate based on his ability to inspire youth turnout, Sanders’s nomination would need to increase youth turnout by even more.

Broockman/Kalla 2020" data-portal-copyright="Broockman/Kalla 2020" />

There is no way to be sure whether Sanders’s nomination would produce this historic youth turnout surge — but it seems doubtful. Turnout in the 2020 primaries so far has not exceeded 2008 levels, including among young voters. If anything, research suggests the opposite is more likely to occur: In response to an extreme Democratic nominee, Republicans could be inspired to turn out at higher rates to oppose him.

What if Sanders’s nomination doesn’t stimulate youth turnout enough to offset the votes it would lose to Trump? In an academic working paper based on this survey, we consider this possibility.

In one analysis, we disregard what voters say about whether they would vote, and use their demographics and party affiliation to infer the shape of the likely electorate. In particular, we base their guesses about who will vote on the demographics of the 2016 voting electorate instead of what people tell us about whether they will vote (and assume people who don’t list a preference will vote for their party). With this approach, Sanders trails all three leading moderate candidates in head-to-head polls against Trump.

Would Republican attacks knock the more moderate candidates down to Sanders’s level?

One concern about our findings is that Republicans who say they would vote for Biden or Buttigieg might not really do so in November, after the general election campaign has heated up. After months of sustained attacks from Trump and Republicans throughout the general election, would the more moderate candidates still be more electable than Sanders?

To examine this possibility, we first conducted an experiment to identify effective attacks against each of the Democratic candidates. For example, Biden’s historical support for freezing Social Security benefits undermined his support, but hearing about Buttigieg’s sexual orientation and the fact that he met his husband online did not decrease his support.

Then, to examine the resiliency of each Democrat’s support in the general election in the face of effective attacks, we showed some of our survey respondents the three attacks that were most effective against each Democrat before asking them who they would vote for in a contest between that Democrat and Trump.

After showing three attacks against each candidate, we find that Sanders would still need the same large youth turnout surge to overcome his deficit relative to the more moderate candidates against Trump. When we analyze the data using the same approach described above that disregards what voters say about whether they will vote, we find that, after being shown the attacks, Buttigieg, Bloomberg, and Biden still do better against Trump than Sanders does. (Warren still performs even worse than Sanders against Trump in this test. We did not include Klobuchar in this survey.)

Prediction markets have been sounding the alarm, too, as Trump’s probability of winning reelection has steadily risen as Sanders has looked increasingly likely to be the Democratic nominee.

Early polls are never a surefire guide to what will happen in an election months later. But Democrats should not be very reassured by early polls that find Sanders faring as well against Trump as the more moderate candidates: These numbers may only look decent for Sanders because they assume he will inspire a youth turnout miracle. Our survey data reveals voters of all parties moving to Trump if Sanders is nominated, a liability papered over by young voters who claim they would be inspired to vote by Sanders alone.

The gamble Democrats supporting Sanders based on his early polls against Trump must be ready to make is that, despite the evidence to the contrary, the lowest-participating segment of the electorate will turn out at remarkably high rates because Sanders is nominated.

David Broockman is an associate professor of political science at the University of California Berkeley. Joshua Kalla is an assistant professor of political science and statistics & data science at Yale University.

Donation disclosure: Broockman has donated $27 to Bernie Sanders’s 2016 campaign, $5 to Julián Castro’s 2020 campaign, $5 to Amy Klobuchar’s 2020 campaign, and $2 to Pete Buttigieg’s 2020 campaign. Kalla has donated $250 to Buttigieg’s 2020 campaign and $100 to Elizabeth Warren’s 2020 campaign.

Update: Broockman and Kalla have published a detailed response to a critique of the data analysis in this article and its accompanying working paper written by Seth Ackerman for the People’s Policy Project. They conclude:

We therefore do not see any reason to revise our paper’s conclusions. At the same time, given that three of the five candidates we examined in our paper have dropped out since we wrote it, we also agree it is reasonable for observers of the Democratic primary to pay renewed attention to a finding in our paper about Joe Biden specifically, although we think the right interpretation of our data is that it is statistically ambiguous with respect to Biden and so other polls are better suited to answer this question — albeit read with the lessons of our paper in mind.

For more, read the whole response here.

We discovered one of social science’s biggest frauds. Here’s what we learned.

