Transcript for How the media projects the winner of an election | FiveThirtyEight Politics Podcast
Hello and welcome to the 530 politics podcast I'm Galen Rupp. One of the big questions heading into Election Day this year. Is not just who will win. But when we'll know the winner. Unlike past elections when would be at least went into election night expecting to know the results by the end of the night. Americans are being cautioned that it could take d.s world longer to get results this year. States have changed their voting rules in response to the corona virus pandemic and more people are voting by mail than ever before. How owning all those mail ballots will be a new task for many states that could take time. Also some states except mail ballot after Election Day so long as they've been postmarked by November Ferrer. And the data that we usually look to as an indicator of how wrapped up a race might be like the percent of precincts reporting. May not be particularly useful if it doesn't take into account mail ballots so in order to get a sense of what will know on election ad in the days after. I'd invited the executive director of elections at EBC news Dan Merkel to join the pod cast he oversees the decision desk act EEP c. Which is responsible for meeting steep price the election projections. Yet welcome to the podcast. Thank you for having me. And I'll also mention that as many listeners probably no 538 is part ABC news. So let's start with the basic stare what is the role of the decision desk on election night and potentially in the days after. So the decision desk is a group of experienced social sciences statisticians. In election experts. And our job is to look at all of the available data on election dale the statistical models. And project the winners of elections when we have a high degree of confidence that the leading candidate will in fact be the winner. What is the difference between protecting. The result of an election or race act calling a race. Well so at ABC news we use the term project. Because. It still is based on data and probability is so we don't we don't call winners we don't declare winners that's that's a little more definitive. So what we're doing is we're estimating based on all these Dana. When we think the leading candidate is strong enough it has a strong enough lead to be projected the winner but it isn't until all the data are counted and certified. That you actually have a have a real winner. And that's in large or ops secretaries of state around the country you said that when you have a high degree of confidence you can project a winner. How hot of a degree of confidence do you have to happen when are you ready to meet a projection on election. So is is a statistical criterion that we sat and it's 99 point 5% confidence so the models give us an estimate of that confidence. And so what that means just on chance alone. Is that you know 99 point 5% youth based on chancellor you could be wrong once I'm out of 200. Over the last twenty years we haven't had any mistakes we've done hundreds of projections. And that's because not only do we use the statistical models but we bring our own independent experience and judgment to it as well. So it. Order to jets of that 99 point 5%. What data or you booking. So we cattle a wide variety of data on election night. We first start out by looking at the exit poll data this is a survey of people who were interviewed as they leave the polling place. Combined with that her phone poles of early and absentee voters. In addition we also have exit polling of in person voters that are now happening right now leading up to Election Day. So that's all polling data. And that's data that we combine in the right proportion and get an estimate of of what things look like before poll closing time. If a race is. Not close of its it has a large margin in these data we can project the race off of this polling data and poll closing time. Apple closing time. Then we start getting actual votes in a two types of actual votes in a precinct level data. That is a sample precincts in a state. That are that are fed into precinct models and then separately we get the vote at every county across the country and those are fed in the county models. And so then all of these data you know it's combined with yes a poll date on the phone poll data the actual vote data them. Various models look at it in different ways and then we're looking for consistency across the models. And amber looking for good representation across the state. Do we have a you know to be up all the different parts of the state do we have a good mix of absentee. Vote in early vote in in phone. And Election Day vote in and so what we get what we have a really good mix of everything. And the model tells us. That is its projects of all that's what we'll do it but we could also have other situations where the model thinks it's ready some to be projected. But if it's only data from a certain part of the state or it's only election date date. Then we know based on that that we need to wait for a better mix of data. Part so there's a lot that I watch you on pack in all but let's begin with the exit polls because. As we talked about on this podcast and seen in past elections icicles and do you notoriously. Inaccurate it was the keys in 2016 for example that the early exit polls reflected the polling in the days before the election which suggested that Hillary Clinton was going to win that auction. So why rely on that exit polling Dieter and how heavily do you roll out. Well we we rely on it a fair amounts but like I said we would never project a race. Simply off the exit poll data if it's if it's a close race we we will wait for actual vote data. Then what does not quite a close race. Well you know if it's you know like a safe race which is double digits we we