This follows a Karl Rove book of several years ago and Catherine Shaw’s Campaign Manager. She is a Democrat. The first thing to realize in polling is that not everybody votes. 240 million adults over age 18, 187M registered, but only 126.46M voted in 2012 election. (75% of adults are registered. 65% of Registered vote in typical Pres. election.) The second thing you need to know is that only about 1/3 of the people asked to respond in a poll do so. Thus, the task of making an accurate data base is difficult. If Hispanics represent 14% of population but vote 8% of the time, and your poll of 1000 voters had 7% Hispanics, do you weight them by doubling or do you multiply their influence by 1.14? Af-Ams typically vote at 12% of the total vote, but represent 13% of the population. However, in the Obama years, they voted heavily representing 14% of the total vote. And do you weight it according to minorities or age or something else? Seniors vote heavily and millenials vote sporadically. Jews are just 2% of the population but vote at 3% of the total. Altogether, there is a huge guessing game of weighting the groups and putting in fudge factors to reflect participation.
You are excused if you are thinking at this point that polls mean nothing. Gallup has dropped out of the Presidential polling, but for none of the above reasons. It was once easier because the pollsters could get a copy of county voting records. Voters are listed by party, address, and number of times they have voted. So if a phone call to an address was picked up and the voter identified themselves, the pollster could assess likelihood of voting. “Mrs. Jones voted twice out of the last 4 general elections so she has a probability of showing up this time of ½.” Cell phones, disposable phones, unlisted numbers have shredded this scheme. On-the-street interviews are good, but people don’t have time, answer (in public) according to what they think the pollster wants to hear and have very low response rates. Internet polls have no controls for partisans who do multiple votes. All of this is chancy science.
Poll results are often all over the map in variation. Real clear politics takes an average of recent polls to assess a race. But even if they get more accuracy, they don’t tell you about a very important fact. When an incumbent is below 47%, even if he’s ahead, he is probably in trouble. Rove summed this up by studying outcomes versus final polls taken. If an incumbent polls 48% he has a better than 50% probability of winning. If it is 43%, this shrinks to less than 10%. The reason for this is that when people answer a poll and give a well-known incumbent, say, 46%, this means that 54% are looking for “other alternatives.” Typically the number of undecideds will go 60-85% for the challenger. So as I write this, Senator Burr of NC and Sen. Ayotte of NH lead their races by a percentage point. But with 45 and 46% respectively, they are less than half probable of winning. I’ll give you a self-depreciating example of this. When Steve ran for the House in 2010, the State Republican Party did a 100 person phone poll of our district. No demographic corrections were done. They found he was behind 39 to 43%. Steve thought he had lost and I would have agreed if he’d have shared this with me. But the state party rejoiced—he had a chance!—and soon he was talking to some big donors for direct mail funds. It turned out he won by 53-47%. That is, of the 18 points undecideds (100-39-47=18), he won 14 of them. (Golly, we were so uneducated about political science when we won that first race! We didn’t know that the probability of beating a 3rd term House member was less than 1%. And Luttrell had won with 60% of the vote in 2008!)
They do exit polls on many things. One question is “how far in advance did you decide to vote for the guy you did?” Answers vary from “I decided 40 years ago, I’d never vote for another Democrat!” to, “Never heard of him. I didn’t know who I was voting for when I checked his name.” Results are consistent however. The number of people who decide within the final two weeks of an election is usually 1 or 2%. Almost everyone has their minds made up two weeks out. But Television never says this! Indeed, they broadcast a myth about polls changing radically the last week. Why? They get advertizing money from campaigns wanting to convince that last tiny group of undecideds. And campaign season is like the Christmas season for broadcasters. They make big bucks. And they spread a myth of the Independent as a careful standoffish voter who can’t decide. This is only true of about 1 out of 5 Indies. Indies come in a lot of flavors. Some are hidden partisans who just don’t want to get in an argument with their dad or wife by joining the other party. So they play coy. As an Independent they can “constructively” criticize dad’s party. Or they are at odds with both parties—fiscally conservative but socially liberal, or have some single issue that neither party addresses. Independents vote only half as often as partisans.
The pollsters and TV guys play a game with viewers, claiming that “polls are tightening in the last week!” More likely, the pollster knows that his results versus outcome will be measured closely and decide his future business. He is trying his best to weight for “likely voters” (previous polls were ‘registered voters’ or ‘general public’) and get an accurate poll. Previous polling done in earlier months for a new organization are often just to feed their talking heads. Pollsters give the sponsors polls that reflect the sponsors point of view. They tell them what they want to hear. Changes in polls can then be cited by the news organization for reason for the candidates to spend more money in last minute advertizing.
And then pollster just sometimes blow it. Brexit was supposed to have Remain win by 5%. They lost by 5%. Reagan was supposed to lose to Carter by 5-10% and won by 12%. In both cases, enthusiasm was sky high for the eventual winner compared to the loser. As mentioned, Romney lost because his base didn’t turn out. That affected the weighting factors. JFK had huge Catholic turnout which swamped anti-Catholic turnout, but he still wouldn’t have won had he not carried Illinois which was determined by dead people voting in Chicago. Gore won the popular vote but it was not just FL that killed him with narrow loss. He narrowly lost his home state of TN—which would have won the election for him. Botton line: It’s the election that counts.