"It is enough that the people know there was an election. Those who cast the votes decide nothing. Those who program the computers decide everything." -- Joseph Stalin (updated quote)
At 8:00PM the polls closed in Massachusetts. At 8:01, the CNN web site briefly displayed their exit poll results:
52% for Sanders vs 46% for Clinton.
(Thanks to Ted Soares for collecting these numbers.)
Pollsters choose a few communities that are demographic microcosms, proportionately representative of the full range of political opinion in the state. People are asked as they leave the voting booth, "who did you vote for", and the results are quarantined until polls are closed, to minimize the chance of affecting the outcome.
But within minutes after 8, the official returns begin to arrive, and then these network guestimates can be corrected to conform to the reality of how people actually voted, or how the voting machines recorded their votes. The numbers on the screen may still be reported as "exit polls" but their content has been adjusted based on the official returns.
50% for Clinton vs 49% for Sanders.
In Eastern Europe, when Western NGOs were monitoring the restoration of democracy after decades behind the Iron Curtain, exit polls were used to check on the accuracy of the reported vote counts. A 7% disparity would raise eyebrows. But in America, beacon of democracy, where election integrity is beyond reproach, we use the official results to gauge the accuracy of the exit polls instead of the other way 'round.
The networks have gotten the message. They noticed over the years that the "real" results always tended to be shifted to the Republican side (or the right, in primaries) compared to the exit polls. So they adjusted their methodology to change what is considered a "representative" sample.
For example, definitely not representative of the state are the 71 townships where votes are counted by hand. They went for Sanders over Clinton. They are all rural, and by all demographic measures should be more conservative than the rest of the state; but election results say that people whose votes were counted by machine voted for Clinton, and those whose votes were counted by hand preferred Sanders. And election results don't lie.
First a little history
12 years ago, the story about exit poll reporting was unearthed by Jonathan Simon, a lawyer-chiropractor-activist from north of Boston. He captured exit poll numbers on his computer screen from the first minutes after polls closed during the presidential election of November, 2004. Based on these polls, many people became concerned that Bush's victory over Kerry was the result of manipulation or foul play. Two years later, their concerns were vindicated when the story broke that vote counting in Ohio had been taken over in the middle of the night by a private firm in Tennessee that was owned by Bush's IT guru, Mike Connell. The White House had been provided with a back door to the computer in Tennessee, so that numbers that were officially reported from Ohio's capital in Columbus could actually be dictated from the privacy of Karl Rove's White House office. (Connell was subpoenaed to testify about the events of election night, 2004, but his plane dropped from the sky just weeks before his deposition.)
This situation was an embarrassment for all concerned, and in its wake, procedures were implemented to assure that it would not happen again. Were the computers that count votes subjected to more scrutiny? No, the exit polls were skewed to the right and scaled way back. Shoot the messenger.
So, twelve years on, we have votes counted by computer almost universally through the US. And steps have been taken to assure that those computer results cannot be challenged. Our Supreme Court, in its wisdom, has ruled that the companies that program the computers that count our votes are entitled to protection, and that their right to company secrets trumps the public's right to know how their votes are being counted.
In much of the country, votes are recorded on pushbutton machines, so the results go directly into computer memory and there is no paper record, hence no recount is possible. Almost all the rest uses optical scanners that read the paper ballots and count up the votes. So there exist paper ballots that can be counted by hand to check on the op-scan machines, but almost everywhere there are onerous legal procedures that keep the paper ballots sealed from independent scrutiny. In Massachusetts, the paper ballots may only be looked at if the official margin is less than half of 1%, and if one of the candidates gathers 2500 signatures on a petition within 6 days following the election.