Something stinks when exit polls and official counts don’t match
Media exit polls in last Tuesday’s election suggested Democrats were going to win the White House and the Senate, yet the reported vote counts brought a GOP landslide. While theories abound about what happened, election integrity activists say the exit poll descrepancy underscores the need for a far more transparent and accountable process. AlterNet’s Steven Rosenfeld interviewed Jonathan Simon, a longtime exit poll sleuth and author of Code Red: Computerized Election Theft and the New American Century. Simon explains why exit polls are a critical clue in the breakdown of the voting process.
Steven Rosenfeld: Let’s start by telling people about your involvement with election integrity and tracking exit polls.
Jonathan Simon: I’ve been working in this field which we call election forensics for about 15 years, since the 2000 election. Certainly things kicked in with the 2004 election and the exit polls there. I was actually the person who downloaded the exit polls that were left up on the CNN website which then made it possible to compare the unadjusted exit polls—and we can explain that in a bit—but comparing the exit polls with the vote counts and show through all those disparities that there was reason to suspect possibly manipulation of the vote counts.
It has deep roots and basically looking at every election since has found varying, but at the same time, fairly pervasive patterns of what we call the “red shift” and where the exit polls are to the west of the vote counts. We track that, we record it and we attempt to analyze it and get some sort of handle on what has caused it as a phenomenon. Then we look at all sorts of forensic data, accumulative vote share, tables and hand counts where we can find them. I’ve always been particularly conscientious about trying to take whatever baseline we’re using and validate that baseline, so that if we have an exit poll for instance, we try to make sure something that has been skewed by over-sampling one party or over-sampling people of color or something to that effect and validate it by that.
We try as carefully as we can. I’ve been doing this pretty steadily now for the last 15 years along with some of my colleagues, and I would be the first to acknowledge that there is a lot of smoke there and there’s a lot of probative value to this work, but that bringing it forth as ironclad proof is very problematic. So we’re stuck at a place where I pivoted to is looking at the risk involved in having a computerized, privatized, unobservable vote counting system and just taking on faith that that system is not being manipulated when there is such a obvious vulnerability (on which the experts strongly agree) of the system to malfeasance and manipulation. That is where I’ve tended to go, is to look at that risk rather than screaming fraud from the rooftops and claiming proof.
SR: Let’s go through this piece by piece, because it’s a lot for people to really understand. You get the raw state-by-state exit polls that are commissioned by a big consortium of national media organizations. What did you find this year, that happened this week? What do you see in the raw data?
JS: Of course, we don’t get the raw data. The raw data would be… we have three definitions here. There’s raw data, which is the actual questionnaires and the simple numerical toning up of answers on the questionnaire. That is never publicly released. It’s if you want to characterize it as such, it’s what’s inside the sausage of exit polls, and we are not privileged to see that. I’ve had one opportunity in my life through an inside source to actually look at some of the raw data, but that’s a very rare thing. It’s not generally accessible to the public. Many of us have clamored for the public release of that raw data, certainly in the aftermath of the 2004 election and have been denied it.
Then there is the weighted exit poll data and that’s what the exit pollsters put out as soon as the polls close. This has been demographically weighted to their best approximation of what the electorate looked like and it is very valuable information. That’s what I was able to download in 2004 and that’s what I was able to download in many of the elections since, and that’s what I was able to download this Tuesday.
Then you have adjusted exit polls and what happens is they take the vote counts as they come in and they use the term as the art of “forcing,” they force the exit polls to [be] congruent with that vote count data so that by the end of the night or by the next morning when you have your final vote counts and final exit polls the exit polls and the vote counts will match, but that’s only because in essence they’ve been forced to match the vote counts.
SR: I’m looking at the New York Times website right now, at its election 2016 exit polls interactive. What are the totals then that I’m seeing?
JS: I’m not looking at the New York Times. I’ve pulled these off of CNN and I’m also looking at MSNBC. Because the firm that does this, Edison, contracts with the consortium of major networks and then has some lesser clients such as the New York Times. When I say lessor, they’re still very major clients, they just don’t have the prime membership that these five networks and the AP have, but all these major clients get the same feed of weighted exit poll data.