would project off of it but if you're talking. You know this year like a Michigan Pennsylvania Wisconsin anything like that Wii U we would wait for actual vote data before projecting that. I'm and so to answer your question about the accuracy of the exit poll it's like any survey it has sampling error and has not sampling error non response that kind of thing. We have seen over the years. That the eyes of poll. You know can't overstate a certain candidate some democratic candidates for example. This varies election by election in a very state by state precinct by precinct and we find typically. That if people with more or less education vote differently that could that could impact that's a poll estimates. Because we fight of people with more education tennis tend to be more likely thought the questioner. So after the 2016 election. At a send out though the company that does or exit polling. Has implemented and education. Adjustment in the gossip also starting in 2016. So we have that in 2018 and that was an improvement and will be using that again in 2020. As so that that gives us a little about our read. We had already been adjusting for age race and sex. Demographics in the exit poll but like it's in a 2018. But we've also started adjusting for education as well given the importance of education and in the 2016 election. How else are exit polls different this year mainly as a result of the pandemic and changes in voting patterns particularly an increased number of people voting by mail or Peruvian person. So that well the first thing obviously who is with the pandemic Addison has had to change some of the procedures in the exit polling to make it safer for people to participate. So typically exit poll interviewer world that they sample voters as they come out every fourth or fifth voter based on a pre set interval. And they approach a voter typically and they go right up to them and hand on the questionnaire and a pen and asked them if they participate in fill it out we cannot do that this year so instead. The interviewers are approaching from at least a six foot distance. Asking people to participate and then if they if they agree that they're attracted to a table that has the question there. It has single use a golf pencils so people have to use the same pencil. It has hands sanitized sir. It has sanitizing wipes and in a single use gloves for people and so in that way. Were able. Make sure that it's us assay procedure and Edison did test out this procedure in the primaries in the spring and found that. It it really was an effective way and people still were able to able to take the example survey at about the same rate as we've seen in the past. The of the other big change like you mentioned is because of the increase in male and early voting we've had to increase the percentage of interviews done on telephone. And Don are early in person in the early in person exit polls so so yes so in in 2000 and sixteen about 4042%. Of people voted absentee or early so we were still doing the phone surveys then. But we're expecting it could be 60% this year so as a result we've had to really increase the people for the interviews with people who are not voting on Election Day. Art so that's the exit poll data how is the other data that you collect in order to make projections different this year you mentioned precincts of course. That can be somewhat complicated write a fact that there are many mail ballots that may not be included in that precinct level data so how are you adjusting for all. So so at the precinct level. You're right a lot of and a lot of states the precinct data will only be. The Election Day vote. But in some ways that's good because when we get that model when those didn't commence that model we get a very good read on what the Election Day Porsche tennis. And then we can combine that with the telephone polling of absentee and early voters or the in person early voter. In addition to that though some states like Texas and Florida. They released big Tom's of early and absentee of data. Either right away or early in the vote count and so we can take a look at that as well and so. It's become more complicated over the years when it first are doing projections in 1984. It was it was like 9% of the vote was early and absentee so all the precinct models were very accurate. Now what we have to do was kind of triangle they. On what what the answer is so we'll be looking at the precinct model Election Day vote will be looking at the absentee vote that's reported. Early vice mighty states that break it out some states break it out better than others and legacy looking at the phone poles and the early voter exit polls. And and really just trying to piece it together that way. All right so there's a lot of moving parts there which states are you actually expecting you could our results from on election night. And of course you know I assumed that you'll be calling states like. New York for example org you know Kentucky when the polls close. But in terms of the State's overall going to be watching war you know competitiveness and who wins between Biden and trop. What states might we get results from reliable results auction. So you're right the safe the say states we expect to have. Either projections a poll close but on some states we don't have exit polls and so if it's a safe state we do not project to pull clothes that doesn't mean. Necessarily the best he's going to be close it means we're just gonna wait for some actual vote count data there. But but there are you know there are a number of states that have really good experience counting early and mail in vote. And depending on the margins. We should be able to make projections in those states now obviously if if it's very close Wilson have to wait for more data. But I'm thinking no states like Texas. Com