Methodology & strategy
Why this site frames the primary the way it does.
The top-two math
California's June 2, 2026 primary uses a nonpartisan top-two system: the two highest vote-getters advance to the November general election, regardless of party. There is one Republican, Steve Hilton, consistently polling at 17–20%. Eight Democrats are splitting the other 60% of the vote. If Democrats spread their votes across multiple candidates, it is possible for two Republicans to finish top-two — even though Democrats outnumber Republicans 2-to-1 in the state.
The way to prevent that is to concentrate the Democratic vote on one candidate. The "best" Democrat to back is the one most likely to clear Hilton and the second-place Republican — in other words, the highest-polling Democrat at the time you cast your ballot.
Why "as late as possible"
Polling will shift over the next several weeks. Candidates will drop out (Eric Swalwell did on April 12, 2026). Endorsements will reshape preferences. Late-deciding voters tend to consolidate. If you vote-by-mail today based on April polls, you lock in a choice that may no longer be the highest-Dem by election day. California lets you return a vote-by-mail ballot through June 2 — use that flexibility.
How the banner is computed
Sample-size-weighted average of vote shares across all polls ending in the last 14 days. If fewer than three polls ended in that window, the three most recent polls are used instead. Each poll's weight is the square root of its sample size — the standard inverse-standard-error weight, since a poll's margin of error scales with 1/√n. A 1600-person poll therefore counts twice as much as a 400-person poll (not four times). Polls that did not report a sample size fall back to a neutral √1000 weight rather than being dropped. Withdrawn candidates are excluded.
No weighting by pollster quality or recency-within-window, no house-effect adjustment, no partisan-pollster discount. Those would all be defensible in theory, but they introduce judgment calls that are hard for a voter to audit. 538-style modeling and Bayesian priors would give a more precise estimate, but they require trusting a methodology that is harder to reproduce. The formula here can be replicated with a calculator and the poll table below.
The same sample-size weighting is used for the trend arrows in the race panel, which compare a candidate's weighted average across the first half of the last 30 days against the second half. An arrow appears once at least four polls are available in the 30-day window.
Where the data comes from
Every six hours a Vercel cron job fetches Wikipedia's 2026 California gubernatorial election page via the MediaWiki API, parses the polling table, and stores a normalized snapshot. Wikipedia updates within hours of each pollster's press release — we piggyback on that editorial process rather than running our own aggregator. If you spot a missing poll, the fix is to edit the Wikipedia article; we will pick it up on the next refresh.
Caveats
- Not every poll tests every candidate — the "sparse" rows in the table are real, not parser bugs. Candidates marked "—" were not asked about in that poll.
- Margins of error for individual candidates below 5% are comparable to the noise, so minor reshuffling at the bottom of the field is often not meaningful.
- Pollster quality varies. Partisan-sponsored polls (marked with a
(D)or(R)next to the pollster name) tend to flatter their sponsors. We include them but do not weight them differently. - Polls before January 2026 are included in the trend graph for historical context but may reflect very different candidate fields (some tested candidates — Padilla, Caruso — are not running).
Further reading
- Blue Voter Guide — full down-ballot Dem endorsements
- Mobilize — volunteer to help Democrats get elected nationwide
- PPIC, Emerson College, and Berkeley IGS — direct pollster sites