How to avoid getting fooled by the polls – 2026 California gubernatorial primary campaign

Author David Kornahrens, Senior Vice President of Research

Disclosure: The author has not worked for any campaign committee, political action committee, or other group involved in the California gubernatorial election during 2025-26 election cycle.  The author has no access to private survey polling results from any such group.

We live in a time of ever-growing streams of data. And yet, despite all that information, it can be exceedingly difficult to cut through the noise and discover the real signal. Understandably, everyone wants a forecast of what will happen before acting. What will the weather be like when I walk outside? What will my child experience if they attend a particular school? How will my investments fare in the face of current events? Survey poll research can provide a forecast or preview of how voters may be thinking — but only if the data is interpreted correctly.

That distinction matters. Survey poll results are useful, but it is not a scoreboard. The deep desire to know the future now has been seen in the abundance of chatter about the outcome of the California primary election for governor. In a crowded top-two primary like the 2026 California gubernatorial race, survey polling can be misleading when stripped of its methodological context or accidentally misinterpreted.

As public opinion researcher who advises clients on strategy, I want to step in and clarify a few concerns and share some insight earned from previous electoral cycles. I have worked as a campaign manager, director of polling teams, and a senior vice president directly advising candidates running for all levels of office. I’m here to help you interpret electoral survey research and not be surprised on Election Night.

Survey Polling: When, who, and how? And remember survey poll numbers ≠ scoreboard

You can pat yourself on the back if you have heard the phrase that a poll is “a snapshot in time.” It is overused because it is true.

A survey poll is an interview questionnaire whose resulting measurements depend heavily on when the interview is conducted, who the population of respondents being interviewed are, and by which method(s) are the interviews conducted with respondents from a population. Each of those methodology choices can materially affect the survey results.

Even when every step is done correctly, there will always be a margin of error. That margin of error creates a range of uncertainty around each data point. It should not be treated simply as the distance between two candidates’ percentages. If Candidate A is at 21% and Candidate B is at 18%, the meaningful question is not who is “ahead.” The better question is whether their possible ranges overlap once sampling error and other sources of uncertainty are considered.

In the context of the California top-two primary in June 2026, survey results from May 2025 provide far less insight than survey results from May 2026. Interviews conducted through only one method are usually less robust than survey research conducted through a mixture of methods. And, vitally, survey research is strongest when it is derived from or matched back to a detailed voter file database with information on how an individual has voted in the past.

The population being measured is just as important. A poll of all adults is not a poll of likely primary voters. A poll of registered voters may still be too inclusive of individuals who do not vote. The relevant population for this election is not “Californian adults,” or even necessarily “California registered voters.” It is the narrower and harder-to-model population of likely voters who will actually cast ballots in the 2026 primary among the wildly crowded field of candidates – with six well-funded Democrats and two well-funded Republicans, and over 50 others who have registered their candidacy.

If this sounds over-complicated or prescriptive, then you’re beginning to understand how difficult it is to achieve statistical validity when conducting survey poll research of a future population. That phrase — “future population” — is vital. The fully verifiable data does not exist today to know exactly which voters will cast ballots by June 2, 2026. Pollsters can model that future population, and good pollsters can model it well, but they cannot observe it in advance.

That is why survey poll numbers are not a scoreboard. They are measurements with uncertainty attached.

Name Identification: Time is as valuable as money, but having both is best

Early in a campaign for statewide office, the primary objective is simple to describe and extremely difficult to achieve: a candidate must become known, and preferably known positively, by enough primary voters across the state.

If a candidate is not already well-known statewide with a positive impression, building that profile becomes a multi-million-dollar exercise requiring months of sustained communication.[1] This is one reason many incumbent elected officials often fail on their first two or three attempts before winning elected office.

These fundamentals require successful candidates to announce their candidacy early enough – and raise enough financial resources – to communicate long enough to primary voters. Candidates who enjoy enviable financial resources but not enough time, like Mayor Matt Mahan, are said to have “run out of runway” like an airplane attempting an ill-fated landing. The phrase is apt. Successful campaigns are not simply about the ability to spend money; they are about having enough time for that spending to create familiarity, positive association, and ultimately voter support.