North Carolina Georgia Florida. Those are states that we expect to counselor to in the past. Well of course those are also expected up close races so if it's a water two point race then yes we'll be waiting for more data. But if we do have a few point margin you know 345 points then those races should be project of all event other states like Virginia. New Hampshire. There are expected to have lower absentee and early vote but but also. As a result will be expecting that they'll be counting similar to how they've counted it in the past. And then you have states that are key states that could be more problematic this year Wisconsin Michigan Pennsylvania. These are states where are the in her the election I'm sorry the early voting and absentee will be quite a bit higher. And you know specifically Pennsylvania. And Wisconsin they're not going to be processing. The results before election they like and some other states and given the big increase. We expect it'll take longer to get those results there especially in the bigger companies and given that the key states and expected to be fairly close we could be waiting. Quite a while offer for a projections states like that. For slower states that we mentioned in the upper midwest when are you expecting results there. You know it it could be in the days following I just I have no idea like this and before we could be surprised and they could count sooner than we expect. But were weren't you we just we don't know because you're the one thing that that helps us normally an election years. Is that states are pretty consistent in content or vote and and companies are consistent in content to vote. Election to election unless there's been a big rule change and if there's a rule change mutate take that into account but pretty much we can look at past reporting in his state. To see what them the current reporting in this election will be. Here were here we don't have that option a number of states and so it's really is our expectations we just got to be more cautious and we have to be ready for anything so as. Fair to save it you're going into election night not expecting to call states like Michigan Wisconsin Pennsylvania art department. That that would be my expectation certainly we could be surprised. The margin could be bigger than expected. The states could count faster than expected so I wouldn't count out projections in those states. But but like it like anything you know it's hard to predict what states will do and it's hard to predict what the exact margins will be. But will be looking at a very closely and if we can make the projection election date based on all the data we have. And and knowing what's out then we'll do that and the other part of is you know we know we have but if we also know we don't have. That also may enable us into a projection. Before all the before a big chunk of some pieces of the RN it really just depends on the circumstances. Do you think it would be possible to project an oral winner on election night without results from Wisconsin Michigan and I think they'll be very difficult. I just you know again nothing's impossible and at the but. That that it seems likely that if we're not able project those we have to have to wait on the overall because the way we project the president. Is you know it it really is when someone gets a 270 electoral votes so it's not like we're modeling it and saying. Well we think we think one of the candidates will get there we actually have to have the State's projected added up to 270 in order to have a winner. So are you going into election night and thinking that you bought a project an overall winner on November 3. Or in the early hours of November 4. Well I guess that in I think it's a distinct possibility but I'm not that I'm not gonna say never because I don't want to set that expectation. You know we've we've seen unexpected things happen before so it's possible but not probable. So let's go back to the states where we might be expecting Fuller results on actual actually say the results from. Florida and Texas come and and obviously hypothetically Biden were to win those two it would seem like there's no path for president trump to. Win battle action would that still prevent you from rejecting the overall winner. Or if if Biden wins Texas has about a scenario in which you can probably project and world. It was at the 270. So so we're we're just vote we're just based on the vote the data in the projection in specific states adding it to seven. So we would not project a winner for the presidency but we believe it's of the pond and sense of the people talking on the air to explain here's what it means. In terms of you know we we have in projections for Biden or are trop in certain states what does that mean. For the overall outlook and that's kind of more of an analysis en route and a projection and so we believe that two other people in the news divisions and to make that analysis. Can we did a little deep. Enter exactly why states like North Carolina. Florida Texas were Georgia. Might be able to powerful results on election night that little are your project and. Yeah really what it comes down to his states like that have already had high numbers of early voting and absentee voting in the past. So first of all. They're used to processing that data quickly secondly a lot of those states. Allow. The the colonies to start processing the ballots. Ahead of Election Day. So that when poll closing time hits they're able to dump a lot of that voting in other states like I mentioned. Com Pennsylvania and Wisconsin are not able to start processing it before Election Day and so you've got that these big you know big stacks and big. A numbers of votes that then need to be process so a state that's processing for the week or two before or as they come in. Is gonna get