Coming this November, most of us will be watching the Election Day coverage and watching many races projected within minutes of the poll closing. My question is how do they do this? I know a few decades ago, they used exit polls. But today, with much more early/absentee voting, isn’t that much harder to do? (For example, my home state of Washington has no in person voting at all, with everything done by mail.)
I see no one has answered this yet. I am far from an expert but I think a lot of it is just knowing the voting preferences (Dem/Rep or lib/con) of a particular area, (perhaps zip code), and looking at the turnout for the race. High turnout in a Dem area, Dem candidate likely to win. Add up the areas and make a prediction.
Sticking a wet finger up in the air is probably just as effective.
they still use exit polls even with early voting . They probably assume early voting people are about the same as people who vote in election day. If that is not the case then their projections could be off.
It’s mostly exit polls and a look at swing districts. There are some that usually match the final result – Republicans when they’re winning, Democrats when their winning. So if those districts are trending Democratic, it will be a good night for the Democrats.
you have states that are fairly quick at reporting their results.
These results are used at the precinct/county/block level are used to “impute” data for missing areas in states with demographically similar populations.
Each network/organization has their own standards for when to call a state. If the exit polls, pre election polls, and actual/imputed data are all leaning a certain way - and there is enough of a margin - they will call that state.
Sometimes they will even call a state with say 2% reporting - for candidate x - even if candidate y is higher in that 2%.
Exit polls by themselves aren’t useful for close races as they have a known bias towards younger voters. But they can look at the subgroups and combine that with other data to make a decision.
Sometimes the results are too close to call - and may even change later - such as Bush in Florida in 2000 - or Al Franken in 2008.
Some of the news media went overboard in 2004 - if memory serves - every network, except FOX, withheld calling Florida “early” in 2004 - even though it was obvious.
It is also childsplay these days to purchase demographic data mined from FB and other social media.
FB can tell you how many FB users in a given area have “liked” a given candidate as well as alot of background demographics.
It’s strictly the demographics and comparison to previous polls. If an area voted a certain way last election or two, then how much has the poll changed in this time? Up or down for a certain candidate? Add in demographic analysis - polls indicate Senor X is trending high with Latinos, say; take the polls with a given concentration of latinos (or gays, or protestants, or welfare people, or millenials - depending on the campaign appeals); is the turnout higher or lower? Did that candidate pick up votes. We already see on CNN, what percentage of men, women, assorted ehtnics and races - who voted for whom.
But the main thing is exactly that - swing polls. These are the neighbourhoods that flip back and forth from election to election, i.e. are more easily persuaded to change votes. Once there’s a hint how they are voting, that tells you the general trend.
For example, the Connecticut primary yesterday. Clinton was trailing well into the evening, but only by a percent or three. The missing polls were around the bigger citries and close to New York City, and analysts knew those tended to vote for Clinton based on opinion polling - so no call even when it got to 80% until they saw those results. OTOH, when several polls all over the state are 60% Trump, that’s a good indication the whole state will not suddenly vote completely different from the already reported polls that share that demographic. .
In the 2012 election in OH, Fox called the result for Obama with an absurdly small percentage of precincts reporting and, in the reporting polls, Romney was ahead. Karl Rove went into orbit. How can you call the state for Obama with only 2% reporting and Romney leading. He insisted on visiting the statistical analysis team. I don’t think they carried the actual explanation, but it certainly would have gone along lines like this: Yes, Romney is leading in those precincts, but by not nearly as much as he should be if he is to have a chance to win the state. Rove kept foaming at the mouth, utterly unconvinced. Of course the statisticians were right. This was Fox mind you whose bias would have been for Romney. Not that it mattered. The voting was done and what was, was.
Oregon has had vote-by-mail since either 2000 or 2004 (I forget which). In that year, I asked on this board how they would do election projections in Oregon. Unfortunately I can’t find the thread, but no one had any idea.
It turned out that they didn’t do any projections for Oregon that year and as far as I can tell, have not done so since then. So if you’re wondering how they will do it in Washington, the answer is they won’t. I think Colorado also has 100% vote-by-mail, so they won’t get any either.
Say what? These predictions are almost 100% correct. It’s very, very rare for them to mis-call a race.
Facebook would be highly biased towards a specific segment of the population
I would be astonished if any reputable network used this to call an election.
