Opinion and choice

In plain words

An opinion is what you think about a specific idea. A vote is a final choice between the options on an official ballot, made in private. They are not the same thing.

People often hold mixed views. You might like one part of a proposal and dislike another. You might support an idea but vote against the version on the ballot because of its cost or wording. You might change your mind after learning more. None of that means you lied.

Polls and surveys also make mistakes for reasons that have nothing to do with honesty, such as who chose to answer. So when results and votes differ, the first question is about measurement, not about character.

This framework keeps opinions and votes apart on purpose. You can choose to share an opinion. Your vote stays secret, and there is no feature that can show it to anyone. That rule comes from long experience with public voting, described in The secret ballot.

Defined terms

Opinion. A view about a specific, worded proposition. In this framework an opinion is optional to share, can be revised, and is always tied to the exact question (see Opinions).

Preference. A broader ordering of what a person wants, which may be expressed through many opinions and priorities.

Ballot. An official choice among eligible candidates and measures, cast under election law. In this framework the ballot is a separate act with requirements of its own (see Private voting).

Secret ballot. A voting method in which the content of a person's vote is not disclosed or linkable to them. International human rights law describes elections "held by secret ballot, guaranteeing the free expression of the will of the electors"¹.

Misreporting. A survey answer that does not match an external record, such as a voter file. Misreporting can result from memory error, social pressure, or misunderstanding, and it is measurable only where records exist².

Measurement error. Differences between an observed answer and the underlying view caused by the instrument, such as a single ambiguous question³.

Person-level ballot linkage. Any record, method, or inference that associates a specific person with the content of their ballot.

The framework's stable distinctions between opinion, ballot, and ranking are in the glossary.

Evidence

The law separates opinion from the ballot

Article 19 of the International Covenant on Civil and Political Rights (adopted by the UN General Assembly on 16 December 1966, in force from 23 March 1976) provides that "everyone shall have the right to hold opinions without interference" and protects freedom of expression. Article 25 separately provides for elections by secret ballot that guarantee the free expression of the will of the electors¹. The two rights sit in different articles because they protect different things: open expression on one hand, and an unpressured private choice on the other. National law varies, and the Covenant is not a description of any one country's procedures. See Jurisdictions.

Stated and actual behavior can differ without lying

Ansolabehere and Hersh conducted what they describe as the first fifty-state validation of survey vote reports against public voter records. They found that politically engaged nonvoters with politically relevant resources consistently reported having voted when records showed they had not, that faulty registration records did not explain the pattern, and that misreporting appeared on items with socially desirable answers. They concluded that survey-based studies can substantially misestimate differences between voters and nonvoters². Pew lists voting among behaviors people tend to overstate, alongside church attendance and charitable giving, and uses a face-saving question wording for that reason⁴.

This evidence concerns whether people voted, which can be checked against public records. It does not concern how they voted, which in a secret ballot system cannot be checked against any individual record. That limit is intended.

Poll and result gaps are often about who answers

After the 2020 U.S. general election, the AAPOR task force found polling error of unusual magnitude across presidential, Senate, and governor races. It concluded that error "was not primarily caused by respondents' reluctance to tell interviewers" their choice, noting that the overstatement occurred regardless of interview mode and was larger in races where that candidate was not on the ballot. It added: "among those who chose to respond to polls there is no evidence that respondents were lying," and that "a separate and likely problem is that some people chose not to respond to polls at all"⁵. The task force also said the exact cause could not be identified conclusively with available data⁵. The point here is not about any party or candidate. It is that a gap between stated preferences and results is, in the best available analysis of a recent large case, better explained by who participated than by dishonesty.

Single questions understate stable views

A long tradition in survey research held that many voters have unstable, incoherent opinions. Ansolabehere, Rodden, and Snyder argued in the American Political Science Review that much of this apparent instability is measurement error tied to individual survey items. Averaging many items on the same broad issue area produced issue preferences that were "well structured and stable," with stability approaching that of party identification, and those preferences explained vote choice much better³. If single questions are noisy, then a gap between one opinion answer and a later ballot choice is weak evidence of anything about a person's honesty.

