This Tuesday, American voters will face an unprecedented day of primary voting. More than 20 states—and American Samoa—will hold caucuses or primaries. Major news organizations are preparing for a very long night, with results pouring in from American Samoa to Alaska. Delegate victories are likely to be determined by previously obscure party rules and thin margins in individual congressional districts.
And no news organization is larger than The Associated Press. Their data and decisions will set the pace for everyone else’s coverage. We asked Senior Managing Editor Mike Silverman how the AP plans to keep on top of Tuesday’s challenge.
Clint Hendler: What’s different this time?
Mike Silverman: Last time, it was nothing like this; you have 24 states, all voting at the same time, and so early. It’s taken on the air of a national election, and our planning for it is on the same scale as the November general election. We’ll have similar numbers of people involved.
Our Washington bureau is our headquarters for our presidential election coverage. I just talked to Sandy Johnson, our bureau chief there. She’s going to have about 40 people that day and night working on the election. I asked her how many would she have in November. She said about 50. So it’s on that scale.
CH: Maybe this is an impossible question, but do you know how many people will be involved in the AP’s coverage?
MS: Well, in the Washington Bureau alone there will be 40 or so, as far as editors an reporters in New York, we’ll have about a dozen, because we’ll be looking at some stories from some of the individual states—although our main focus will be on the over-all national picture, and that will be done out of Washington. And in every state where there is a primary there will be half dozen or more AP staffers involved in some way, from the bureau chiefs who are working with our Washington people to call the races, to somebody who will be writing a story on the exit poll results that we get from the particular states, to political reporters who will be writing about the outcome of the races. That’s just AP staff. Then we have literally hundreds of stringers who will be fanning out around the country getting vote counts, and then a smaller, but still large, number of vote entry people who will be taking their phone calls. So it’s quite an operation.
CH: The AP is a client-based organization; the networks rely on you, a lot of papers rely on you. What do they need, exactly?
MS: Some newspapers can buy the exit poll material, but they’re not intimately involved in the consortium, so many of them will be relying on us for the exit poll analysis as well. They have to pay a substantial amount of money to subscribe to the exit polls, and many of them are happy to rely on AP to describe the main points.
CH: And what are you expecting that the networks—which is how most people will be getting this news—will want?
MS: The networks don’t rely on us for the exit polls, because they are equal partners in the consortium. But they rely on us to quickly and accurately report the vote returns as they come in from various states. There is no official apparatus in the government that does that. Each state will eventually release a vote count, but everybody wants to know as soon as the polls close—for instance, in Florida on Tuesday night. Everyone wanted to know how many votes was McCain getting, how many votes was Romney getting. AP answers that question.
We have stringers who go to every county, even on the precinct level in some places, and as soon as votes are posted—partial returns and then later—they call in to an AP vote entry clerk, and the vote tallies are kept running in our computer system, and made available to our clients and customers to see as they’re reported. So we do as close to a real-time running count of the actual returns as is humanly possible. Now, you multiply that by 24, this Tuesday, and we’ll have hundreds of people in various data centers across the country—from Spokane to New York City, and the DUMBO center that we have near the Manhattan Bridge. And we have numerous systems experts who oversee the operations to see that there aren’t problems, and, if there are, to address them.
The networks do not try to independently collect the votes—it would be a huge, huge waste of resources and duplication. That’s the first and foremost thing that everyone relies on us for: a quick, running tally of the raw vote as it’s coming in.
Beyond that, our call, when we make it, will be something that the networks are looking for. I don’t know if you are a cable news addict, but
CH: …usually you guys call first.
MS: Well, not always. If we do it, we feel a competitive pride when we’re sure we’re right and able to do it, and then it’s always fun to see others react and match us. But there have been times when, for whatever reason, we’ve felt we needed a little more time—and we’ll take it. So we’re not afraid to get beat, either. If there’s any question in our minds, we’d rather get beat than be wrong.
CH: When you make that call, is there actually a statistical confidence interval that something has to edge over, or is it more that you look at everything in the stew and you know it when you see it?
MS: As Sandy Johnson would put, it’s a combination of journalism and science. The statistical models are there and we pay a lot of attention to them. And we don’t call a race until we’re confident. So if it’s close, if it’s within a few percentage points, we’ll wait until we see the trend and make sure there aren’t any unanswered questions or big gaps that might leave us surprises. And this year, turnout is very heavy, setting records in some of the early races, so we’re obviously keeping that in mind very much, since the results will differ in some ways from past elections. If there are more younger voters, that can affect things. All of that we take into account.
CH: So it’s part statistics, but there are more moving parts than the statistics can hold.
MS: Absolutely. And that’s why the role of our state bureau chiefs is key. They live there, they know the make-up of the different counties and areas of their state, and the past voting patterns. It’s just helpful to have someone who’s familiar with the territory first-hand, from living there.
There’s also the other aspect, the exit polls, which you didn’t ask about, which are an incredibly important aspect of our day and night—of our election operation. We’re involved with the networks in a consortium that conducts the exit polls that allow us to both get an early sense of how the races are going, but also provides really important and interesting data about the issues and the demographic and racial and income breakdowns of the voters and who they supported, and why, and all of that stuff.
CH: When will you actually first start seeing exit poll data?
MS: The procedures are now that each organization has its analyst in a quarantine room for much of the day. They are seeing stuff, but are hermetically sealed from the rest of the world, and there have been really extreme measures to try to guard against early leaks. When there’s a primary where the polls open at 6 or 7 a.m., we start getting some data by early afternoon.