a head start on on some of the states that are not allowed to do that it it varies widely by. Across the country by state. Are there particular. Regions or counties in those states that we could have full results for or close to fall results for. Relatively early that won't give you we're good sense that maybe you when your team starts to zoom in on and checked out in order to get a sense of how the races aren't. I think specifically. You know in some he stays where we're expecting it's maybe reveal slower specifically. We expect the smaller counties. To be able that process the vote and count the vote quicker. And answer oh yeah bully looking at that closely by region by county. And the other the other important part of the models that I I didn't mention is were were comparing comparing the precinct data. In the county data to pass data from recent races so that gives us a good indication we're things are changing. So we will get a sense of where things are going even if we can't necessarily per jacked. The race because we're missing say Milwaukee city or some errors some B area of the state. You know it is if if something big is missing we still may not be able to projected but we'll we'll have a sense I'm on how things are looking at least in indoor and the other part of the state that's report. And in say it Florida for example is there one county event when you are a result there you might know how that the state has honors that. A bit of a stretch. Yeah that's a stress we don't really look at it that way we we look at everything. You know all different regions all colonies. And and what the way the models worked the model stratified the State's instant instant geographic and party regions are party categories. And so really what we're looking for instead of specific counties is good representation within these provocations. And then good representation of the strata vacations across the state. And so there aren't situations where you can make a projection without certain quote unquote important colonies. Depending on how the rest of of the data has reported an in with the March Madness. So that's something along the lines of if you see that in Florida trump is over performing his numbers in rural counties by three points how to RD one. Florida and 26 team that would look like an election that's on track to have trumpet the runner for something like that. Right and and and that you know that comparison is very helpful. And in addition the models really it uses a ratio less than that to use the colonies in the precincts that are reported in the past data. To adjust. Can't estimate the colonies in precincts that have not report it so it's it's like Robert we're using as much data as possible to kind of I just four areas that we don't have data from yeah. All right so you are overseeing our you know elections coverage here act ABC news overseeing the decision to ask. What kinds of things are you worried about going wrong in Q election night or are you worried about out. You know even and it's our debt not something goes wrong in what may be stresses you router keeps you up. A lot of things keep me up at night and I would say for the last number of weeks I I've I've woken up every night save between 1:30 AM and 4 AM every night. And is in I'm thinking about something it's often something not that important but we are sleeping it's kind of seems important. And so the only way I can get back to sleep. Is I have a notepad by my by my bet and I write dollar taking so I don't have to keep thinking about it and M I'll wake up in the morning usually they'll be something actually is not that important. You know. I can't I don't have a lot of specific things that were worried about one thing. That we normally do and what we did this year is we start rehearsing trip testing and practicing. Really early and we normally start in August this year because of all these changes we started in July so we've done. We've been testing our systems rehearsing and practicing. Since July and this is gone on it's a September and October where we were hers and practice three and now for five times a week. And so I think from from that standpoint I think we're prepared for anything. That you know the team has been practicing we've been training. And so what we have no idea what to expect so just to give an example you know crazy things can happen in 2012. With hurricane sandy out here. Well you know I lost power in my house for for two weeks right before the election. That was unexpected. So what I have to do I had I headed to go to plan B and had to move in to New York City into a hotel because I couldn't do anything from home with no power. Com we lost our T one lines as a result of hurricane sandy that's those are the pipes the data pipelines that take in the data from Edison. And so the first of for a little bit of time that we had no way to get the data. Let's leave it back came back up that we had backed up BP MV PN lines. But again he has never know what what crazy thing will happen so we try to plan for every contingency and every scenario. And and did you know it in things will be a little harder this year with corona virus because some members of our team we'll have to work remotely. Just given the number of people conducted the in the room and so yes there could be technology issues and that kind of thing. But with all the practice we've been doing we think we have enough workarounds and backed up scenarios where. We don't we'll deal with the situation when it happens one thing. That I think a lot of newsrooms have discussed is cyber vulnerability. And how. Even if the results are secure that a cyber attack on the infrastructure that reports the results. Ranging from secretary of state web sites to actual news organizations like ours could cause misperceptions amongst the