Other candidates, like Tom Steyer, have run for higher office before. In the case of Steyer, he announced eight weeks before Mahan, has given his campaign over $200,000,000, and spent well over $120,000,000 on paid advertising. This is the right mixture of resources to be competitive, though not necessarily assure success. The effect of this communication on the electorate is seen, in part, in the increase of Steyer’s public polling percentage since his paid communications began at volume in early 2026. The effect of that communication can often be seen in public polling. When a candidate’s support increases following paid communications, the movement reflects persuasion, but it may also reflect something more basic: voters are finally becoming aware of the candidate as an available and attractive option.

This simple law of political gravity, name identification, buoys some candidates who have it or can create it, while pinning down others fundraising and therefore chances, especially in the eyes of individual donors, endorsers, interest groups, and political action committees.

One political party locked out of the General Election?

A great deal of digital ink has been spilled discussing the possibility that two Republican candidates could finish first and second in the June primary, locking Democrats out of the November general election. That scenario is possible under California’s unique top-two election system. But the level of alarm around it has often reflected a misreading of survey polling. The public survey polling in early 2026 which showed two Republicans with the most support, contained an important detail that was overlooked: undecided voters.

By virtue of a ballot initiative passed by voters in 2010, in California primary elections[2] candidates for a state or federal office all appear on the same primary ballot. The two candidates receiving the most votes advance to the general election, regardless of party affiliation[3]. Two Democrats can advance. Two Republicans can advance. A Democrat and a Republican can advance. Voters do not have a write-in option in the general election for these offices.

This type of election structure heightens the risk of too many candidates from one partisan party affiliation such as the Democrats, splitting their larger partisan voting base while a fewer number of Republican candidates individually amass the highest and second highest amount of ballots cast.

In 2026, this scenario was effectively mitigated by President Donald Trump’s endorsement of only one Republican candidate, then-congressman Eric Swawell’s campaign implosion after criminal allegations from multiple accusers, and other Democratic candidates backing out – like former State Controller Betty Yee’s principled withdrawal from the race. However, the initial fear of only two Republicans advancing was always exaggerated in the public discourse because of those undecided voters.

Any statistically valid survey questionnaire design should allow respondents to select “don’t know” or “undecided” when asked a vote question. Good researchers will often follow up with undecided respondents or push them once more to choose a candidate, but the initial undecided share remains highly informative.

Most public survey polling results were often reported in terms of which candidates were the first, second, third, fourth, etc. highest selected choices by respondents without mentioning the upwards of 1-in-4 respondents who selected “undecided”. Even less attention was paid to the partisan composition of those undecided voters. That omission matters enormously.

If the undecided voters are disproportionately Democratic registered voters, then the apparent Republican advantage among survey respondents who did select a candidate in the polls may be overstated. Those voters may not yet know which Democrat they prefer, but many will eventually cast ballots for a Democratic candidate.

When documents were publicly provided, the S360 research team examined cross-tabulations. We came to the conclusion that because the “undecided” numbers were heavily Democratic registered voters who had not made their minds up yet, they would still eventually cast their ballot for a Democratic candidate. That is very different from saying the undecided voters were genuinely open-minded and available to support any candidates of all party affiliations equally. In a top-two primary, undecideds are not just a blob of uncertainty. They are a pool of voters whose choice is concealed for the time being. Their existence must be recognized and their future action interpreted through party registration, vote history, ideology, geography, ethnicity, age, and other variables that shape actual candidate vote choice.

For context, voters in California must register with one party affiliation option when registering to vote. These party affiliation options include Non-Partisan Preference (NPP) which is effectively “independent”. According to Political Data Inc., the gold standard of voter file data for California, while the registered voter population is:

• 45% Democratic registered voters,
• 25% Republican registered voters,
• 23% Non-Party Preference registered voters, with 7% registered voters of miscellaneous third parties,

The last two midterm primary electorates on average were composed of voters who were;

• 51% Democratic (+7 pp[4]),
• 30% Republican (+5 pp),
• 14% Non-Party Preference (-9 pp), with 5% registered voters of miscellaneous third parties.

This +7 percentage point increase between the registered voter population and the recent historical average for midterm primary voter turnout underscores the fundamental advantage for Democratic candidates – so long as the ballots of the roughly 50% primary voters, who are registered Democratic voters, are not spread across too many Democratic candidates.

