In the late 1980s, when my firm, Price Research, prepared a voter survey for the Bakersfield City School District, we began with a basic administrative record from the Kern County elections office. To clean up addresses and append phone numbers, we turned to Political Data Inc. (PDI)—then a young company operating out of two converted dressing-room trailers on the Warner Bros. lot in Burbank. PDI helped us find and organize voters. But to learn what those citizens actually thought about school facilities, taxes, and local bonds, we still had to do something simple: we had to ask them.
Today, that basic administrative list has become the foundation of a multi-million-dollar commercial enterprise. Major vendors such as L2, TargetSmart, Aristotle, i360, and Data Trust assemble public records into vast dossiers, appending anywhere from 600 to over 2,500 individual data points per record—income estimates, housing specifications, vehicle makes, magazine readership, firearm licensing, travel habits, and predictive behavioral scores. What began as a tool for locating voters has evolved into a system for estimating them.
This evolution has created an acute democratic tension. Citizens register to vote to exercise a civic right. Few realize that this single public registration record becomes the anchor for a commercial profile packed with data they never provided and political judgments they never expressed. The public nature of a voter roll should not place commercial inferences beyond civic accountability.
Two Different Kinds of Information
Modern voter files conflate two fundamentally different categories of data:
- Verified Vote History: Official election records showing turnout in previous primary, general, or special elections. This administrative paper trail allows vendors to categorize citizens as high-, medium-, or low-propensity voters based on actual civic conduct.
- Predictive Models: Algorithmic scores derived from commercial consumer databases estimating candidate preference, partisan lean, or policy stances.
One measures a verified civic act; the other constructs a commercial prediction. Because the secret ballot protects actual choices, official records show how often a citizen participated, but never how they voted. Everything else—including candidate scores and ideology rankings—is simply a vendor’s educated guess.
According to federal Election Assistance Commission data, more than 40 percent of voter registration transactions now occur through routine "Motor Voter" interactions at state DMVs. Completing that errand does not mean a citizen understands how their record will be enriched, nor does it provide them any mechanism to inspect the resulting profile. Political organizations routinely purchase and act upon commercial claims about citizens, while the citizens themselves are denied any opportunity to review or challenge them.
Independent Voters and the Fallacy of Binary Modeling
Independent voters expose the flaws of this machinery most clearly. Gallup’s long-term polling found that 45 percent of U.S. adults identified as political independents in 2025. While many lean toward one party, they deliberately claim an independent identity; for them, leaning preferences and independent self-conception coexist.
These self-identified independents overlap with—but are distinct from—formally registered unaffiliated voters. Under Oregon’s automatic voter registration system, for instance, citizens are registered as nonaffiliated unless they affirmatively select a party. Others deliberately opt out of party registration. Neither path reveals where they stand on specific candidates or ballot measures.
Yet an official registration marked "unaffiliated" routinely acquires a commercial label of "likely Democrat" or "likely Republican." That label then dictates whether and how campaigns engage the voter.
Consider an unaffiliated voter in Colorado who typically votes Republican but strongly favors a local school bond. A commercial algorithm branding them "likely Republican" may lead a bond campaign to write them off entirely. Conversely, an unaffiliated voter who leans Democratic might oppose a municipal tax measure; if partisan lean is treated as a catch-all proxy for policy stance, campaigns will miss the specific concern driving their vote.
A flawed estimate does more than waste ad dollars. It sidelines citizens from civic outreach or subjects them to appeals built around beliefs they do not hold. Most never discover why they were targeted—or ignored.
From Polling People to Predicting Them
At Price Research, we specialized in California school bond and tax measures governed by the state’s stringent two-thirds approval requirement. Survey inference was central to our work, but an answer supplied directly by a voter and a conclusion drawn by an analyst remained fundamentally distinct. We asked voters about project priorities, tax thresholds, and counter-arguments before revisiting the ballot question. Even voters who shared an initial response rarely arrived at it for the same reason. "Undecided" reflected a temporary position, not the underlying values shaping it.