2019-03-05T02:02:03-05:00

Earlier this year, we discovered one of social science’s most widely publicized scandals: The data for a celebrated study showing that a face-to-face conversation could radically decrease prejudice toward gays and lesbians seems to have been fabricated.

We didn’t set out to debunk the original study. Quite the opposite, actually. We wanted to replicate and extend it. It was only when we began doing so that we realized the data for the original seemed too good to be true; within days, others determined it was, and the study was retracted.

This episode grabbed headlines not only because of its implications for political advocates, but also because of broader concerns it raised about science. Outlets such as the New YorkTimes argued that a rise in similar retractions should erode public trust in science. “Cheating in scientific and academic papers is a longstanding problem,” the Times editorial board wrote, “but it is hard to read recent headlines and not conclude that it has gotten worse.”

We disagree, and think this is the wrong conclusion to draw from the rise in scientific retractions. But what are the right lessons? After reflecting on our experience over the past few months, we’ve come up with a few.

1) The rise in retractions shows the state of social science is strong, not weak

A rise in scientific retractions might indicate misconduct is on the rise — but it also might reflect how changing scientific norms have made scientific misconduct easier to detect and expose.

Concluding the opposite is an example of survivor bias, an error in reasoning best illustrated with a famous story from military history. During World War II, the US Navy asked Columbia University statistician Abraham Wald to analyze patterns of damage from Axis anti-aircraft fire on planes returning from bombing runs. Wald found that certain areas of the planes, like the wings, were especially likely to come back riddled by bullets. He then recommended that the Navy reinforce the other areas — the ones that were consistently unscathed.

The Navy was puzzled with Wald’s advice. If the wings were so often damaged in battle, wouldn’t that suggest weaknesses in the wings, not the consistently pristine windshields? Wald responded wisely. Enemy bullets land everywhere, he explained. But planes that take fire only return to base for inspection when they’re hit in areas where the planes can withstand it. On the other hand, if no planes that return to base have bullet holes in the windshields, it doesn’t mean that windshields somehow avoid getting hit — rather, it suggests planes struck in the windshield don’t make it back.

In the same way, we see scientific retractions as indicators of science’s unique and growing penchant for telling the truth. Just like enemy bullets struck Navy aircraft everywhere, fraud and mistakes crop up in all walks of life. The fact that we are especially likely to see and hear about the errors scientists make does not suggest scientists are especially prone to making mistakes. Rather, it shows that scientific errors are increasingly likely to be detected and corrected instead of being swept under the rug.

2) The norms of academic science make it much easier to catch fraud and mistakes

One reason scientists often get caught lying is that they are required to publish in ways that make it easier to catch lies and mistakes, and they publish in an environment that celebrates those who catch them.

We were able to detect irregularities in the study’s data because of these scientific norms. Within our discipline — political science — norms to make data sets and code publicly available were first adopted in the early 1990s, recognizing that the pursuit of scientific truth is a community enterprise and that the sharing of data and code is vital to that pursuit. By comparing one publicly available data set to another, we were able to detect results that seemed too good to be true. Had these norms not existed, these irregularities likely would have gone undetected.

Then, after making our discovery, we could come forward because we knew our colleagues valued telling the truth over eschewing controversy or embarrassment. The article’s senior author and Science’s editors readily retracted the study because they did, too.

3) Research is best when it’s independent, public, and not-for-profit

In a time when privatization is a watchword of political debate, our experience also illustrates the importance of protecting these norms and institutions, which remain unique to academic science.

Research and development is a $400 billion industry, of which academia comprises only a small share. Proprietary, for-profit fields of inquiry such as market research, pharmaceutical and agricultural research, and consulting are much larger than academic science. Yet we hear about new scientific retractions daily and scandals in for-profit research only rarely. Why?

We suspect the reason academic scientists are responsible for an outsize share of research mea culpas isn’t because academics are especially dishonest but because academia features unique norms and institutions — like blind peer review and access to replication data — that succeed in allowing scientists to discover and disclose difficult truths, including those that concern others’ work.

Like Wald noted about Navy plane wings, the fact that we can observe the mistakes that occur in academia shows its strength, while the relative silence of private sector researchers about their mistakes should prompt questions, not confidence.

Returning to our story, the fraud we discovered could easily have happened in the private sector — but without open data and academic freedom, it just may never have been detected or exposed. Scientists’ proclivity for admitting our errors thus doesn’t mean scientists commit more of them; rather, it showcases the scientific community’s special dedication to telling even embarrassing truths.