public. They could also do damaged even if again the results are secure Howard. You thinking about back in terms of you know you're getting data in from Addison as you mentioned you're getting data in from. The State's secretary of State's and then of course you know ABC news and I thirty we have our own cyber vulnerabilities as well so how are you two regional bank. Well we we do have really good technology teams both an ABC news and it had an at Edison research. And they kick the tires on all this we have different back up scenarios. For our systems. And for the Edison system so multiple ways to log in if if certain at certain ways a logging in have been compromised or are problematic. And in editions that the way Edison gets the county vote is really a decentralized method. So in many places what Edison does. Is sends actual people to the counties themselves. And so Edison is getting data directly from the company's now there are some counties where we take data feed and there are some states where we take its eighty feet. But we also have the option of relying on people on the ground. And so if a secretary of State's web site is hacked or compromise or the data feed is not usable. There are still people who are able to go to the county and then phone in the results Addison so in that kind of a decentralized situation we we would have confidence in the in those results. We mentioned at us and a couple times here and I just wanna make sure it is that our listeners understand what we're talking about what does Addis in. Do you. That's in the search is accompanied we've been using since 2003. And they collect the exit poll they conduct the example for us as long as well as the phone calls they mention. The also vote count the vote for a so let's let some people are people may not realize is that there is no national organization or national government and says he that counts the vote. Across the nation and that's why news organizations have to tabulate the vote. Ourselves and so we use Edison research deep into the vote tabulation so went on our website or. On our air you're seeing in a 62% of the expected vote in here is the vote. That is vote that Addison is collecting for us. In in and timely manner such that were able to report the results time in a timely manner in if it weren't for that we did it the vote count would be a lot slower. And so so as it does all that forests and they also need to provide us with background information on the election as well so we we rely very heavily on them. Each election than they do a great job for us. And do all networks use Addison. The four networks that use Addison are ABC news CBS. CNN and NBC. As well as a number of subscribers across the country and addition a Reuters and and a number of you know local subscribers but also other other bigger bigger Internet companies as well who doesn't use. So so the associated press and fox. They are both using AP Dana for this election and I I think for a 2018 as well. And so though those will be the two big ones that I'm aware of and I'm sure there are other local ones as well. So. Another funeral. Contingency that we are planning for in news rooms around the country. Is how president tropical reacts to the results he said that he thinks the results of the election must be reported on November 3. As we've discussed here that's simply not how elections work but there's some concern that the president may declare victory based on partial results. We're something else along those lines are of course we don't actually know won't happen and we can only echo off of what he sat. Is that something you're considering the decision dusk and would that affect you know your work at all. No that's not something we take into account in and it's not something we we would have Pate who you pay attention to it that decision desk certainly the news division would would would. Have you know be entrusted in that and they have to decide how to report that. But at the decision desk what were not influenced. By internal pressure or communications from within the news organization nor are we influence or paying attention. To external communications so what the only thing we're looking at. Is that data as its reporting on election night. The models and when we have confidence in it and so soap people externally can sane and do whatever they want. But that does not factor into how we project via the races. I watch you while I have you here on this podcast talk a little bit about. Past election you already mentioned 2012 in dealing with hurricane sandy. I think going in to use this election I a lot of people have the 2000 election in mind. And how difficult it was two. Project that race and Europe of course in particular the troubles at Florida. Were you working out the decision dust back then in 2000 and what was that like. Well yes yes I was an ABC news actually came to ABC news C news in 1989 so that was my first election working the decision desk at ABC. And it were I'd worked decision desk previously at voter news service for five years before that. But he I was it was in have. An interesting election to say the least. You know that then networked in the news organizations projected gore the winner shortly before 8 o'clock. A doubt was based on partially on exit poll data and some limited amount of precinct data they had come in. That that projection had to be retracted it looks solid at the time. And and you know in in in a retrospective. Post world that was done in a what what we found. Was that the exit poll overstated gore and in addition the model did not adequately take into account the absentee vote. Which was more for bush. And so one of the changes that that came out of that. And and debt service well in the last twenty years is that we are a lot more cautious projecting offered the exit poll. And