I’m not saying there isn’t useful info to be gained, but not to call an election. And there have been relatively successful attempts at internet polling that have worked out better than many would have guessed, but Facebook likes would need lots of finessing to turn into something useful for this purpose.
Despite people saying stuff “sticking a wet finger in the air” - general election polling is relatively accurate - especially when adjusted ala Nate Silver/538. Combining known precincts along with demographic data, prior polling, and proper modeling - it is relatively trivial to accurately call the vast majority of states in a presidential election.
Of course you want to have enough accurate data on Election Day to be sure - and it will vary from election to election. Probabilities can be assigned to every state and I believe I’ve heard that one of the Networks waits for 95% confidence - I would pick a higher number than that.
I was going by our experience in the UK.
But what that actually demonstrated was something more specific and more germane to the OP, namely the crucial difference between conventional opinion polls and exit polls. Because the exit poll used by the major UK broadcasters in May 2015 turned out to be pretty accurate, allowing them to make reasonably good attempts to call the result as soon as the polls closed. The BBC’s prediction underestimated the number of seats that the Conservatives would get by just 14, which admittedly did make the difference between a narrow majority and narrowly missing one. That those predictions were so out of line with all the previous opinion polls was why they were then met with such astonishment.
But it was no real surprise that an exit poll was more accurate than other polls. That’s mainly because they’re counting people who have actually turned out to vote. They’re not perfect, but they do give pollsters a significant edge when it comes to calling results.
My understanding is that they typically rely on exit polls, gathering data throughout the day, and they often know who the winner is even before the polls close. However, it’s generally considered unethical to announce the winner while the polls are still open. This was one of the many missteps in 2000 when Fox News announced on the air that they were calling FL for Bush just a few minutes after polls had closed in the eastern portion of the state, forgetting that the western portion is in a different time zone so they should have waited another hour.
In nearly every case, there is polling well in advance of election day. Before the polls even open, they already have a good idea of what’s likely to happen. If the exit polls align well with the previous information, and one candidate has a big lead, they might not even wait for any precincts to report actual results. But if the race is close, they would be likely to wait for precincts to start calling in the actual vote totals for that precinct, and if those results don’t match the exit polls, they’d be inclined to wait until more complete information is available. The networks hire statisticians who have algorithms set up in advance to estimate what’s the probability that the final result will match the previous information. Most of the time, it passes 99.9% certainty very quickly and they can just call the election. The really tricky part is, what do you do when the certainty is only 75% at 9pm and 93% at 10pm and 96% at 11pm. Do you go ahead and call it when it passes 95%? Or should you wait for 98%? Remember, if you say 98% is good enough, that means once out of 50 elections you’ll make the wrong call. There is not agreed-upon standard. Each network has to make their own decision about whether it’s better to be right or better to be fast.
Fox actually held off for a lot longer than that…about 75% of the vote was in when Rove had his famous meltdown. Trouble was, the bulk of the uncounted votes were in Lucas County (Toledo) and especially Cuyahoga County (Cleveland), and they were going to boost Obama’s small lead to comfortable margins…anyway, about three quarters of the vote had been counted by that time.
I was actually reading about polling today at the library (Exit Polls:Surveying the American Electorate by Samuel Best) and it had a partial explanation. The organization running exit polls now does phone interviews a couple of days before the election, asking if people have voted already. If they answer yes, they give them essentially the same poll people actually visiting the polls are given. Then on election day, when the election officials know the number of early voters versus the day-of-election voters, the numbers from the phone survey are weighed accordingly and included in the exit poll results.
Absentee and pre-election day votes can be counted early. Also, do they wait until the polls close to start tallying the rest of the ballots? It’s possible that at the time the polls close at night, they may have already counted the morning’s ballots.
Unfortunately.
It is becoming harder to find an unbiased source of information on many things these days. Polls are also becoming less accurate. There are many ways to construct and conduct a poll to produce a result that you desire.
Even exit polls can be biased. Where are the polls being taken? What time of day?
Who is more likely to answer the pollster? Truthfully?
I suppose this could vary by states, but most places wait until all votes are cast before counting starts.
There are huge potential problems with early counting. If the parties knew of partial results they could step up efforts to get people out to polls in the evening or try to block the other party’s voters. Given the history of voting, early counting is a gigantic red flag.