4) Working with academics is the best way to learn what really works

Science’s unique norms and institutions also explain why nonprofits, governments, and corporations increasingly partner with academics to evaluate their effectiveness and test new ideas, rather than relying on our for-profit counterparts. The same norms and institutions that encourage us to tell hard truths about other scientists’ work also encourage us to do the same about non-scientists’ efforts. For groups that work with academics, the uncomfortable truth we are in a unique position to tell is that something the group is doing doesn’t work. Academics deliver this news routinely.

It’s also a matter of culture; open criticism is inherent in an academic’s job description. Anyone who has served on a dissertation committee, sat in an audience at an academic seminar, or completed a peer review has had to tell someone that something is wrong with a project that he or she has spent months or years working on. It’s core to the job.

That’s not always true in the private sector. Fear of losing sales might mean private sector researchers are not able to deliver the news that their firm’s drugs, advice, or advertising doesn’t work — and their consumers or funders would have a hard time finding the fib.

By contrast, when academics do produce evidence that something works, you can be much more confident we’re not fudging the numbers — we’d almost certainly be found out if we did.

5) Academia can still do better — but only if its unique position in society is protected

For all that science gets right, there remains room for improvement. As an example, younger researchers like ourselves should be more clearly encouraged to speak out when they suspect something is amiss in published work. Academia is a close-knit yet highly competitive community, and raising suspicions about your colleague’s work still bears some risk. It needs to be even easier to be a whistleblower.

Only one thing makes us doubt whether science will rise to challenges like this one. At the federal and state levels, politicians are threatening science’s funding and independence more aggressively than ever. (As one of many examples, Congress is trying to cut federal National Science Foundation funding for the social sciences in half.) As political scientists, this is no surprise to us: Politicians are worried about the danger the truths we might tell could pose to their agendas, or to their contributors’ interests. These politically motivated attacks on science are reckless and shortsighted; they endanger the ability of one of humankind’s singular engines for progress to keep telling the uncomfortable truths that make progress possible.

So long as science does continue to receive the public support that ensures its independence, we are confident it will only get better at finding and telling the truth — and, as a consequence, that we’ll keep on hearing about the mistakes scientists make.

David Broockman is an assistant professor of political economy at the Stanford Graduate School of Business. Joshua Kalla is a PhD student in the Charles and Louise Travers Department of Political Science at the University of California Berkeley.

Experiments show this is the best way to win campaigns. But is anyone actually doing it?

2019-03-02T17:46:50-05:00

U.S. Senator Mark Udall glances down at phone packet and talking point materials while waiting to be introduced at a canvass kickoff campaign on October 25, 2014 in Thornton, Colorado. | Marc Piscotty/Getty Images

There’s a refrain we hear about political campaigns every election cycle: “this year, campaigns waged an unprecedented ground game, having a face-to-face conversation with almost every single voter.”

Baloney. As academics who study campaigns, we hear this claim all the time. But we also know it’s important to investigate whether data backs it up. We did. And it doesn’t. In fact, there’s a paradox at the heart of American campaign craft. Mountains of rigorous research show that campaigns should be having personal conversations with voters at their doors. But, campaigns spend almost all their money on TV ads — and, every year, most voters say they’ve never had a conversation about the election at their door. What gives?

Why campaigns’ "ground game" matters

By far the most effective way to turn out voters is with high-quality, face-to-face conversations that urge them to vote. How do we know? Nearly two decades of rigorous randomized experiments have proven it.

Alan Gerber and Don Green ran the first of these “field experiments” in 1998. The professors randomly assigned voters to receive different inducements to vote: some received postcards, some received phone calls, some received a visit from a canvasser, and some received nothing.

The experiment found that voters called on the phone or sent postcards were not noticeably more likely to vote than those sent nothing. But canvassing was different. Just one in-person conversation had a profound effect on a voter’s likelihood to go to the polls, boosting turnout by a whopping 20 percent (or around 9 percentage points).

The nearly two decades since Gerber and Green’s first experiment have consistently borne out their finding that personal conversations have special political potency. Hundreds of academics and campaigns have tested the impacts of various campaign tactics with randomized field trials. High-quality canvassing operations emerge as consistent vote-winners. On the other hand, impersonal methods have consistently failed to produce cost-effective results, no matter how you slice the data or which populations researchers examine.