Modern political analytics operates on an entirely different scale. As political scientists David Nickerson and Todd Rogers noted in their landmark 2014 study on campaign big data, survey responses from 1,200 individuals are routinely leveraged to project preferences onto millions of voter records. The 1,200 supply real answers; the millions receive algorithmic assignments. That statistical scaling is technically impressive, but it is no substitute for democratic dialogue.
Empirical research reinforces why independent voters defy simple modeling. A North Carolina study by J. Michael Bitzer, Christopher A. Cooper, Whitney Ross Manzo, and Susan Roberts revealed that unaffiliated registrants consistently differed from major-party registrants in both attitudes and voting behavior. Among more than 207,000 unaffiliated voters who participated in three consecutive presidential primaries (2012, 2016, and 2020), nearly half—47 percent—voted in different parties' primaries across those cycles. While primary participation does not dictate general election behavior, it demonstrates an electoral fluidity that static party-lean scores in commercial voter files invariably blur or erase.
Similarly, a 2026 study by Philip Moniz, Kyle Endres, and Costas Panagopoulos confirmed that commercial policy predictions deteriorate rapidly when a voter’s views diverge from orthodox partisan lines. While their study excluded nonaffiliated voters and thus did not prove independents are uniquely unpredictable, it demonstrated that partisan heuristics are unreliable guides to specific policy views.
Data vendors themselves recognize these vulnerabilities. A GoodParty.org campaign guide cautions candidates to validate commercial scores through real door-to-door conversations. That is sound advice for the campaign buyer; the profiled citizen deserves a fair mechanism to inspect those claims as well.
Giving Citizens a Voice in Their Dossiers
Campaign access to public voter files and vendor accountability to the electorate need not be mutually exclusive. States can balance both by explicitly bringing commercial political data brokers under state consumer data privacy frameworks. Such rules should apply strictly to commercial vendors licensing compiled dossiers to multiple campaigns, leaving internal party and volunteer databases distinct.
A clear precedent already exists. In a 2022 formal opinion, the California Attorney General affirmed that qualifying inferences drawn by a covered business fall under the California Consumer Privacy Act’s (CCPA) "right to know"—even when those inferences stem from public records. This precedent provides a legal framework for treating commercial inferences separately from the public voter rolls used to generate them.
A comprehensive, administratively feasible transparency framework—mirroring consumer data protection standards in credit and employment reporting—should establish four baseline standards:
- Data Provenance Labeling: Every field in a political file must clearly state its origin: administratively recorded (e.g., official turnout), self-reported (e.g., donor records or direct survey responses with dates), commercially appended (e.g., credit bureau or vehicle data), or modeled (e.g., ideological or candidate scores).
- Score Plain-Language Standards: Predictive metrics must disclose what they measure, explain their numerical scale (a score of "60" is meaningless without context), and report their production date.
- Right of Inspection and Correction: Citizens should have secure, verifiable portals to review their profiles and correct factual errors in commercially appended source records. (Factual errors originating in official election records would remain subject to standard correction procedures through local election authorities.)
- Contested Inferences: While a citizen cannot simply demand that a vendor delete a proprietary model, a voter should have the right to register a formal dispute against modeled labels. Whenever that profile is licensed, a dispute flag—or the voter's self-reported affiliation—must accompany the record.
Vendors can satisfy these requirements by disclosing general modeling methodologies and input categories without disclosing proprietary source code or model weights. State privacy regulators already oversee similar disclosure standards across credit and employment reporting.
Accountability to the Citizen
These protections will not eliminate campaign missteps or stop every targeted political ad. They will, however, give citizens a practical mechanism to inspect and challenge consequential commercial assertions made about their civic lives.
Forty years ago, the administrative voter file was used simply to locate citizens so campaigns could talk to them. Today's commercial dossiers decide what to expect from citizens—and whether they are worth talking to at all. That shift demands accountability not just to the political clients buying the data, but to the sovereign citizens described within it.
Independent voters do not need to be easily predictable to deserve democratic respect. Americans deserve to be engaged as citizens who can speak for themselves, not as commercial estimates engineered to speak for them.
Kent C. Price is a retired political science professor with four decades of experience in applied public‑opinion research, including extensive work on large two‑thirds revenue elections in California.
