Stop looking at the needle. Stop refreshing the map. Most of all, stop believing that "84% reporting" means we are 84% of the way to a result.
Every election cycle, major newsrooms—The New York Times, CNN, the AP—perform a high-stakes magic trick. They present a "percent of vote counted" metric as if it were a hard physical reality, like the amount of gas left in a tank. It isn’t. It is a statistical guess, often a bad one, wrapped in a UI designed to trigger a dopamine hit and keep you glued to the broadcast.
The industry consensus is that transparency about data leads to an informed public. That is the first lie. Transparency without context is just noise, and in the case of election "scrapers" and "expected vote" models, that noise is actively misleading. I have sat in the rooms where these models are built. I have seen the panic when a "dump" of mail-in ballots from a blue urban core arrives and the "expected vote" denominator shifts by 200,000 in an instant.
The reality of vote tracking is far messier, far more political, and far less "scientific" than the slick dashboards suggest.
The Myth of the Denominator
To calculate a percentage, you need two numbers: the votes counted and the total votes cast.
Here is the problem: on election night, nobody knows the second number.
When you see a graphic stating that a county is "99% reporting," that 99% is not a count of physical ballots. It is an estimate based on historical turnout, voter registration data, and early voting trends. It is a ghost.
If a county had 100,000 voters in 2020, the model might assume 105,000 will show up in 2024. If 104,000 votes are in, the dashboard screams "99%." But if a local controversy or a surge in youth turnout actually brings 120,000 people to the polls, that "99%" was actually 86%.
The media calls this the "Expected Vote." I call it the "Expectation Trap." By anchoring the public to a fake denominator, newsrooms create a false sense of certainty. When that denominator inevitably shifts as late-night boxes arrive, the public smells a conspiracy. It isn’t a conspiracy; it’s just that the math was built on sand.
Why the Scrapers Are Broken
Competitors brag about their "sophisticated scrapers" that pull data directly from secretary of state websites. They want you to think there is a direct fiber-optic line from the voting machine to the graphics on your screen.
There isn't.
The "pipeline" of election data is a Rube Goldberg machine of human error. It looks like this:
- A precinct worker finishes a count and calls it into the county office.
- A tired staffer types it into a legacy 1990s database.
- The Secretary of State’s website updates a JSON file or, worse, a raw HTML table.
- The media’s "scraper" grabs that data.
This is where the "glitches" that fuel social media meltdowns happen. A typo at the county level—adding an extra zero—can make it look like a candidate just gained 50,000 votes out of thin air. The scrapers catch it, the dashboard reflects it, and ten minutes later, when the county fixes the typo, the "vote flip" becomes a viral conspiracy theory.
By prioritizing speed over verification, newsrooms are sacrificing institutional trust for the sake of being first to update a progress bar that doesn't actually matter.
The Rural-Urban Lag is Not a Flaw, It’s the Feature
The "lazy consensus" in election reporting is that we should report votes as they come in. It sounds logical. It’s actually the worst possible way to inform a viewer.
Because of how geography and infrastructure work, small, rural, deep-red precincts report fast. They have fewer ballots to count and fewer bureaucratic hurdles. Large, urban, deep-blue precincts report slow. They have millions of ballots, complex mail-in processing requirements, and more scrutiny.
When a news outlet shows a map that is 70% red at 10:00 PM, they are technically "reporting the data." But they are lying through omission. They know the blue votes are sitting in a warehouse in Philadelphia or Detroit, but because the "percent reported" metric is tied to precincts and not ballot volume, the visual representation is skewed.
We should stop reporting "percent of precincts." It is a meaningless metric in an era where one precinct might have 50 voters and another has 5,000.
The Model vs. The Reality
Let’s dismantle how "The Times" and others actually determine that "share of the vote." They use what is called a Multilevel Regression and Poststratification (MRP) model.
In plain English: they aren't just counting. They are simulating the rest of the night.
They look at the votes that have come in from "indicator" precincts—neighborhoods that historically mirror the rest of the county. If Candidate A is overperforming their 2020 margin by 2% in a specific suburb, the model adjusts the "expected" totals for all similar suburbs.
This is brilliant when it works and catastrophic when it doesn't. These models struggle with:
- The "Shy" Voter: People who don't fit the historical profile of the precinct.
- Cannibalization: When a candidate’s surge in early voting is just a shift from their game-day turnout, not an increase in the total pie.
- Late-Mail Variance: In many states, mail-in ballots can arrive days later. The models try to "guess" how many are in the mail. They are often wrong.
Stop Asking "Who is Winning?"
The "People Also Ask" section of Google is filled with variations of "Who is leading in [State]?"
If you ask this before 2:00 AM, you are asking a flawed question. A lead in the first four hours of election night is a statistical hallucination. Instead of asking who is leading, you should be asking: "What is the remaining vote's partisan lean?"
If a candidate is down by 100,000 votes but the "uncounted" 15% of the vote is coming from a district that traditionally goes 80% for their party, they are actually the favorite. But the "percent counted" bar doesn't show you that. It just shows you a red or blue bar that is longer than the other one.
The Professional’s Playbook for Election Night
If you want to actually understand what is happening, you have to ignore the flashy percentages and look at the underlying margins.
- Ignore the "Needle": It is a jittery representation of a model's uncertainty. It moves based on small batches of data that may not be representative.
- Watch the "Benchmarks": Look at a specific county. Did the candidate need to win it by 10 points to win the state? If they are winning it by 8, they are losing the state, even if they are currently "ahead" in the raw count.
- The "Voter Drop" Rule: Whenever a large batch of votes is uploaded (a "drop"), look at the percentage split of that specific batch. If a batch of 50,000 votes comes in from a blue area and it’s only 55% blue, the night is over for the Democrat.
- Accept the "Grey Zone": There is a period between 11:00 PM and 3:00 AM where the data is fundamentally unknowable. The media fills this with "expert" panel discussions. Turn off the TV.
The industry treats election night like a sporting event where the score is updated in real-time. It’s not. It’s more like a giant jigsaw puzzle where the pieces are being delivered by different couriers at different speeds, and some of the couriers are lost.
The "share of the vote counted" is a narrative tool, not a factual one. It exists to provide a sense of progress in a process that is naturally chaotic. If the media were honest, their dashboards would just have a giant sign that says: "Data Insufficient. Go to Sleep." But you can't sell ads against a "Go to Sleep" sign.
So they give you the needle. They give you the "99% reporting." They give you the illusion of a finished count while the actual ballots are still being pulled out of envelopes in a basement three counties away.
Stop being a consumer of the illusion. Understand that the denominator is a guess, the scrapers are fallible, and the percentage is a placeholder for the truth that won't arrive for days.
The most accurate way to track an election is to wait for the certification. Everything else is just high-budget theater.
Turn off the tracker. The math doesn't care if you're watching.