Quality counts: field operations’ "knock" numbers don’t tell you much

Volunteers from the Yes campaign speak with a voter as they canvass for support for the Yes vote in the Pilton area of Edinburgh on September 16, 2014 in Edinburgh, Scotland. (Matt Cardy/Getty Images)

Given the widely acknowledged importance of a good “ground game,” campaigns like to tout statistics that show they’re knocking on huge numbers of doors. These statistics can make their ground games sound quite substantial.

But, in reality, large “knock” numbers often conceal lackluster ground games. Why? Campaign operatives often rush through neighborhoods, hurrying to rack up impressive numbers of “knocks.” However, these hurried efforts often fail to reach most voters at all and entail only perfunctory interactions with the voters they do. Campaigns’ ground games can thus sound sizable in terms of “knocks” when they haven’t had any conversations with voters at all.

And, to actually affect voters, research shows that having an actual conversation is crucial. Canvassing seems to work best when voters who don’t care much about politics engage in a genuine conversation about why voting is important. So, when canvassers rush through scripted interactions, just trying to cram their message into voters’ minds, the impacts they leave are minimal — voters might as well have been sitting through a television ad. On the other hand, research has consistently found that authentic interpersonal exchanges usually have sizable impacts.

But facilitating that breed of genuine personal outreach isn’t what many “field” campaigns actually do. Green has seen this in practice. He has found that many canvassing operations have effects “smaller than what we obtained from our initial study or in our follow-up experiments with seasoned groups such as ACORN.” But, Green went on to say, “When I’d inquire about the details of these sub-par canvassing efforts, I would often discover that the scripts were awkward or that there was limited attention to training and supervision.”

This suggests a picture that should frighten candidates, campaign managers, and donors alike. Even if field operatives have racked up millions of “door knocks,” when one looks under the hood of these operations, there often isn’t much reason to believe they’re having many quality conversations with voters at all.

Voters aren’t seeing the ground game

Another reason to doubt campaigns are running good ground games? Voters don’t appear to be seeing them.

A political organization running a field campaign shared data with us that helps quantify just how invisible the ground game is to the very voters supposedly being inundated with it. (The organization requested anonymity when making public their internal research findings.) During October 2014, this organization ran a field campaign in a hotly contested Midwestern gubernatorial race. According to most accounts, this gubernatorial race witnessed the same all-out ground game as other elections this year, and this organization should thus be thought of as only one of many blanketing supportive voters with personal conversations urging them to vote.

What this organization did allows us to critically evaluate how widespread the ground game was — it ran an experiment in which some voters it was targeting were randomly assigned to receive a knock on their door from organization field staffers while others, a “control group,” received no contact from the organization (but still received identical efforts from other groups). After the election, this organization conducted an ostensibly unrelated survey in which they asked voters in the two groups what campaign contact they recalled receiving over the last few weeks from any political organization.

The results? The “control” group who received no contact from this organization remembered getting a knock on the door from any campaign only about 21% of the time. But just one conversation at the door from this organization doubled that figure, to over 40%. With just one contact yielding such a large increase, it’s hard to believe the ground game in that race reached anything near saturation.

Source: Survey of voters targeted by campaign in midwestern gubernatorial race.

But what about for mail, phone, and television-based appeals? The numbers for these modes show exactly the opposite: voters were saturated. This organization found that around 9 in 10 voters it targeted recalled receiving phone calls, mailers, or seeing TV ads about the same election. The disparity between these numbers and the same figures for field raise questions about the idea that the ground game is already in full swing.

Source: Survey of voters targeted by campaign in Midwestern gubernatorial race in control group who did not receive canvassing.

(Data from the Cooperative Congressional Election Study reveal a similar disparity — a majority of Democratic voters in swing states in 2010 and 2012 recall receiving phone and mail contact, but almost no voters recall someone knocking on their door.)

Spending more on unproven TV ads than on canvassing

The same disparity between field and other techniques manifests in patterns of campaign spending. A recent investigation by the New York Times provides a window into how SuperPACs spent their dollars over the last three election cycles. The results are puzzling. Over 80% of these groups’ spending went to TV advertising, followed by mail and online advertising to round out about 90% of spending in total. Finally, in a distant fourth, came field work — capturing less than 5% of campaigns’ budgets.

Somehow, many campaigns aren’t managing to spend much more on the most effective form of voter contact than on radio.