CH: And that’s demographic data?
MS: It’s everything. It’s interviews with voters leaving polls from the morning wave of voting. But it’s early and they often don’t reflect the final outcome. Those get added throughout the day and the late afternoon and early evening. But the idea of the quarantine was to give our people early knowledge, but also to make sure that it doesn’t get out to the rest of the world. On a typical day, the quarantine room is sprung around five o’clock, and then the fellow who heads up our exit polling operation will come into the New York office, if the quarantine room is in New York, and start doing an early story using the information about issues, but under the understanding that there will be nothing put on the wire that reveals how people are voting, who’s wining.
But there are things you can write early on, like “people said the economy is the most important factor,” etc. When the polls close, that exit poll material is available for us to use. And in some cases, in some states, the exit polls will indicate some sort of landslide or blowout vote.
We don’t just look at the exit polls—they’re cross-checked against past voting trends and a lot of other factors. But there are races where we’re able to make a call as soon as the polls are closed, before any actual votes are counted. We did that for Hillary in Florida. We did it for Romney in the Nevada caucuses, and Obama in South Carolina, because the exit polls throughout the day consistently showed a two-to-one margin over Clinton and Edwards combined. You know, we were confident enough of the validity of that within some percentage variation that we called it. But more typically, the exit polls become a factor, and we start looking at the raw vote and then the exit poll material gets weighted in with the actual vote returns. So they’re very useful, but only in extreme cases are they the sole basis for us calling a race.
CH: You’re looking at 24 states. When you cross reference the exit poll data with past performance and demographics, how are you keeping a handle on all that past data?
MS: The past data is fed into the computer. It doesn’t have to be done by someone looking up tables in a manual from 2000 or 2004. It’s in there and the people who run the exit poll operation have the data in the system so that it’s kind of weighted against the current exit poll results, as is the real-time vote returns that are coming in. I’m not an expert on the details of the statistical models that they use, but they’ve been tested and refined over time, so they are a pretty helpful guide to someone who knows a little bit about what they’re doing—they look at it and they can see if the exit poll is proving to be accurate and if the returns that are coming in are consistent with past voting trends.
We look for red flags, to see if there’s something that seems really off from either the exit polls or from the early returns that are coming in. Our New Hampshire bureau chief has been up there for several election cycles and that was very helpful for us to be able to call it for Hillary Clinton, which we were the first to do. That was based largely on us knowing where the votes were still out, as well as where they were in, and being able to calculate the missing votes. There was some concern that Obama was going to do very well in some of the college towns, and those votes were late coming in. But when Clinton piled up big margins in the cities in New Hampshire, we were able to pretty much calculate that there couldn’t be enough outstanding votes to make up the difference. That’s just one example.
CH: With so much attention on the delegate count, it’s not states that are important. States are, of course, somewhat relevant still, but so are congressional districts.
MS: Before an awarding of delegates can take place, we have to know what the vote is. And it varies from state to state, but the awarding of the delegates that are up for grabs will partly depend on the statewide vote, and then, as you say, you look down at the congressional district level and there is some proportional representation. That’s why in Nevada, in the district caucuses, Hillary got more support—but Obama, we calculated, will end up with one more delegate.
So we have an analyst who will have a group of people working with him in Washington, who will be focused simply on the delegates. We expect, though, that our estimates won’t be completed until sometime the next day, because of the volume.
CH: So you don’t think you’ll be able to provide delegate estimates in real time?
MS: No, we will. But some elections won’t be known until early Wednesday. In California, the polls won’t be closed until eleven o’clock. Colorado, Montana—caucuses—could end at midnight. It will be sometime on Wednesday before it’s all wrapped up. But we’ll be doing it in as close to real time as possible.
CH: So you will have someone who’s watching vote returns in primary states and doing conversions to delegates?
MS: Stephen Ohlemacher, who’s our staffer in our Washington bureau who’s kind of our demographics reporter in normal times, he’s been doing this all along, and he’s been steeping himself in the rules and in the procedures because they are tricky and they’re different. And he’ll have people working with him.
CH: So what more can you tell me about what he’s doing to prepare? Getting election laws and party rules in binders from 24 states?
MS: That’s really it: making sure there aren’t any surprises. He also had reported last week that no matter how the delegate allocation goes next Tuesday, it’s impossible for any one candidate in either party to really nail the nomination that night. There is no such thing as a sweep because of the proportional breakdown, especially all the democratic states, which are proportional. Most of the Republican states are. There are huge numbers of delegates at stake, but not enough to get anyone to the magic number that night.
CH: Who has been going and getting all the various state and party rules and collating them?
MS: We have a research arm. We have several staffers who spend much of their time talking to the Secretaries of State and getting the party rules and regulations and making sure that they are kept apprised of changes, because sometimes there are changes at the last minute. There’s a whole staff of people who do that.
CH: That’s been going on perpetually?
MS: It’s a year-round operation now. We have to build these systems and update them. A lot of that work is done when there’s a down period. We don’t really stop.
CH: Have you done any dry run or mock exercise?
MS: We’ve been doing smaller versions of the process all campaign season. But this Saturday there will be an all-day, full-scale test of everything. As we always do, before the big elections, we’ll be mimicking the progression of the day, with actual stringers calling in with dummy results, and making sure everything works, and that the people who are supposed to show up at the precincts actually show up. We try not to leave any thing to chance.