Analyzing the Undecided

Who are the undecided?

They like both candidates more than they dislike both and they care about climate change as much as they care about the economy.

Last week I was handed a juicy nationwide survey taken just after Harris became the presumptive nominee. The primary objective was to characterize the undecided voter. The following is restricted to 3209 likely voters: excluding those who said they will not vote, and excluding those who say they did not vote in 2020 and are unsure if they will vote in 2024.

Harris is up across most demographics save the obvious (whites, males, not metropolitan, less-educated).

What I found striking was (after excluding partisans who like one but not the other candidate) that there are twice as many who like them both Trump and Harris, as opposed to disliking both. For clarity: “like” is “Very favorable” or “Favorable”; “dislike” is “Very unfavorable” and “Unfavorable”

Harris has the edge on the double dislikes. Trump has the edge on the double likes.
But the undecided group is larger in both.

The New York Times recently evaluated the “double-haters,” inferring that many are Haley boosters. But, in fact, whether it’s the population total or the undecided, there are twice as many Ted Lassos as Roy Kents, twice as many Leslie Knopes as Ron Swansons, twice as many Pam Beeslys as Stanley Hudsons, twice as many Lisa Simpsons as Mr. Burnses, twice as many Forrest Gumps as Lieutenant Dan Taylors, twice as many James Wilsons as Gregory Houses, and twice as many Mr. Peanutbutters as Bojack Horse(mans/men)

A thousand words paints a picture.

In the survey including all 5150, respondents were asked what one word they would use to describe either candidate. There were 1414 distinct words. Here I have limited the word clouds to those appearing at least 10 times. “Supporters” is defined as those who definitely will, or probably will vote for the candidate.


I see more diversity of positive word usage among the Harris camp. Shocked to see “wh*re”. And it’s clear that the word choices differ quite a bit for the non-partisans.

The undecided: Economy and Climate Change.


Jobs would appear to be a wash. Trump has the edge on the economy, and it’s significant.
Biden has an edge on handling the pandemic, but it’s not significant (and obviously not a 2024 issue). Of all 3, for the undecided, as James Carville quipped, “It’s the economy, stupid.”

But it’s not “just the economy” anymore. It’s climate change too.

With an eye to characterizing who are the undecided, I ran a couple of machine learning models. Given the imbalance in the data (only 17% are undecided), and that most features are either categorical or nominal, I used boosting classification and decision trees, and then extracted the highest scoring features.

Having no party affiliation and disliking both candidates were obvious. But climate change also popped as being predictive. And the undecided skew in the same direction as those who are already decided for Harris.


This would appear to be a topic worth pressing by the Democrats who have been shying away from it.  It’s also worth rethinking by the GOP.

When a poor ML model may still be useful.

You’ll notice from the confusion matrix that’s it’s not a very good model. It misses more of the undecided than it gets, and it incorrectly labels as “undecided” as many who are actually already decided as it properly labels as undecided.


You wouldn’t want to use a model that has 56% false negative and 48% false positive to decide who to treat for cancer.

But think about it from a campaign ad-spend point of view. If you have $1 million to spend and just targeted everyone, 83% of the money would be wasted on decided voters.

If you use the model, sure, you’re going to miss most of the undecided but with only $1 million you were going to miss most of them anyway. While only about half of the people you target will in fact be undecided, wasting $480,000 is a lot better then wasting $830,000.

Especially in an election year where moving the needle one or two percentage points could have a huge impact on the outcome of the electoral college vote.


Scroll to Top