Possible Election Outcomes

Choose your adventure (and analysts) wisely

The MAGA ceiling is real. That doesn't mean Trump can't win: by my forecast, he has a 1/3 chance. That's not small. But it's smaller than other forecasters, and for good reason.

For a more detailed look at the MAGA ceiling, read this article I wrote that

Mike Madrid

was nice enough to share on his Substack.

I don't think there's a strong case to call the race a tossup, for reasons you're probably aware of - if you follow me here or on social media.

In the states Harris needs to win: Michigan, Pennsylvania, and Wisconsin, her poll average is at or near 49.

Sure, 49-46 would be stronger than 49-48 (which is closer to her status there) but it's still stronger than it feels.

True, her “outs” in places like NC, GA, AZ, and NV are strong enough to overcome a blue wall crack, but I'll save that for my forecast update.

The basics are much simpler than that.

I'll explain why.

By the traditional - wrong - definition that analyzes poll data by how much someone is “up by” says there is no difference between 49-47 with 4% undecided and 44-42 with 14% undecided. Both are “up two.”

Seems silly, right?

Well, it gets worse.

Those same analysts insist on comparing elections that are this dissimilar by how much the polls were “off by” - which is an improper analysis that treats polls as if they should predict the result.

If the poll averages reported above are exactly right, then:

A 44-42 (+2) poll average, given a 65-35 undecided ratio, would yield either:

53-47 (+6), or

49-51 (-2)

Depending on who the 65% of undecided voters favor.

In both cases, the poll is reported as “off by 4” because the margin of the poll “missed the result” by that amount.

Off by four, they say, just because the undecideds decided in a way contrary to their Almighty Assumptions.

Well, what if the same exact 65-35 undecided split happened in an election with an accurate 49-47 (+2) poll average, with 4% undecided?

Either

51.6-48.4 (+3.2)

Or

50.4-49.6 (+0.8)

Not so drastic now, is it? That same, wrongly reported “four-point error” under the same exact circumstances - except fewer undecideds - would only be called a 1.2-point “error” here. Junk.

In both cases from a 49-47 poll average, even with a fairly drastic undecided ratio, the leader won. The same thing would happen from an accurate 49-48 poll average.

That's because…if you get 50%, you can't lose. That's why being close to 50% is good. I know this is not a super advanced concept, but I can say with certainty that

Nate Silver

and other forecasters and analysts don’t understand it.

For one, if they did, they wouldn't compare 2020 or 2024 to 2016.

For two, Silver’s forecast said there was a 1/10 chance Biden would lose Maine in 2020, despite Biden's poll average at 53, AND about 6% undecided.

Regardless of forecast errors like above, the analysis below is what will be in future stats and political science textbooks as “junk” alongside Literary Digest’s unscientific methods.

Here, they're comparing an election with high third parties and undecideds (2016) to one with very low third parties and undecideds (2020)

It's just this same junk analysis repeated over

And over

And over

If you can't tell by my tone, I'm getting a little tired of it :)

People who are clearly very bad at math are freely, with no accountability, misinforming people about how math works. In many cases, they get paid a lot of money to do it (not that I'm jealous or anything).

*Deep breath*

The probability of a four point “error” (improperly defined) is substantially higher as undecideds and third-parties increase.

Yes, it is literally that simple.

This is bad analysis. In 2016, Trump “outperformed” his polls by winning over a large percentage of undecideds AND converting many people who were polled as supporting third-party candidates.

In 2020, he did so again with a favorable undecided ratio.

That's not as easy to do when there are very few of them.

That's why I illustrated the difference between a drastic undecided split from 44-42 vs 49-47.

Both are “up two” but could yield very different results based on how undecideds decide alone - without even getting into the possibility of poll or poll average error.

So, what does this mean for 2024?

To slightly oversimplify, there are approximately nine possible scenarios. Those nine possible scenarios are generated from three possible poll average scenarios, combined with three possible undecided ratios.

Poll averages can:

1. Be accurate

2. Underestimate Harris

3. Underestimate Trump

Undecided voters can:

A. Split evenly

B. Favor Harris

C. Favor Trump

True, and importantly, not all of the above have equal probabilities (and in a close election, these aren't the only variables that matter) but it makes for a very clear way to demonstrate a baseline of possible outcomes.

Kind of an “intro to forecasting” if you'd like.

Remember, the analysis below is based on current poll averages. With 4 weeks left, things could change in which case this chart would need updated.

---

Scenario 1: Polls averages are accurate and…

1A. Undecideds split evenly

Result: Harris wins

1B. Undecideds split in Harris' favor

Result: Harris wins comfortably

1C. Undecideds split in Trump's favor

Result: Harris wins narrowly

---

Scenario 2: Polls averages underestimate Harris and…

2A. Undecideds split evenly

Result: Harris wins comfortably

2B. Undecideds split in Harris' favor

Result: Harris wins in a landslide

2C. Undecideds split in Trump's favor

Result: Harris wins

---

Scenario 3: Polls averages underestimate Trump and…

3A. Undecideds split evenly

Result: Trump wins narrowly

3B. Undecideds split in Harris' favor

Result: Too close to call/Trump probably wins narrowly

3C. Undecideds split in Trump's favor

Result: Trump wins comfortably

---

Scenarios of interest

Any poll average of 49 that has underestimated that candidate (Scenario 2) will result in that candidate winning.

If 49 is an underestimate, that means the candidate's true support is 49.5, 50, or even higher which (even with very, very, little undecided support) guarantees victory.

This cannot be said when that poll average is 45 or even 47.

In Scenario 1, “accurate” is subjective. How accurate is “accurate.”?

User’s choice. If your goal is to find a number such that each of the three scenarios happens 1/3 times, that's fine.

If you're trying to detect how often poll averages have very accurately measured a candidate's base of support (what a poll average tries to measure) then maybe +/- 1%.

The reality is it's not as straightforward as picking a scenario, it's more about assigning a probability to a range of outcomes.

Because there are so many possible outcomes, even the most likely outcome is highly unlikely.

Likewise with the undecided ratio.

If we limit “split evenly” to “literally only 50-50” we are never going to have that scenario.

Should we expand “split evenly” to include 55-45? 60-40? Again, user's choice.

All of that is to illustrate the same damn point I'm a broken record about:

The math gets a lot easier when a candidate has a poll average of 49% or higher.

As always, let's sprint through the finish line.