we are a lot more explicit as to how the absentee votes are incorporated. Into the models and so as a result in last twenty years and the hundreds of races we projected we've not had a a recurrence of of that type of situation. Are there any other election nights that come to mind in planning for were. 120 that make you think like oh we definitely learn some lessons there you know the after all these decades are projecting elections. You know. This visit Weisel confident in twenty tort. Nothing really sticks out from a specific year. Since Florida. You know things have have gone pretty smoothly and there's nothing that's six out of something that was unexpected. Because really what it comes down to his every state is different. And in every election every state is different and that's the case again this time the -- that are different might be a little dependence than last time. But well so we don't just you know prepare overall we actually have to look at and prepare for every single state. And and that's what we've done in the past and that's what we're doing this time just based on what we expect to happen in in in the bank robber searched them that we have. I'm curious I mentioned today in 2016 the initial exit poll results showed Hillary Clinton favored to win. That election at what point was it clear that those whole lacks those exit polls were wrong and that the result was headed in Trump's. While sending to the the of the early adds a poll results. Had had Clinton up. You know fit in some of those midwest states but it you know it was it that those were not big leads in those were leads definitely within the margin of error of the ads have Paul. And so there it was within the range of you know anything could happen. But it yet wasn't until later in the night when. We started projecting some of the races for trump. That that Hillary was Hillary Clinton was leading in earlier. You know when the both started coming in and we saw. The small Hillary could lead in the exit poll become a small Donald Trump lead in the vote count is when we kind of had a sense that. Things where we're going another way. And the race for the senate is also our competitive. This year do you have thoughts on you know when you might be able to call those races are saying overall who's likely to take control of that chamber. Yeah it's it's it's the same thing with the presidential race that we will will only have a winner projected winner in the senate when we have. Enough projected senate seats. Or. You know. It's it's it's so it's exactly the presidential race in that in that we're not gonna say okay well the Democrats have flip these two. And so we can now project that it's it's likely that the take that the senate we do we just have to wait for the number of seats unlike in the house over a hospital we project that that is one. World we will project debt. Based on. Our early counting states. In what we expect to happen later Connors counting states so in the house races. We don't have to get to the exact number of scenes that is more of a projection because we're dealing with a lot more. House races then for example senate and in presidential races. Lastly here. For our listeners that are going to be I'm sure watching the results con man on election night or maybe not maybe people who listen to those accounts aren't actually he hit it going to be paying attention on election night. Although I imagine that's a very small number of people. What should it viewers keep in mind what would you advise. People to think about as they're waiting for your projections from the decision ask. Well I would say. Keep in mind that. You know anything could happen you know polls are in general pre election polls are a good indication. Of what may happen. But like any thing polls are not perfect and so. I would I would just before getting too far down a path of this person's gonna win or that's person is gonna win just really. Take a step by step and and and go where the projections lead us and so yeah I think he's state where that Donald Trump visit is knee is needing. Is projected refer referred Biden that may be a good sign for buying but we we need to see with the other states. What happens in the other states as well so I think you know I think. They just don't get locked into any. Any one scenario as what I would say. It just just wait for the projections to be made and he doesn't I say is if you're looking at the vote count on abcnews.com or wherever. Just realize that the early vote that's coming in May be very un represented oath of what it will be in this state. So if you seen trump or Biden to head somewhere that's supposed to be a close race. It could because it could be because. It's just the Election Day vote in and trump is doing very well there or could be is just the a mail in vote and -- doing well there. So again. Just you know not getting too excited about specific vote count results until there's a projection because with the projection we're looking to see that there is a good mix of data in. And and what's what may be reported on election especially early is likely not be represented of how that how how the state meet me turn now. All right well I think that about doesn't so thank you so much debt for sharing retirement Austin. All right thank you. Dan Merkel is the executive director of elections at BBC news my name is Dillinger Tony chow is in the virtual control room clear bitty Gary Curtis is on audio editing. You can get in touch by emailing us podcasts art thirty dot com you can also of course greeted us with any questions or comments. If you're a fan of the show leave us a reading or review and the apple podcast store war tell someone about us. Thanks for listening and wells. And I are. Yeah.
This transcript has been automatically generated and may not be 100% accurate.