What you’re probably looking at now would be adjusted exit polls and they’re very close to, if not congruent with the vote counts. But if you had looked up Tuesday night, for instance, if a poll closed at 7pm Eastern Time and you had gone online to a network site at 7:01pm Eastern Time, what you would have seen at that point was a weighted poll that had not yet been adjusted to match the vote counts. They would tell you the number of respondents. They’d give you all the cross tabs, by which I mean broken down by gender, age, income, party affiliation, usually 30 to 40, sometimes 50 questions … Pretty detailed stuff that indicated how each subgroup of the polled population had answered these various questions.
Some of those questions are demographic questions: What is your race? What is your income level? What party do you identify with? Who did you vote for in the last election? etc., etc. … Then there are the current choice questions. Who did you just vote for this evening and/or this afternoon? Those are all presented in sort of a scroll fashion. You can pull that up on all these websites.
However, they will change over time as the vote counts come in. That’s why we screen-capture these initial public postings, because that contains the purest information in terms of not relying on the vote counts and if we’re approaching this with a certain amount of suspicion of the vote counts we’re trying to verify or validate the vote counts we want exit polls that are independent as possible from the actual vote count data, which then becomes blended in as the evening goes on from the time the polls close until whenever the final vote counts are available. That vote count data becomes blended in with the exit poll algorithm and gradually pulls the exit polls in congruence with the vote counts, at which point they’re used for academic analysis of demographics, but they’re not anymore used for validating the vote counts.
SR: Tell me again what the ‘red shift’ is and how you saw this shift again this year.
JS: The red shift is a term that I coined back in 2004 after the Bush-Kerry election, because the familiar term the “red shift” when we mean astronomy, that’s what brought it to my mind. But the reason it’s called the red shift is that it was very directional in that election where you saw vote counts coming out more in favor of Bush, more in favor of Republican candidates. Since Republican by that time had been designated red as in red states and blue states, that’s how it got the moniker the red shift.
What we found from that point forward is that it’s almost a singularity, very rare, that we find any significant blue shift anywhere. When we look at exit polls and vote counts, what we’re almost always seeing are vote counts that come out more in favor of the Republican candidate than the exit polls and in the case of intraparty nomination battles, more in favor of the candidate that is, I guess you’d have to say, to the right of their opponent.
For instance, in the 2016 primaries, a massive shift of exit polls state after state after state, in favor of Hillary Clinton. The vote counts were more in favor of Clinton than the exit polls, which were more in favor of Bernie Sanders. We saw a very consistent pattern of that.
In this past Tuesday, again we saw a very consistent pattern of exit polls that were more in favor of Hillary Clinton, more in favor of Democratic senatorial candidates and then vote counts were shifted from the exit polls to the right towards Donald Trump, towards the Republican senate candidates. Those are the figures that I pulled down and did a very basic analysis of. You have a column of numbers of state by state showing the degree of that shift and we’ll eventually do that for the national vote for the House of Representatives as well.
SR: When you see this discrepancy, without being overly simplistic, the question becomes, why is it there and what caused it? You’ve been through this four or five times and not even counting the midterm elections. What do you think is really going on when you see this general one-way shifting? Does it mean the polling is wrong? Does it mean the voting machinery is being tampered with? Does it mean both? How do you explain or understand this?
JS: What it means to me is that neither system is self validating. Neither system can be trusted. If you look at accounting, you do double entry accounting. I’m not an accountant so my terminology may be off, but you basically audit by checking one column of numbers against another column of numbers. If they disagree, you know something is wrong somewhere. There is some arithmetical mistake, some failure of entry, possibly fraud … you don’t know. You just know that if two things that are pretty much supposed to agree had disagreed, there’s a problem somewhere. I can rule out mathematically and scientifically, by this time, errors due to random chance. Errors due to random chance, sampling errors, what we call margin of error issues, would not be expressing themselves so consistently in one direction. They’d be going in both directions and they’d be much smaller.