But, even though campaigns spend a very large share of their budget on TV ads, the research on the impacts of TV ads doesn’t bear out the idea that they powerfully influence elections:

First, there’s little evidence supporting the idea that TV ads can mobilize voters to turn out. In 2008, Jon Krasno and Green exploited quirks in media market boundaries to measure the impacts of presidential advertising. Ryan Enos and Anthony Fowler have examined the impact of the presidential campaigns’ TV ads in a similar manner. The results? Voters who receive the heavy volume of TV advertising associated with presidential campaigns are no more likely to vote than voters who see barely any. When it comes to turning out new voters, there’s not much evidence TV ads are of much use. (One exception is a study by Vavreck and Green on Rock the Vote’s television ads, which found mild effects among young voters.)

There is similarly limited evidence that TV ads have an enduring impact on voters’ attitudes towards candidates. Yes, ask voters how they feel within hours of seeing a TV ad, and we sometimes see evidence that they’ve been swayed. But these effects usually fade quickly. In one study, Seth Hill, James Lo, Lynn Vavreck, and John Zaller examined the impacts of presidential television advertisements and found that their effects disappeared within days at most. Likewise, in a collaboration with the Rick Perry for Governor campaign, Gerber, Green, and James Gimpel and Daron Show found that Perry’s ads had a noticeable immediate effect but left no lasting trace.

Even if TV ads provide fodder for much punditry or look impressive in a focus group, there’s not much reason to believe they have lasting impacts on voters’ views or behavior.

Why aren’t more campaigns focused on having personal conversations with voters?

Campaign consultants, like Republican Mike Murphy (left) and Democrat Bob Shrum (right) typically make more when campaigns spend on ads than when they spend on field operations. (Alex Wong/Getty Images for Meet the Press)

We academics are still scratching our heads about this one. Here, we’ll mention just a few possibilities.

First, managing a canvass operation is difficult and requires considerable recruitment, training, and supervision. It’s a lot easier to write a check to an ad agency or mail firm.

It’s also gotten harder to raise a field army. Decades ago, a rich network of civic organizations — think churches, Elk Lodges, and labor unions — could supply ample volunteers for field work. But, as these organizations’ memberships have flagged, professionally-managed, centralized, DC-based groups with weaker grassroots ties have tended to take their place. As a result, knocking on millions of doors now requires recruiting tens of thousands of temporary field staffers or new volunteers. When faced with a logistical challenge of that scale, it becomes mighty appealing to write a check to an ad firm instead.

There’s also a more cynical possibility — campaign consultants urge campaigns to spend more on ineffective tactics because it boosts their bottom line. Candidates and campaigns rely on consultants’ expertise when allocating precious campaign resources. But many consultants take a cut of ad fees, making a healthy commission when campaigns squander their resources on TV. On the other hand, waging field campaigns tends to be a low margin business and thus prove less financially appealing for consultants to recommend to clients.

Campaigns can do better

As effective as high quality field campaigns are today, they’re likelier to get even better as the research improves. Successful turnout interventions also seem to have lasting impacts on individuals, leading them to become lifelong voters, as well as on their cohabitants. But to take advantage of these innovations, campaigns need to seriously increase their focus on field.

The good news is, the necessary financial resources for waging real ground games are already available — campaigns just have to spend their money right. Consider what would happen if campaigns diverted just some of the money they currently spend on TV towards field. Nearly $1.2 billion was spent on TV ads during the 2014 election cycle, capturing about a third of campaign spending. Imagine if campaigns diverted just 30% of that amount to field, for a $350 million ground game — many times more than the amount campaigns actually spent on field this year. Field operatives can often be hired for about $20 an hour (including overhead) and could have two high-quality 20 minute long conversations with voters every hour, for about $10 per conversation. That all adds up to a staggering reality: campaigns could have had a 20 minute conversation with every single registered voter in a state with a close Senate race — and still afford to blanket the airwaves with ads.

Waging a high-quality ground game isn’t easy — but no one said winning elections was. Before 2016, candidates and campaign consultants need to take a hard look at the science, lest the ground game take a back seat yet again. We may need to knock on their doors, too.

David Broockman and Joshua Kalla are graduate students in the Department of Political Science at UC Berkeley.

Update: An earlier version of this post cited research by Michael LaCour and Donald Green on the effects of canvassing campaigns on opinions about gay marriage. That research has since been discovered — by this post’s authors — to have been faked by LaCour.