Can We Trust The Polls? The 2024 Election Forecast Breakdown
With less than 50 days until the election, everyone wants to know who is going to win.
I don't know. No one knows. If someone tells you they know, you should consider adding them to the list of people you shouldn't trust as a reliable source.
Before I dive into this election, I want to talk about some previous elections. The fact is that most elections just simply aren't that close. Most elections In the last 45 years, the presidential elections of 1980 (Reagan vs. Carter) and 1984 (Reagan vs. Mondale) are generally considered blowouts, with polls accurately predicting Reagan's landslide victories. George H.W. Bush won the popular vote by nearly 8 points and gained more than 400 electoral votes, sailing to victory despite at one point being behind in the polls. Bill Clinton's elections were closer. He did not win the majority of the national votes in 1992 or 1996, though he won more votes and a large amount of electors. The 2008 election (Obama vs. McCain) also saw a significant margin, though not quite as large as Reagan's, and Obama won all the states he was predicted to win, and very slightly outperformed the polls, making that election no close, and not as close as predicted. Similarly, it did not come as a surprise that he won in 2012.
Conversely, several elections have been notably close, including 2000 (Bush vs. Gore, decided by the Supreme Court, hanging chads, ect), 2004 (Bush vs. Kerry), 2016 (Trump vs. Clinton), and 2020 (Biden vs. Trump), where the final outcome was uncertain until late in the process or even after Election Day.
In other words, most of the time it's pretty clear who the winner is going to be ahead of time, and most of the time the polls are accurate enough to spoil the surprise. In fact, it was the Obama-era pollsters who gave rise to the kinds of intense statistical analysis we are used to today specifically because polling had become so accurate that it made it possible.
And here we are, just three election cycles later, and we can say that the polls are probably all garbage. What the heck happened?
Trump, technology, and the breaking of the world order
In 2016 and 2020 the polls were way off. In fact, they were worse in 2020 than they were in 2016. One reason is that many polls failed to accurately capture the level of support for Donald Trump, possibly due to his supporters being less likely to participate in polls or openly express their preference. Another reason -- the way we communicate has completely changed in a short amount of time and we're all rushing to catch up.
Pollsters don't call millions of people, they call thousands, conducting a sample that they extrapolate using fancier math than I feel like explaining (I'm not the math guy). That means they also can analyze their polls and tweak their math after an election so the next election gets better results. But for generations, until not that long ago, people had landlines in their homes and most of them answered them when they rang. Now we have cell phones, and some of us answer them all the time, and some never, and most somewhere in between. Text-message polls have similar results. There are generational, regional, racial, gender-based and partisan-based differences for who is eager to pick up the phone. The result is that pollsters have had to reinvent their workflow, and they simply don't have all the data they need to really figure it out.
The second problem is that the public doesn't understand statistics, probability, or margin errors. If they saw "Clinton up by 2%" that's what they expected. Before the 2016 election the popular polling analysis website Fivethirtyeight gave Hillary Clinton roughly a 71-72% chance of winning the election. She, uh, didn't win. But that mean that Trump won in roughly 30% of the simulations run by their algorithm. Because many of the polls were off by 3-4 percentage points, enough to change the result in many states. The polls, pollsters, and analysts were then widely criticized for being inaccurate, both fairly and unfairly so. If I told someone that the polls were 96-97% accurate, they would be happy with that number.
The fact is that our country is too divided, and the polls are mostly well-within the margin of error. So, we just don't know. We turn, then, to another fancy mathematical term -- conditional probability, a term not as scary as it looks. What conditional probability involves is that if something happens, something else is likely to also happen. Here are some examples:
Let's say you drop a carton of eggs off of your counter. Bummer. You're afraid to look inside because you know there's a probability that some eggs broke. The carton, and the shell of the egg, will withstand a certain amount of force, so maybe you'll get lucky. But if you see yoke running out of the carton, there's a probability (close to 100%) that one egg is broken (otherwise how to runny egg get there), but there's also a very high probability that more than one egg broke, since if there was enough force to break one eggs the rest are likely also at risk. In this example there are a two examples of conditional probability. The first is that if you see egg running out of the container (a condition), there's a near certainty that an egg broke (a probability). The second is that if one egg breaks (condition) there's a high probability that others broke as well.
Let's give an election example. Vermont is more liberal than New Hampshire, so if a Republican is polling ahead in Vermont then there's a very high chance that New Hampshire will vote Republican.
This is helpful in polling because we can see other trends at play. For instance, before Harris entered the race Biden's approval numbers were very low, and voters had a negative opinion on how he was handling the economy (unjustifiably so, but that's a subject for a different article), and likely voters were saying that the economy was one of the most important issues to them. Trump had an edge in the polls before Biden dropped out, but these other conditions also indicated that Biden would struggle to overcome the deficit, meaning the stage was set for a Biden loss.