If you take a mathematical sample of a whole … if you take a blood draw in a person and you look at 1,000 or so blood cells as represented in of all their millions of blood cells, that’s guaranteed to be a random sample. It’s not like all the bad blood cells hide out in a single vein or something. From that, you get a very clear and crisp mathematical margin of error and it tells you how likely you are to be within X number of percent about what the truth is about the entire target that you’re looking at of the blood of the whole body. That’s how you can make a diagnosis based on a pinprick.
In exit polling it’s not that simple. In exit polling you have sampling that is not purely mathematically random. First of all, it’s done in clusters because it would be an impractical matter to catch people all over the state randomly coming out at the polls. You’d have to have a person at each precinct, etc. We’re not even talking about early voting and absentee voting. Let’s just leave that out of the equation and assume everybody votes on election day. You’d still have to go to thousands of precincts. It would be prohibitively expensive. What they do instead, and I was a pollster for a couple of years quite long ago, but the methods haven’t changed that much, you basically cluster sample. You pick 20 or 30 precincts that are representative politically and demographically of the whole state and those are the precincts in which you do all your interviews.
That adds mathematically about a 30 percent increase to the margin of error, to the inaccuracy if you want to call that of the poll. It’s certainly a tolerable change or loss of accuracy that can be factored in mathematically, but the real problems come up in exit polling with selection bias, response bias, the possibility of people lying to the pollster, etc. These are the things that have been seized on by those who have debunked the exit polls and said they’re worthless. They’re not worthless and at the same time they’re not best evidence. Best evidence would be the voter marked paper ballots. Best evidence would be the memory cards in the computers and what program is actually determining how these votes are counted, what the code is on those memory cards.
Exit polls are indirect. They’re statistical evidence and they have flaws that are difficult to quantify. When you see pervasive patterns where it is substantial well beyond the margin of error repeatedly in the same direction, in particular when you’ve been able to independently validate the demographics of the exit poll sample. This is the work that I did. It’s in my book, Code Red: Computerized Election Theft in the New American Century.
SR: So this is a persuasive and recurring pattern and not just in this week’s vote?
JS: In the 2016 primary, we compared the performance of the exit polls in the Republican primaries with the performance of the exit poll in the Democratic primaries. There was a glaring difference. I call these “second order comparatives.” Second order comparatives are very important because you’re essentially validating your baseline by doing that. If you’re conscientious about election forensics, that’s the work that you try to do. Does it add up to ironclad proof? No, but it’s a very consistent pattern that is absolutely probative enough that it says, Okay, we want to now take a look at the other system and how the votes are being counted. When you look at that other system and how the votes are being counted, your hair stands on end because it’s so vulnerable to not just outsider hacking, but to insider manipulation as well.
There are certainly a lot of anecdotal instances of this. For instance, just in this particular election, they bought machines in Ohio that had a feature in them that was basically capable of self auditing. It was a security feature. The Republican secretary of state of Ohio allowed the counties to switch off that feature. You have to ask why. You bought it and it had that feature. They said, Well, it would create chaos. You look at things like that and say hmm. You scratch your head and say, what is going on here? What may be happening in that darkness of cyberspace that the exit polls are giving us a pretty good hint about, but the vote counting system itself completely conceals?
SR: Let’s talk about what you found this week. I’m looking at your 2016 presidential chart. I’m looking at North Carolina for example, where it says the exit poll margin was 2.1% ahead for Clinton, but the final vote count showed Trump with a 3.8% lead. You have similar 4.4% Clinton lead in Pennsylvania but then losing by 1.2% to Trump, a 5.6% shift. You have Florida where she was ahead in exit polls by 1.3% and ends up losing by 1.3%, a 2.6% shift.
Is there any reason you can point to as to why you are seeing that in so many different states?
JS: First of all, let me preface it that what they’ve done since 2004 is exit poll fewer and fewer states. I think there were about 30 states exit polled this time, 20 states were left out because they were considered to be locks, non-competitive. What that does for a forensic standpoint is that it cuts our baseline… It’s as if they had a certain limited amount of resources, and they decided to really plow it into getting larger sample sizes in states that they knew were going to be competitive and possibly controversial.