That difference is not academic. It means Democratic registered voters and Republican registered voters have historically made up larger shares of the midterm primary electorate than they do of the registered voter population overall, while NPP voters have made up a smaller share. This is why the top-two risk is very real but often overstated in public discourse. A fragmented Democratic field can create danger. But a heavily Democratic likely voter universe, combined with undecided voters who are disproportionately Democratic, changes the interpretation of early toplines.

What is right and wrong about “swim lanes”

Another objective lesson about the dynamics of this primary election is through the concept of political “swim lanes.” Swim lanes are the concept of the stylistic, ideological, and factional lanes within a partisan party base that candidates occupy – whether they intend to or not. An example of this is the electoral aftermath of the then-Congressman Eric Swalwell following the suspension of his gubernatorial campaign and his resignation from Congress.

One of the few publicly available surveys to examine how Swalwell’s previous supporters would reorganize themselves after his campaign suspension was the Emerson College Polling/Inside California Politics survey released in April. The poll found former US Secretary of Health and Human Services, Xavier Becerra, gaining the most visible advantage compared to previous public polling. Becerra overall support rose to 10%, and he posted a 15-point surge among Democratic voters.

Swalwell’s suspension and exit did not simply release a generic bloc of primary voters back into the pool of undecided voters; it appears to have opened space among voters looking for a combative, institutionally connected Democrat with experience in federal office and a posture of direct opposition to Trump-era politics. In that sense, Becerra may have been the more natural successor for a meaningful share of Swalwell’s support than candidates like Mayor Matt Mahan or Speaker Antonio Villaraigosa, even though both can plausibly be grouped within a broader centrist-friendly universe of candidates. The movement toward Becerra is a useful reminder that while voters do pay attention what type of “lane” a candidate represents, the concept is not defined in the way that the media and academia describe “swim lanes”.

Embracing a degree of uncertainty: Researching a “future population”

Considering everything discussed above, what can it tell us about the California gubernatorial horse race based on limited[5] public polling?

• First, look at when the poll was conducted. In a dynamic race with changing candidate fields, fundraising, endorsements, earned media, paid communication, and late-breaking events, older polls can quickly become stale.
• Second, look at the survey population. Is the poll of adults, registered voters, likely voters, or likely primary voters? For this race, the closer the survey population is to likely June 2026 primary voters, the more useful the poll is.
• Third, look at how the survey was conducted. Interviews conducted through multiple methods are generally stronger than interviews conducted exclusively through a single online panel or SMS text approach. Methodological diversity in interview tactics is not magic, but it can help reduce the risk that one mode systematically excludes or over-represents certain types of voters.
• Fourth, look at whether the survey is informed by a voter file. In California, where turnout history, geography, language, age, ethnicity, and party registration all matter, voter file-based sampling and weighting are major advantages for survey polls that use this methodology.
• Fifth, review the margin of error. Then actually use it. Add and subtract the percentage point amount from each listed data point to understand the cone of uncertainty around that candidate’s support. When candidates are close together, their ranges may overlap substantially.
• Sixth, pay attention to undecided voters. Do not simply ask how many undecideds there are. Ask who they are. Their party registration, past turnout, and demographic profile may provide more insight than the top lined results themselves.
• Finally, remember that a June primary and a November general are entirely different electorates, especially compared to the overall registered voter population. Absent a major candidate-specific event, such as criminal allegations or another disqualifying scandal, far more Democratic registered voters will participate in the November general election than in the June primary. That means if one Democratic candidate and one Republican candidate emerge from the primary, the November general election may be relatively anti-climactic in a state like California.

The real question, then, is not simply who is “leading” in an early survey polling. The real question is whether candidates have enough time, enough resources, enough name identification, and enough partisan consolidation to survive the peculiar incentives of California’s top-two primary system.

Footnotes:

[1] This is exactly why many observers expected a previously elected official like Kamala Harris – Former Vice President, Senator, and State Attorney General – to deter others from continuing to run had she entered the gubernatorial race.

[2] Except primaries for U.S. President which remain semi-closed partisan elections, and county, municipal, and other local elected positions whose primary election structure remain defined under local charter law.

[3] Being “locked out” or not advancing any candidates to the general election happens to one party or the other every election cycle in California, though mostly in state and federal district races. 2018 was the last time a statewide general election featured two election candidates of the same party.

[4] pp = percentage point comparison between registered voter population and primary electorate average

[5] Limited in depth of methodology details made available, not in frequency.