Conditional probability can also let us see when the polls are flawed. If Harris is ahead in national polling by 5-7%, there's a very small chance that she loses enough battleground states to cost her the election. But if the national polling is closer to even, it's unlikely that Harris can win in places like Pennsylvania.
What does the crystal ball say?
The polls are noisy, contradictory, and entirely too close to call. I'll allow Harry Enton to explain:
In recent days, Harris has received a number of good polls, and generally she's been gaining or holding ground since the debate. But some analysts, like the ever-controversial Nate Silver, thought it was more likely than not that Trump, not Harris, was going to win the election before the debate. Now most of those analysts have changed their tunes and think Harris has the slight edge.
The frustrating answer is that there is no way to tell who is going to win and this will likely remain the case until after the results come in. Momentum is much easier to spot than the actual results, and there's no question that Harris had a lot of momentum coming out of the Democratic convention, stalled or maybe lost a little ground between the convention and the debate, and had a solid bounce after the debate. No matter how much momentum Harris gains, it's hard to tell exactly how deep the hole was that she had to dig out of.
But, sticking with conditional probability, there is suddenly a lot of good news for the Harris campaign:
1. Harris's favorability rating has skyrocketed, while Trump's remains steady -- and low. Before becoming the nominee many people simply didn't know much about Harris and her favorability polls followed Joe Biden's fairly closely and have been negative for a long time. Since being named the nominee, Harris's favorability numbers have risen faster than any political figure since 9/11. She now has an favorability rating of about 46.9% and an unfavorability of 46.6%, giving her a net favorability of 0.3%. Trump, on the other hand, had a favorability rating of 42.7%, about 4.2% worse than Harris. Perhaps more importantly, he has an unfavorability of 52.8%, leaving him with a 10.1% deficit. And while Harris has all the momentum, Trump's numbers have been fairly stable.
2. Americans traditionally trust the Republican candidate to do a better job on the economy (Americans are wrong, there's absolutely no evidence to support this argument and a ton of evidence to support the opposite but, again, that's a story for another day). This was especially true with Joe Biden (again, a topic for another day). But Harris has closed that gap and is currently effectively tied with Trump. With momentum on her side, she might top Trump before the election. Trump has very few issues where he resonates with most centrist voters. The economy is the main one (people think because he's a billionaire he must be great at the economy, again, a topic for yet another day).
3. North Carolina. For a long time now, many analysts have suggested that Pennsylvania is the most important state and would likely determine the outcome of the election. But while Trump has lost a lot of ground to Harris there, most polls show that the race is deadlocked within the margin of error. Some recent polls that have come out in the last two weeks show Harris might be gaining ground, but again the numbers are moving so slowly that it suggests it may be impossible to know the outcome there until after the votes are counted.
All of a sudden, however, Trump's support in North Carolina has collapsed and Harris has surged through the polls. She was climbing out of a very deep hole, 5-6%, but polls show her effectively tied.
Then there's the news cycle. You've probably read all about it by now but Lt Gov Mark Robinson, a Republican running for governor, is in hot water after very alarming remarks were found calling himself a "BLACK NAZI" and defending slavery on a porn website. Most of Robinson's campaign staff has quit this week, and while it's too early to see any polls on the issue, we can see Google searches in real-time:
All of this is terrible news for Robinson. Typically, committed voters have a tendency to ignore scandals about their preferred candidate. Human nature says that if a voter has already made up their minds they'd just rather not know. The fact that online searches are exploding about Robinson indicates that people are tuning in. That's likely to reduce Republican turnout, and possibly likely to motivate those disgusted with Robinson, particularly African Americans or others who are particularly disturbed by this rhetoric. Remember, with margins this small every single vote counts. And as the video above shows, while Harris can win if she loses NC, Trump probably
can't.
4. A popular vote pullaway? Again, we turn to conditional probability. With razor thin margins in swing states, it's highly likely that if Harris wins the nationwide popular vote by 5% or more, she will win enough swing states to be President. If Trump wins the popular vote, it's very unlikely he can win the swing states. But if Kamala Harris wins the popular vote by less than 4%, we're back to a toss-up, and it's very hard to guess who is going to win. Several recent polls have shown Harris gaining ground, but we haven't seen anything close to a pull-away.
New news every day
So now I've said that we can't trust the polls, but we can still gain information from them, and all the recent news is good for Harris and bad for Trump. So Harris will win, right?
Today's headline from the New York Times:
Polls Find Trump Shows Signs of Strength in Arizona, Georgia and North Carolina - The New York Times
I guess we’ll just have to wait for the actual voters to actually vote.
