On April 23, Hillary Clinton, with a net gain of 214,000 Pennsylvania votes in her back pocket, set off an election firestorm by claiming she had slipped ahead of Barack Obama in the popular vote total.
“I’m very proud that as of today, I have received more votes by the people who have voted than anybody else,” she told a rally in Indianapolis.
Journalists immediately fact-checked the statement, pointing out that Clinton’s math relies on counting votes from Florida and Michigan. (Of course, the DNC has stripped those states of their delegates for scheduling their primaries before the party’s calendar allowed, and both Clinton and Obama had agreed to not campaign in either state. Furthermore, in Michigan, Obama asked that his name be removed from the ballot, which naturally kept him from earning a single vote.) Even The New York Times, in a headline, gently called Clinton’s claim “New Math.”
But whether or not Florida and/or Michigan should be counted is only the most obvious pitfall in determining who’s actually winning the popular vote. News audiences—and superdelegates—want to know the popular vote, a simple number that in almost any other election cuts through the intermediation and let’s you know who’s winning. Is that too much to ask?
Well, maybe. The Democratic party’s nominating process is a kaleidoscope of caucuses, conventions, and primaries, sometimes all in the same state. And there’s no obvious best way to estimate a popular vote from it all.
The biggest difference, of course, is between caucuses and primaries. Not only are caucuses low-turnout events (you must be in the room to participate, and they take place in a limited time frame), but four caucus states don’t report the individual caucus-goers’ preferences. Together, those states have a population of more than 11 million, and there is no precise information on how individuals “voted” at those caucuses.
And into this breach steps The Estimate—and a certain amount of disagreement: CNN, MSNBC, The New York Times, and Real Clear Politics all use different strategies to attempt to count these caucus “votes.”
The one thing they all agree on is that, regardless of the strategy, it’s nowhere near an exact science.
“This is an attempt to tally one person, one vote, and to try to get as close to a real number as possible,” says John McIntyre, managing editor of Real Clear Politics, a Web site that’s become a go-to reference for superdelegates, votes, and other election arcana.
“This popular vote thing? The only reason we’ve succumbed to it is that I’ve talked to superdelegates that want to know the total,” says Chuck Todd, NBC’s political director. “But we’ll never get an agreed upon total.”
“We came up with a number. No, not a number, but an estimate,” stresses an off-air member of CNN’s political unit, who was made available on the condition he not be named.
Again, four caucus states—Iowa, Nevada, Maine, and Washington—don’t report raw vote totals. Instead, they report delegate equivalents. These are not the pledged delegates that will be voting at the Democratic convention in Denver. They are low-level delegates—the closest number we have to the raw vote—who will go on to further caucuses at the state, county, or congressional district level; they are the first step toward determining who gets what in Denver. Where the delegate equivalents are selected varies from state to state: in Iowa and Nevada the caucuses take place in every precinct; Maine meets at the town level; and in Washington the first round of caucusing takes place at varying levels of jurisdiction, from whole counties to chopped-up state legislative districts. The delegate equivalents reported from these contests are the only party-sanctioned numbers indicating how strongly a candidate performed.
“These numbers they spit out correlate, to some degree, with how people voted,” says McIntyre.
But exactly how well?
These delegate equivalents are derived from the preferences of individual caucus-goers, or the raw vote. But before becoming delegate equivalents, those votes must essentially first pass through a complicated system of filters that can reflect geography, population, local voting history, and the national party’s requirement that, if possible, any candidate who garners support from more than 15 percent of voters at a caucus site get at least one delegate.
Despite that ambiguity, the delegate equivalents allotted in these four states are the only official numbers out there. As such, they form the basis of most press estimates of the popular vote. How?
Here’s comes a lot of math: take the total number of caucus-goers, and split it up using the percentage of delegate equivalents earned by each candidate. So in Iowa, 37.6 percent of the delegate equivalents went to Obama, about 29.8 percent went to Edwards, and 29.5 percent went to Clinton. The Iowa state party claims approximately 239,000 caucus-goers. Mash those numbers up, and Obama gets 89,836 votes, Edwards gets 71,103 votes, and Clinton gets 70,433 votes. That’s how CNN, Real Clear Politics, and the Times handled Iowa and Nevada, which use nearly identical caucus systems.
But not Chuck Todd at NBC. In Iowa and Nevada, he hoped to avoid the imprecision of the delegate equivalents—especially the 15 percent rule and supporters who switched their choice after their preferred candidate fell under that bar at their site—by instead using percentages from the National Election Pool’s entrance polls, data gleaned from questioning a hopefully-representative sample of people on their way into the caucuses.
Everyone I spoke with, besides Todd, dismissed NBC’s entrance-poll method. “We didn’t do that because exit polls have been terrible this election, so I don’t know why the entrance polls would be any better,” said Real Clear Politics’s McIntyre.
“You’ve got an estimate based on an estimate,” said CNN’s number cruncher.
But as it turns out, in Iowa and Nevada, the methods make surprisingly little difference. By my math, Todd’s approach yields these Iowa numbers: Obama 83,651, Clinton 64,506, and Edwards 59,988.
Amazingly, that means the Clinton-Obama spread in Iowa—which is, of course, what people now want to know—is, by happy accident, nearly identical by either method: about 19,403 by delegate equivalent and about 19,145 by entrance poll.
