Donald Trump has called his election a historic landslide, but it was anything but. Only two other presidents have been elected with smaller popular vote margins since records began in 1824. His edge in the Electoral College, while decisive, depends on less than 80,000 votes across three states (Wisconsin, Michigan, and Pennsylvania) out of more than 135 million cast nationwide. It was a very close election.

In a close election, there are a million reasons “why” it was close. Trump’s popularity with working-class whites. Turnout among the Democratic base. Campaign malpractice in the Midwest. Jill Stein. Millennials. Most are probably true in the sense they could move enough votes.

The Clinton campaign, however, has centered its why-we-lost narrative on the “Comey effect,” along with another outside factor, Russia’s hacking of DNC and Clinton campaign email accounts. The “Comey effect” refers to the impact of FBI Director James Comey’s October 28 letter to the House Judiciary Committee announcing the discovery new emails that appeared pertinent to their closed investigation of Clinton and his subsequent letter on November 6 that absolved Clinton (after millions of votes had already been cast early).  

Many people — most notably Trump and other Republicans — have scoffed at the claim that the letter changed the outcome of the election, suggesting that it’s a convenient excuse for a weak candidate who made some questionable strategic decisions. 

But the Comey effect was real, it was big, and it probably cost Clinton the election. Below, we present four pieces of evidence demonstrating that this is the case.

When we began looking at the data, we were skeptical that Comey’s intervention was decisive. Politicos are notoriously prone to attributing election outcomes to gaffes and other oversimplified causes. It was once posited that a single awkward scream cost Howard Dean his shot at winning the Democratic primary, that the Willie Horton ad destroyed Michael Dukakis, and that the notorious “47 percent” video from 2012 caused Mitt Romney’s loss. Research since has debunked the idea that these incidents were decisive factors. In almost every case, the effects of supposed “game changers” tend to be smaller than broader structural factors, including the state of the economy, the popularity of the incumbent and how long a single party has held the White House. 

But Comey’s letter is unique for a few reasons. First, it was an intervention by an institution that Americans have largely perceived as nonpartisan. (Indeed, the FBI actively works to foster that image.) Second, the intervention was almost perfectly timed to impact Clinton at the worst time — dominating the final week of campaigning as an unusually large number of undecided voters made up their minds. Finally, it aligned perfectly with the narrative pushed by Trump —  and bolstered by the media’s obsessive coverage of how Clinton handled her State Department email, and the slow-drip release of hacked emails — that Clinton was somehow fundamentally corrupt. 

Understanding what happened in 2016 is crucial to understanding how to move forward, as efforts to reform the Democratic Party will be largely based on the stories the party tells itself about its defeat this time around — and those stories will also shape narratives about future presidential contenders.  

Exhibit 1: the state polls

First is battleground state polling data from the late stages of the campaign. One of the reasons the outcome on November 8 was shocking for so many is Trump notched surprise victories not only in purple Florida and Ohio, but also supposedly stalwart blue states like Pennsylvania and Wisconsin. 

Looking back at the widely used RealClearPolitics polling averages shows there was actually a lot of variation in the errors across states. As others have pointed out, one key factor affecting the size of errors was the share of voters in a given state who are white without a college degree. Much of what pundits are describing as error, however, might better be described as the “Comey effect.” States that showed some of the biggest surprises — where Trump outperformed polls — also tended to be states where there were fewer polls toward the end of the race. 

In other words, pollsters didn’t get things wrong so much as fail, in some states, to speak to voters after a decisive event had taken place. RCP underestimated Trump’s margin over Clinton by less than a point (0.8), on average, in the seven battleground states where five or more polls were conducted between October 28 and Election Day. But in the other seven states, where fewer than five polls were conducted, the polling averages undershot Trump by 2.7 points. 

Take Wisconsin, where Trump beat his polling average by more than 7 points: only two of the polls included in the final RCP average were conducted entirely after Comey’s letter was published.

To be sure, the gap with Clinton was narrowing before Comey dropped his bombshell, but the pace also picked up significantly after that. For example, averaging across 14 battleground states, the race moved 1.1 points in Trump’s direction in the week following the third and final debate — but Trump gained an additional 2.4 points after October 28.

Exhibit 2: the national polls

The effect of Comey’s late intervention into the election is also clear in the national polls. As neuroscientist Sam Wang showed, Clinton’s margin over Trump falls dramatically in national polls directly after the Comey letter and never recovers. At the time, statistician Nate Silver noted that the Comey letter coincided with “a swing of about 3 points against her” — a massive swing in a tight election. These public polls are supported by internal polling from both campaigns suggesting that Comey was a massive blow to Clinton at a pivotal moment in the election.

It’s worth noting that Comey also made headlines in July, when he testified in Congress about Clinton’s email server and then announced he would not charge her, while at the same time declaring her behavior “extremely careless.” In the words of Nate Silver, “That period produced about a 2-point swing against Clinton.” In other words, every time Comey and emails were driving the news cycle, Clinton’s national polling numbers took a significant hit. 

Exhibit 3: The early voting numbers compared with the late deciders

Early voting numbers are also suggestive of Comey’s impact on the race. Take for example a noncompetitive blue state like Rhode Island, where there was no contest but the presidential race. In 2012, Obama’s margin between absentee and Election Day ballots was similar, and Obama actually performed about 5 points better on Election Day.