North Carolina was one of those. I believe it had the largest sample size in the country. It was almost 4,000 voters were sampled and the usual sample size in these state exit polls is somewhere between 1,500 and 2,000 if they expect it to be competitive. That was basically a double sampling that reduces the mathematical margin of error, but it also improves in a less quantifiable way the accuracy of the poll. That 5.9% red shift from Clinton to Trump is way outside the margin of error for that poll and therefore very unlikely to occur by chance. What might have made it happen? People could’ve been lying to the exit pollster. The exit pollster could’ve been all young urban college kids and the Trump voters might have been reluctant to comply with their requests. There might have been refusals from Trump voters.
Now Edison usually tries to get these things right and one of the ways they try to get it right is through some expensive training and they try to get a fairly represented sample of polling interviewers. The polls by the way are confidential. They’re not verbal interviews. You’re just handed a clipboard with a poll on it. It’s not as intimate as some people would believe. There’s less of an incentive to lie because it’s basically confidential. You fold your polling sheet up and you put it in the box or you hand it back to the interviewer to put it into a grab bag. There’s no name on it. There is nothing that associates you with it. The incentive to lie isn’t particularly high. We’ve always dealt with the—is there a reluctant [George W.] Bush responder going on here, is there a shy Trump voter? We don’t know. These are possibilities, but we’ve seen the same kind of exit poll pattern in intraparty contests, we’ve seen it year after year, we’ve seen it at the Senate races, at the House exit poll. It transcends an individual race like this where there was so much intensity.
If you want to sleep well at night, which I also prefer to denial, and you want to say to yourself, Yeah, it must have been people just lying to the exit pollsters and I’m not going to worry about it, that’s fine. What you’re missing at that point is the fact that if you challenge me to say, How do you know these exit polls are valid? I would turn right around and challenge you and say, How do you know the vote counts are valid?
The fact is, and this is cold hard fact, neither of us can prove our case. That is the problem. We have an unobservable system that cannot answer the challenge that it might be subject to manipulation. It can’t demonstrate that it is not rigged. Exit polls are just a tool that we use to look at it and say, Well folks, there might be something to dig deeper into here. The problem is virtually never is anyone allowed to dig deeper. We have optical scanner equipment all over this country right now that have the voter marked ballots that drop through the optical-scan reader device and sit in their cabinet below. Those voter marked ballots need to be saved 22 months in theory, although they’ve been destroyed early, in fact, in many cases, especially if when there was an investigation going on in Ohio.
You have these voter marked ballots that would have probably not been destroyed within two days of the election and they’re there. They theoretically could be exhumed and examined. You could go machine by machine, you could look at them in public and you could compare them with machine counts, then you could reconcile those machine counts with the central tabulator. County counts, and state counts … You could say, Yes, this was a valid election or no, this was not a valid election. We had a problem. Might have been fraud, might have been a glitch, we don’t know. The fact is, nobody has access to those ballots. They are corporate property. They are off limits to public inspection. It might as well, in the 99.9% of cases, be a paperless touchscreen that has no record whatsoever.
The fact is, we are denied, when I saw we, the candidates, the public, very often election administrators, by the rules of their states, are denied access to the actual hard evidence we call it, that would allow a determination of whether the election has been accurately counted or perhaps has been illegitimately counted and manipulated. As a matter of fact, in quite a few states and usually under Republican control, but the Democrats have not been tremendously cooperative about this either. The trend has been for ballots to be removed from public record status so that they are no longer susceptible to four-year requests and similar public information requests, Freedom of Information Act requests. They are getting less transparent, not more so.
SR: Where do we stand today? You have your exit poll analysis that says something is not right here. We look at the results. We don’t have access to the ballots. We really don’t have meaningful audit or recounting possibilities. People are just left feeling very upset and powerless and then the days pass and attention goes elsewhere and the election machinery and the polling machinery largely remains the same.
Am I wrong? Am I missing something?