Methods exist for sensitive questions, with limits

Survey researchers use indirect techniques for sensitive topics. The list experiment (item count technique) asks respondents how many items on a list apply to them, rather than which ones, and compares groups that did and did not see a sensitive item. Imai describes it as "a survey methodology that is designed to elicit respondents' truthful answers" to sensitive questions, and notes that indirect questioning "means loss of information"⁶. Blair and Imai show that the method depends on assumptions that must be tested, and they develop methods "to detect and adjust for potential violations of key assumptions"⁷. These methods estimate group-level patterns. They do not reveal any individual's answer, which is part of their value and part of their limit.

Competing interpretations

Gaps reveal concealment. A familiar public explanation for surprising results is that people hide their real views from pollsters or from each other. Social desirability effects are real for some behaviors²⁴, so this explanation is not baseless.

Gaps reveal measurement and participation. The 2020 AAPOR analysis found no evidence of lying among those who responded and pointed to nonresponse instead⁵. Measurement research shows single items carry substantial error³. Under this view, concealment is one possible contributor among several and should not be the default assumption about any individual.

Opinions and ballots should match; mismatch is a problem to fix. Some designs aim to make expressed opinion predict voting closely.

Opinions and ballots should be allowed to diverge. A ballot measure is a whole text with costs and side effects; a candidate contest bundles many issues. A person can support a goal and oppose a specific measure. The framework takes this second view: divergence is normal and the system is not built to detect or reduce it.

Transparency of votes would increase accountability. Historically, open voting was defended in those terms.

Secrecy protects against coercion. The modern consensus, reflected in Article 25¹, is that secrecy is required so that the vote reflects the voter's will rather than pressure from employers, households, or others. The history is in The secret ballot, and current coercion risks are in Intimidation.

Application to this framework

These are design rules and proposals for the framework. They are not claims that any system meets them today.

1. Three separate acts, three separate data classes. Priorities, opinions, and ballots are stored and governed differently (see Privacy). An opinion shared publicly is never treated as a ballot, and a ballot is never treated as an opinion.
2. No person-level ballot linkage, by design and by policy. No feature, export, research product, or administrator view connects a person to ballot content. There is no ballot share button and no public vote history. A vote's privacy cannot be switched off. Requirements for doing this in an authorized election, including secrecy-compatible verification, are in Audits and verification; the current websites conduct no official votes.
3. No inference from opinion to ballot. The framework does not predict, score, or label how an individual voted from their opinions or priorities, and does not report "consistency" between a person's opinions and any vote.
4. Disagreement is not evidence of dishonesty. Publications never describe gaps between aggregate opinion results and official outcomes as lying. They name measurement and participation explanations first, consistent with Measurement.
5. Revision is a normal act. Changing a shared opinion creates a new version; it does not trigger warnings, penalties, or public flags. See Opinions and Representation on explained changes of position.
6. Group-level sensitive measurement only. If future research uses indirect methods such as list experiments, it would report group estimates under a published method, never individual results, and would not be run on ballot content.

Hypothetical example. A participant publicly supports "more funding for parks" and later, in an authorized election, a parks bond fails in their district. The framework does not and cannot tell anyone how that participant voted. A release might note that aggregate support for the goal was higher than support for the specific bond, and would list possible reasons (cost, bond structure, who participated) without assigning motives to anyone.

Unresolved questions

• How can a system make it obvious to users that opinions are shareable and ballots are not, so that no one mistakes one for the other under pressure?
• If a coercer demands proof of a public opinion as a stand-in for a vote, how far can design reduce that risk? See Intimidation.
• Can aggregate opinion and official results be compared responsibly without inviting accusations against groups?
• What research methods for sensitive opinions are compatible with the framework's consent rules in Consent?