The polling industry has been under assault in recent years for “polling misses,” but polling on average has been quite accurate. In 2024, some polling firms were almost prophetic in their numbers. Instead of doing a deep dive into the data within the polls (known as “digging into the cross tabs”), a better way to view the overall polling picture is by looking at polling averages across various news sites. These polling averages add all the polls done on a subject, state or national polls for example, and divide them by how many polls are inputted to get a polling average. Now, within those averages there is a standout polling firm in the 2024 field that one should look to for the exact mood of states and the nation. That standout is a polling firm out of Sao Paolo, Brazil known as Atlas Intel.
Rather than attempting to describe how exact polling was in the 2024 cycle, I bid you to look at the chart below. Atlas Intel effectively told everyone how Election Day was going to look days before results came in. Also notice how the Real Clear Politics (RCP) average, while not perfect, showed a tight race - which turned out to be the case, even with President Trump winning all the swing states.
|
|
|
|
| Election Result | Atlas Intel | RCP Average |
National | Trump 49.8 Harris 48.3 | Trump 50 Harris 49 | Trump 48.6 Harris 48.7 |
Georgia | Trump 50.7 Harris 48.5 | Trump 50 Harris 48 | Trump 49.1 Harris 47.8 |
North Carolina | Trump 50.9 Harris 47.6 | Trump 50 Harris 48 | Trump 48.7 Harris 47.5 |
Arizona | Trump 52.2 Harris 46.7 | Trump 52 Harris 47 | Trump 49.1 Harris 46.3 |
Nevada | Trump 50.6 Harris 47.5 | Trump 50 Harris 47 | Trump 48.2 Harris 47.6 |
Pennsylvania | Trump 50.4 Harris 48.7 | Trump 50 Harris 49 | Trump 48.5 Harris 48.1 |
Michigan | Trump 49.7 Harris 48.3 | Trump 50 Harris 48 | Trump 47.8 Harris 48.3 |
Wisconsin | Trump 49.6 Harris 48.7 | Trump 50 Harris 49 | Trump 48.2 Harris 48.6 |
This chart alone puts to bed the concept of poor polling during this latest presidential cycle with how incredibly close to the end results the RCP average and Atlas Intel numbers were. However, there were some slight losers in the polling aggregator game. Sites that collect and collate data are know as aggregators or in this case polling aggregators. In creating polling averages, it has become in vogue to “weigh” or favor select pollsters when creating an average. For example, The New York Times, 538 and The Silver Bulletin pick their favorites based on various criteria. The stated position of RCP, on the other hand, is to include as many polls as possible, where the extremes tend to cancel each other out. The difference between the two methodologies was slight but noticeable (see below chart). RCP’s averages tended to favor Trump, the eventual winner, while all the others gave Harris the edge.
| Election Result | RCP Average | New York Times Average |
National | Trump 49.8 Harris 48.3 | Trump 48.6 Harris 48.7 | Trump 48 Harris 49 |
Georgia | Trump 50.7 Harris 48.5 | Trump 49.1 Harris 47.8 | Trump 49 Harris 48 |
North Carolina | Trump 50.9 Harris 47.6 | Trump 48.7 Harris 47.5 | Trump 49 Harris 48 |
Arizona | Trump 52.2 Harris 46.7 | Trump 49.1 Harris 46.3 | Trump 50 Harris 47 |
Nevada | Trump 50.6 Harris 47.5 | Trump 48.2 Harris 47.6 | Trump 49 Harris 48 |
Pennsylvania | Trump 50.4 Harris 48.7 | Trump 48.5 Harris 48.1 | Trump 48 Harris 49 |
Michigan | Trump 49.7 Harris 48.3 | Trump 47.8 Harris 48.3 | Trump 48 Harris 49 |
Wisconsin | Trump 49.6 Harris 48.7 | Trump 48.2 Harris 48.6 | Trump 48 Harris 49 |
The misconception of “bad” polling probably stems from a polling bias in which people only look at polls that reinforce what they want to be true. “Well, all the polling I looked at showed my preferred candidate winning by so many points” is a common trope. They did not see or want to see the other not-so-favorable polling, which is why polling averages give a more honest outlook.
The accuracy shown in polling allows for some very plausible inferences. The first one that comes to mind is that Kamala Harris did indeed reset the race and may have won if the election were held immediately after Biden endorsed her. The second probable inference is that Joe Biden had a large, growing polling deficit that was insurmountable and might have led to a true Donald J. Trump landslide victory. Past performance does not ensure future polling precision, but we should not be so quick to condemn polling until we take a good look at those pesky boring numbers.
Comments