Why were the Polls in the 2016 Election so Inaccurate? This year’s election surprisingly ended with Donald Trump’s victory over Hillary Clinton when polls predicted Clinton to easily wipe out Trump. How were these polls so inaccurate? In order to understand this phenomenon we must first look into the mechanics of polling. Polls are a method of research conducted to predict the views of a population using a smaller group of people to approximate the views of the population. This can be done in two ways. The first is called sampling. Sampling is done by strategically picking people from different demographics by gender, age, race, religion, income, education, etc. leading to more accurate results. The next way is called random sampling. This is when you randomly select many people and get their votes. This was is simple and works well normally, but requires far more interviewers than sampling. Creating a poll is extremely difficult. They require a lot of thought. To create an accurate poll you must phrase your questions correctly. You must list the answers in either specific or random orders. Some people feel comfortable with the first answer they read or one that seems simplest. This is why the most accurate polls tend to use basic questions with yes or no answers. Highly ranked polling agencies have professional pollsters who have worked with and studied polls for years. Their questions are constructed by skilled psychologists to make sure the responses can relate to and satisfy the voters points of view. They also have to focus on who to survey requiring experts in demographics. Once all the data is collected these agencies have to then give their data to mathematicians who analyze this data trying to calculate the balance of sides. After this, these experts try to figure out how accurate their data seems and predict the more popular opinion in the large scale. Having this said, how could the polls for the 2016 election have been so inaccurate? On November 8th, almost all polling agencies were predicting Clinton to sweep Trump. The New York Times had predicted Clinton’s odds at defeating Trump to be eighty-five percent, comparing the odds of Trump winning to “an NFL player missing a 37-yard field goal.” Many other organizations like PredictWise, FiveThirtyEight, the Princeton Election Consortium, and the Cook Political Report predicted the chance of Clinton’s win from seventy to ninety percent. Most of these predictions presumed Clinton to be guaranteed to take several major swing states. These of course were all incorrect. There are many theories attempting to explain this phenomenon. The first theory is the “shy Trump” effect. This is the idea that when surveyees were called they would feel embarrassed to be voting for the less popular candidate and therefore choose Clinton on the phone and when voting day came they would secretly vote for Trump.The first theory has received a lot of attention among pollsters. This theory is called the “Shy Trump” Effect. It states that the people surveyed weren’t honest in their responses as to who they planned on voting for. These “Shy Trumpers” might find that supporting Trump is embarrassing or socially undesirable because of his plans to ban Muslims, his racist comments towards black people, or his offensive comments towards women. Therefore, his supporters would feel uncomfortable replying to someone one the phone that they support Trump.Trump Acknowledged this theory back in June, where he says “He was supposed to win by 10 points, and he lost by 5 or something. So it’s a certain effect.” Here Trump was alluding to the “Bradley Effect” in the 1982 election for governor of California between Tom Bradley and George Deukmejian. Bradley was black so voters were afraid to tell interviewers that they won’t be voting for a black candidate. “Now, I have — unfortunately, maybe fortunately –the opposite effect. When I poll I do fine, but when I run, I do much better,” Trump added, “In other words, people say, ‘I’m not going to going to say who I’m voting for. And they get it, and I do much better. It’s an amazing effect.”If this were the case of the “Shy Trumper” Effect could private online studies prove it? Politico and Morning Consult did a study on this to test the validity of this theory. They found that with an online survey there was a slight INDICATION of the “Shy Trump” Effect. They also saw a dramatic increase in high-income educated voters, supporting Trump online. In the phone surveys Clinton lead by 10 points, resulting in 54 percent to 45 percent. When they observed the surveys from online they found Trump to be up a point on Clinton making it almost even, but still, a dramatic increase. Kellyanne Conway, Trump’s campaign manager predicted this earlier when she said “Donald Trump performs consistently better in online polling where a human being is not talking to another human being about what he or she may do in the elections. It’s become socially desirable, especially if you’re a college-educated person in the U.S., to say that you’re against Donald Trump.” In phone surveys, voters with college degrees supported Clinton by a 21 point margin, but online that dropped to only an 8 point margin.In October, a public opinion course at Cornell surveyed a bit over 1500 people nationally over the phone. 20 percent of those voters said they had no intention on voting for either candidate. They then did a separate survey and rephrased the question, saying if they HAD to pick one, who would it be. This time only 2 percent gave no answer. With the original survey they observed that Clinton was above Trump by by a margin of 8 points. In the second survey though that margin had halved, resulting in a 4 point margin lead by Clinton. This gives more evident to out “Shy Trump” theory and also brings us into our next theory, non-responsive biases.Non-responsive biases occurs when people don’t respond to surveys even though they will be voting. This concept is a threat known by all good pollsters and it is hard to deal with especially due to the fact that response rates in 2016 have fallen below 10 percent. This can cause the results to be in favor of one candidate if the people who don’t respond are systematically different from those who do. An example of this is the less educated voters because they were hard to reach by pollsters and contributed a great deal of votes to Trump. One other way that we can see that this may have been affected by non-responsive bais is because nearly all the polls were in favor of Clinton. When the errors are only in one direction, and are never distributed, then we can infer that the data contains a bais.One more idea to support the claim that non-responsive bias affected the polls comes from one of Trump’s classic phrases “fake news, fake polls.” He constantly repeated this phrase throughout his campaign. With extreme confidence he exclaimed “we are gonna win so big,” knowing of his strategy to have his followers believe the polls were fake. This could potentially, and probably did, have a large effect on his supporters’ unwillingness to respond to the polls. This could have resulted in many trump supporters not answering polls intentionally leading to extremely incorrect data displaying Clinton’s lead over Trump.The last idea, like the previous one, is extremely difficult to control by pollsters. This is the result of inaccurate likely voter models. Likely voter models are the tactics used by pollsters to identify likely voters. Even if they were to interview everyone in the country their answer would still not be concrete because they would have to have an accurate likely voter model in order to predict the outcome as to who will vote. A simple approach to controlling this is to ask the voters whether or not he or she plans on voting. This doesn’t work though because most people say they will vote for the same reason the “Shy Trumper” won’t say he supports Trump, because it’s socially undesirable to not vote. To take it a step further many likely voter models include a measure of enthusiasm of the voter into the equation. This allows the pollster to decide whether this person truly meant if he was planning on voting. Successfully predicting the electorate is extremely difficult and usually requires lots of luck on your side. The only issue is that if you make mistakes, small differences in assumptions can lead to extremely large differences in the predictions. In this election we were able to see the pollsters screw up terribly in their predictions of the votes in the Midwestern and Rust Belt States where we saw Trump win almost all of them.