JS: I couldn’t have stated that better. There was a Supreme Court case famously where it had to do with pregnancy and it stood for mootness; when cases became moot because they couldn’t be heard in time so they had to do either abortion or pregnancy. By the time it came up through the channels, the time had passed and the woman was no longer pregnant and the issue was moot.
In a similar kind of way we go through these cycles. Interest heightens right before the election. If there is high suspicion of fraud, which sometimes as there sometimes are. If in presidential elections, 2004 obviously being the major one and then this one, there is a flurry of interest, a kind of window of opportunity, but the public attention span and the media attention span in particular is very short. We live in a sound-bite society. There’s always other stuff to attend to. People are busy. People are working hard. By and large, you couldn’t find a subject that is less sexy than election forensics. The Kardashians are a lot sexier as is the football games on Sunday. So yes, it passes briefly in front of the public eye. There’s a lot of stirring about it and then it dies out and it’s basically left to us hardcore election integrity advocates. This is catastrophic. This is tragic. What we’re left with is a system that was accepted more or less without real proof.
If that’s what democracy is worth to us, then we deserve what we get. Democracy requires support. It requires citizen support. It requires an investment of care and an investment of vigilance and an investment of participation more than deciding, Yeah, I’m going to vote or I’m not going to vote. It requires the fulfillment of a duty to be part of the public that counts and observe the counting of the votes so we don’t have the ludicrous situation where we hand our ballots to a magician who takes them behind a curtain, you hear them shred the ballots, comes out and tells you so-and-so won. This is what we’ve got now and it’s what we’ve accepted. We spend more money in two weeks in Iraq then would cost us for 30 years to hand-count our elections. This is surrealistic, this is absurd, but it’s the very strong inertial reality. Getting the energy up to change that reality, especially when that reality has worked well by definition for everybody who is sitting in office. They’re the people with the least incentive to look under the hood and say, Hey, we need to change this. It’s what put them in office.
You will rarely meet a politician or a journalist with a seat at or anywhere near the adult table who wants to rock this boat, who wants to pull the cover off of this thing, who wants to restore public sovereignty. One other point about that is that the musical Hamilton is popular now and Hamilton was notoriously an elitist. Somebody who really didn’t believe in full franchise as many of the founding fathers, of course. Women didn’t vote, blacks didn’t vote, etc. You had to have property to vote in certain states. There’s a strain of that elitism that pervades the highly educated classes that tend to fill the political offices and tend to fill the journalistic offices around this country. These are people with high education, high capabilities. They look out at the American people and rightly or wrongly, if I channel what they see, they see a mass of Kardashian following and football fanatic uninformed people.
That’s our public. That’s our sovereignty. That’s what they see from their perch. It’s very understandable in a way that they’re not all that excited about restoring the sovereignty of that group because they don’t really trust it. In their minds, [the] last people you’d want making decisions about whether to war with Iraq or Russia or do this or that deal politically are a bunch of people who come on, flip on the TV, tune on “Dancing with the Stars” and drink a couple of beers. There’s not a whole lot of belief in public sovereignty, right or wrong. As moronic as one might believe a good part of this public is, it’s really the experts who keep wandering into the minefield. The public knows more than they’re given credit for.
One of the things that’s going on in this election manifests through Bernie Sanders, manifests through Donald Trump, is a very widespread recognition of that reality, whether they’d call it exactly such or not, a recognition that we are not being listened to, we are not being heard, we are not being considered. So there is this huge populist uprising. Was there enough to give Donald Trump the victory? That remains to be seen and may never be seen, but it was certainly enough to give both Bernie Sanders and Donald Trump the kind of support that was shocking to the journalistic establishment and to the political establishment. They’re not being the elites that they are … and yet they are not advocating for having a vote counting system that accurately, publicly and verifiably translates the will of that public into leadership policy and direction. That’s in large part what’s led us into the place that we’re in today, November 10th.
SR: Thank you, this is very good. There’s a lot here. I hope people will take the time to understand that something is not right with both the exit polls or the vote count machinery, and at a deep systemic level there are questions that are not being answered. I respect the work you’ve done, as discouraging as its findings are.