In 2016, Clinton pulled in the same margin with absentee voters: 60 percent compared to Obama’s 61 percent. But something rather remarkable happened on Election Day — her support collapsed, dropping by a net 13 points. In Florida, Clinton won the early vote 52 to 48, but Trump won the Election Day two-party vote 56 to 44 percent. 

Steve Schale, a Florida political consultant who managed Barack Obama’s 2008 campaign in Florida, found this difference between early vote figures and Election Day ballots even more stark in the swing I-4 Corridor that stretches across central Florida, from Tampa to Daytona Beach. Here, Schale found that Clinton won the early vote with 56.3 percent of the two-party vote, but only won 47.3 percent of the Election Day vote — a highly unusual gap suggesting a pretty significant late surge. 

These early voting numbers are supported by data on “late deciders”: voters who chose their candidate in the last week tilted strongly towards Trump. And no media event was more important in the late stages of the campaign than Comey’s letter, which suggests that the most plausible explanation for the strong break in late-deciding voters was that letter.

Exhibit 4: media coverage of email, email, and more email

The Comey effect dominated media coverage in a way few events did during the campaign, other than Trump’s famous “grab ’em by the pussy” Access Hollywood video. During the final days of the election major newspapers “published 100 stories, 46 of which were on the front page, about or mentioning the emails.” The tone and tenor of coverage shifted markedly against Clinton in the closing week of the campaign.

Coverage of Clinton’s emails eclipsed her policy proposals and ended up being the only story about Clinton that stuck with voters. While 79 percent of registered voters had heard “a lot” about Clinton’s emails, only 23 percent heard “a lot” about Trump’s housing discrimination, 27 percent heard “a lot” about the Donald J. Trump Foundation’s illegal political contribution to the Florida attorney general, and, surprisingly, only 59 percent had heard a “a lot” about the Hollywood Access tape. The word clouds below show, in graphical form, that emails were the central way that most voters understood Clinton:

During the entirety of the general election campaign, June 7 to November 8, Gallup found that Clinton only sustained a “lead” in media coverage, meaning more Americans were hearing about Clinton than Trump, four times. Two were email related: FBI Director Comey’s press conference in late July, in which he called Clinton “extremely careless,” and Comey’s server-related announcements in late October and early November. (The others occurred during Clinton’s bout with pneumonia and during the party’s convention.)

The Shorenstein Center found that negative coverage of Clinton’s campaign was fueled by allegations of “scandal.” As the chart below shows, “scandal” coverage toward Clinton peaked in the final week of the campaign, consuming more than a third of her coverage. The timing was perilous. 

The upshot:

It’s true that there are other possible explanations for a late shift in vote intentions, but thus far there is no alternative explanation of merit. (The cyberhacks were surely important, but their effects would have been felt more steadily throughout the campaign.)

Instead, the evidence is clear, and consistent, regarding the Comey effect. The timing of the shift both at the state and national levels lines up very neatly with the publication of the letter, as does the predominance of the story in the media coverage from the final week of the campaign. With an unusually large number of undecided voters late in the campaign, the letter hugely increased the salience of what was the defining critique of Clinton during the campaign at its most critical moment.

The appeal of big-picture narratives about demographics, along with anecdotal evidence of big mistakes by the Clinton campaign in certain key states, makes it easy to point fingers. But looking specifically at the three “Rustbelt” blue states mentioned at the beginning of the article, no unifying picture emerges. Most stories mention Michigan, where Clinton didn’t campaign, rather than Pennsylvania, where she campaigned intensely. Indeed, these three Midwestern states (Wisconsin being the third) provide essentially an A/B/C test of different campaign strategies — and in each state she came up just short.

We do not intend to exculpate the Clinton campaign — in hindsight many decisions were flawed — but rather to note that the decisions were not abnormally bad (all campaigns make errors, and Trump’s made far more than others). However, the historic intervention into the election by James Comey means three major things:

Use caution when drawing lessons from 2016

Academic research will eventually yield important findings, but there is the potential for Democrats to overcorrect following this historic presidential loss. Introspection is important — and while still early, it’s already underway — but understanding exactly what led to the loss is foundational to understanding how to move forward. Lessons should draw from a broader pool of data than the results of the extraordinary 2016 election.

Democrats cannot rely exclusively on the presidency

Democrats must focus down-ballot where the problems are more acute (for instance, failing to run candidates in winnable elections). Because presidential elections are so variable and are so strongly dependent on outside forces, the low-hanging fruit for the Democrats is focusing on organization and mobilization down-ballot.

Something disturbing happened in 2016

Along with the Russian-linked theft and publication of emails from the Clinton campaign and the DNC, the Comey effect is of a different category than the usual investigative reporting or opposition research that campaigns have to contend with. Comey broke a decades-long norm of not intervening in presidential elections. The fact that his interference alone almost certainly swayed an election is indicative of a broader and disturbing breakdown of political norms.

Sean McElwee is a policy analyst at Demos. Following him on Twitter: @SeanMcElwee. Matt McDermott is a senior analyst for Whitman Insight Strategies (@mattmfm). Will Jordan, a former elections analyst at YouGov, is currently studying public policy at Columbia University (@williamjordann).

The Big Idea is Vox’s home for smart, often scholarly excursions into the most important issues and ideas in politics, science, and culture — typically written by outside contributors. If you have an idea for a piece, pitch us at thebigidea@vox.com.