Elections
By Matthys Van Leeuwen, PH.D in Statistical Analysis
In a close race, shifts beyond this would be statistically improbable without systematic bias in the remaining ballots.
Last night, I was watching the first election results roll in for the City of Burien with my good friend Joseph. He did outreach at homeless encampments at Burien before the camping ban took effect and was now working in Renton. I quipped, “he’d soon be back working in Burien soon.”
The voters have spoken, at least in part and they have chosen progressive government leaders. But as a statistical data scientist, I don't dwell on the political ripple effects. Instead, I focus on what the data tells us about the likely final outcome once all votes are tallied.
I often wonder why can’t we predict results so confidently from the initial count?
Last week, I followed the national election in my home country Holland, which uses a voter-ID paper ballot system with in-person voting. Elections there run like clockwork: An exit poll drops right after polls close, and 99% of ballots are counted within eight hours. By contrast, Washington's mail-in system feels archaic—it often takes a week or more for results to solidify.
But let's break it down with the numbers from King County. As of the latest update (November 4, 2025, 9:01 p.m.), 292,499 ballots have been counted out of an estimated 356,499 total, leaving about 64,000 still to process. This first-night tally is far more reliable than a typical exit poll. as it already captures over 82% of the vote.
To gauge predictability, we can calculate the margin of error (MOE) for the vote share, assuming a worst-case scenario of maximum uncertainty (50/50 split). Plugging in the values yields an MOE of 0.0764%. In practical terms, this means the projected final vote share from tonight's count is accurate to within ±0.0764 percentage points at 95% confidence—assuming the initial batch is representative of the whole.
That's a razor-thin window: In a close race, shifts beyond this would be statistically improbable without systematic bias in the remaining ballots. This calculation may temper the optimism of trailing candidates hoping for a late surge, but it's grounded in established statistical principles.
I recall a striking statistical example from four years ago in the City of Burien: Hugo Garcia trailed Martin Barrett by over 300 votes on the first night. Garcia voiced hope that the remaining counts would deliver the turnaround—and remarkably, just two days later, a single day's batch provided exactly the 300 votes he needed. As a scientist, that sequence still baffles me.
School Board Races:
Two candidates who prioritize academics and evidence-based decision-making appear poised for victory: Katie Kresly holds a 7.7% lead over Joe Van in Highline School Board District 3 (based on her votes exceeding his by 7.7% relative to his total), while Sue-Ann Hohimer leads Angelica M. Alvarez by 5.6% in District 2 (similarly relative to Alvarez's share). Damarys Espinoza is leading by a wide margin that will be impossible for Ken Kemp to catch up.
State Rep Race:
Likewise, for those rooting for Burien Mayor Kevin Schilling to advance to the 33rd Legislative District, he's up by 6.2% over Edwin Obras (relative to Obras's vote total). All three margins dwarf the 0.0764% MOE, suggesting high confidence in these outcomes.
To clarify for the skeptics: These aren't raw percentage-point differences but relative margins (lead divided by the opponent's votes), which aligns with how MOE is contextualized in proportional analyses.
If my projections prove wrong, it would rightly raise questions about the mail-in system's integrity. If my home country can nail exit-poll predictions across millions of voters in a multiparty chaos—often within a few points—why can't a real first-night count, with 82% of ballots in hand, reliably forecast a two-candidate race?
Science demands better, and so should we.
Members Discussion