Nevada is a little different, but not much. Again, by my math, Todd’s entrance-poll method would yield a Nevada spread of 8,082 in Clinton’s favor, the delegate-equivalent method gives Clinton a 6,040-vote advantage.
That closeness means that both methods are, if you’re optimistic, equally good, or if you’re pessimistic, equally bad. Both raise obvious red flags. But absent the real votes, we’ll never know how close the estimates track to what actually happened.
And unfortunately, Washington and Maine don’t give us any more data to test that question. Todd’s entrance-poll method is useless in those states, because there were no entrance polls.
Washington’s Democratic party claims that 244,458 people caucused, resulting in 21,629 delegate equivalents for Obama and 9,992 for Clinton. In this case, Todd simply added the delegate equivalents to the each candidate’s overall popular vote tally, as if they were raw votes. This means that—poof!—more than 200,000 Washingtonians go missing in NBC’s count.
The Times and RCP stick with the delegate-equivalent method, which, by my math, produces these popular-vote estimates: Obama 165,058 to Clinton’s 76,246—a spread of nearly 89,000 votes.
CNN introduces another Washington wrinkle. Ten days after the Washington caucuses the state held a primary—of sorts. Washington Democrats did not use the primary to allot delegates, but the Democratic candidates were nonetheless listed on the ballot. CNN uses this pseudo primary contest, rather than an estimate based on delegate equivalents, in its popular vote tally. That beauty contest had almost three times the caucus turnout, but was considerably closer: 691,381 votes were cast, but Obama only amassed a 38,386 spread. That tally nets 27,000 more votes to Obama than Todd’s, but 50,000 less than the Times’s and RCP’s counts.
Finally, to Maine. The state Democratic party claimed about 44,000 caucus participants, which produced statewide delegate equivalents of 2,079 for Obama and 1,397 for Clinton. Todd, again, just adds them in as one-to-one proxies, giving Obama a spread of just under 700 votes. RCP was joined by CNN in using the old delegate-equivalent percentage method, producing, by my math, an estimated 26,387 popular votes for Obama and 17,574 for Clinton, for a spread of about 8,800.
What about the Times? Well, Maine, while still not reporting true popular votes, still gives more information that the other three states we’ve examined. Delegate equivalents and turnout numbers are available from each caucus site. The paper combined this data at the county level, did a delegate-equivalent percentage calculation for each of the state’s sixteen counties, and added them up to get these numbers: Obama 27,459 and Clinton 16,606—a spread of about 11,000. “It gets you closer,” says Amanda Cox, a graphics editor at the Times.
The difference between the Times’s county-calculated number and the other outlets’ statewide-calculation is around 2,200—again, not so big in the scheme of any attempt to estimate the total national popular vote. But as a percentage of total Maine participants, it is a big difference—about 5 percent. That big of a difference makes you wonder, again, how accurate the delegate-equivalent method was in Washington and or Iowa, where a combined 5 percent variation could account for 25,000 votes.
So using my math, and the method described to me by each outlet, here are my estimates of their popular vote estimates for Nevada, Maine, Washington, and Iowa:
Obama Clinton Spread NBC: 155,422 132,013 23,409 RCP: 334,953 223,970 110,983 Times: 336,025 223,002 113,023 CNN: 524,007 463,468 60,539
As you see, a few choices here and there have a big impact. And the choices don’t stop with Maine.
What about votes from caucuses in territories like Guam, American Samoa, and the Virgin Islands? What about Democrats Abroad, the party organization for expats that held a so-called “global caucus”?
Real Clear Politics and the Times include them. “I look at this like there are X-number of sanctioned contests where American citizens are voting,” says McIntire. “And the DNC has determined they are entitled to delegates.”
Not CNN and NBC. “It’s not going to tell you about their electability,” says Todd.
What about Puerto Rico, where pundits suggest a million voters could take part in a June 3 primary?
CNN remains undecided. The Times and RCP will include them. NBC’s Todd is less enthusiastic. “The superdelegates probably aren’t going to care,” he says, adding that he may include them, but segregate the results like Michigan and Florida, allowing viewers to indulge in a bit of choose-your-own-tally.
What about individuals participating in Texas’s caucuses, which took place just after voting closed in that state’s primary?
So far, no one includes them. The caucus was only open to people who, earlier in the day, voted in primaries. “You start to double count people,” says McIntyre.
But the Times’s Cox remains open to the possibility. It was a bit of a moot point at first because the state’s caucus data—compiled by snail mail—was incomplete. But now the numbers are there. “If you argue that votes that result in delegates should be counted, then maybe the Texas caucus should be included,” she suggests
And what to do about Idaho and Nebraska, which long ago allotted their delegates via caucus but will be holding their own nonbinding primaries in May? How different would including those numbers be from CNN’s decision to count Washington’s nonbinding primary instead of an estimate based on earlier caucus turnout?
Some of these choices would have a sizable impact on the totals, others are negligible, but all reveal interesting semantic and procedural differences between different outlet’s methods and philosophies.
“Once you start trying to do it, it’s much more complicated that you’d think it would be,” says Cox. “Just like so much in this election.”