The Influence of Polls

If Trump wins, we need to disregard all current polling organizations because they will have been vastly outside of their margin of error for two presidential election cycles (This is based on my personal analysis comparing polls vs. results in key battleground states, not nationwide polls. See P.S.). Their methodology is clearly flawed and that must be intentional or due to negligence. While I am saying people should completely disregard the polls after a Trump victory, I do not believe they will. The media and the people who follow election campaigns love polls, and it is hard not to. As far as I can tell it is the best information we have on the outcome of a future event and it is numerical and thus easy to comprehend. What would happen, though, if we did disregard the polls in the next election cycle?

Do polls themselves influence the elections they predict? I believe they do. For the first time this year I was closely following the Democratic Primaries and I had the privilege of learning about Iowa’s bizarre caucus method. To explain it briefly, Iowans gather in local gymnasiums/meeting halls and group up in a designated section to show support for a candidate (e.g. “If you support Biden go to the corner of the gym with the basketball hoop”). There is an initial round of standing, the votes are counted, and then another final round is commenced. Observe what occurred between the first and the final round in the 2020 caucus:

In the first round of voting, the top 5 candidates received 162,487 votes (92.1% of total). In the final round, those candidates received 168,475 votes (97.8% of total). This increase in votes came from mostly from the lowest candidates (especially Andrew Yang. You may be able to guess why I noticed this). Yang, Steyer, Gabbard, Bloomberg, Bennet, and Patrick received 12,701 votes in the first round (7.2% of total). In the final round, they received only 2,207 (1.2% of total). To me, this is a real-life example of the influence apparent indications of how an election will go have on the results of said election. In the Iowa Caucus, viable candidates drew wavering supporters from less viable candidates. (You may observe that the low tier candidates dropped 10,500 votes and the high tier candidates gained only 6,000. I think we can assume that some people who came to support lower tier candidates went home when they saw the lack of support) (See P.P.S. for ranked choice voting)

In 1951, Solomon Asch created an experiment (known as the Asch Conformity Experiment) to show how social pressure from a majority group can cause a person to conform. He put 8 people in a room: 7 actors, and one naïve subject. The subject believed the other 7 people were test subjects just like him/her. The group was then shown problems in which they had to answer which of 3 lines presented on a screen was longest. The answer was always obvious, but the 7 actors preplanned to uniformly pick one of the wrong answers (on 12 of 18 questions). Asch found that the subjects would often conform and say the incorrect answers.

“Over the 12 critical trials, about 75% of participants conformed at least once, and 25% of participants never conformed.”

I believe that the first round of voting in the 2020 Iowa Democratic Caucus was much closer to people’s actual wants than the final round. I believe a presidential election without polling or low emphasis on polling would yield similar effect on the results.

P.S. I have seen more than one articles that take the odd position that the polls in 2016 were “not that wrong”, and they usually cite that the national results were within the margin of error for the national polls. I think someone forgot to tell them that we have an electoral college, and the state polls are the important ones. I used the state polls taken closest to the election and compared them with the election results, and I found that on average, the polls underestimated Trump’s support by 6% with only 5 states overestimating or correctly estimating the level of Trump support. More importantly, in the battleground states, Trump’s support was underestimated by an average of 3.8% and this allowed him to nab 83 of the 125 electoral votes from that group. One of the more devastating upsets was Michigan, which is considered “light blue” or “somewhat more democratic” state where Trump’s support was underestimated by 5%. Perhaps many of us learned in 2016 to be skeptical of polling numbers, but they are predicting that Joe Biden is beating Trump by even more than they predicted Hillary was in 2016. If they get it “wrong” again, I am going to guess that they are intentionally misleading us.

Contentious states that Trump won in 2016 (from my personal analysis)

P.P.S. The 2020 Iowa Caucus also highlights how important ranked choice voting is. If you only get one vote, you must be strategic and vote for the candidate you consider most viable who you would like to see in office. With ranked choice voting, less strategy is needed. Imagine there are 10 voters and 3 candidates. Let’s say you can see the results as the election is happening and after 6 votes have been cast, Candidate A has 3 votes and Candidate B has 4 votes. You like Candidate C the most, but you like Candidate B the least. It is already mathematically impossible for Candidate C to win and if you vote for him, you increase the likelihood that Candidate B is elected. You cast your vote for Candidate A and Candidate C ends up receiving 0 votes.

If this election involved rank choice voting, Candidate C might show up on many voters rank #2 spot. This doesn’t mean Candidate C would win, but the results might more accurately reflect the electorate’s affinity for him/her.

Latest Posts

Follow My Blog

Get new content delivered directly to your inbox.