Measuring trust: what survey trust items capture, how reliable they are, and how to tell trust from partisanship

Measuring trust means turning a judgment people make about institutions into survey numbers that can be compared over time and across groups. The best-known U.S. measure asks how often people trust the government in Washington to do what is right; Pew Research Center reported 17% answering “just about always” or “most of the time” in 2025, among the lowest readings since the question began in 1958. What such items capture, and how much of it is partisanship, is a long-running research question.

In plain words

“Do you trust the government?” sounds simple. It is not.

Some people answer about the people in charge right now. Some answer about the system as a whole. Many answer partly based on whether their party holds power: trust goes up for the winning side and down for the losing side, and then flips.

Researchers also distinguish trust (what people feel) from trustworthiness (whether an institution actually behaves in ways that deserve trust). A site could make people feel confident without being reliable, or be reliable while still distrusted.

Good measurement uses several questions instead of one, checks that a question means the same thing to different groups, and asks about specific behaviors people can check. This page explains those methods. The broader problem of declining trust in elections is in Trust in elections; how the framework earns trust through verifiable releases is in Published releases and Audits and verification.

Defined terms

• Trust. A judgment that an institution or actor will act in one’s interest or according to its stated rules, even without constant monitoring.
• Trustworthiness. The actual quality of acting reliably and honestly. Levi and Stoker’s review addresses “the meaning of ‘trust’ and ‘trustworthiness’” as distinct concepts.⁠³
• Specific support. Approval of current officeholders and their performance.
• Diffuse support. Attachment to the political system and its institutions regardless of who holds office.
• Political cynicism. Distrust of government, the term used in Miller’s 1974 analysis.⁠⁴
• Reliability. The consistency of a measure, including test-retest reliability (same answer on two occasions).
• Internal consistency. How well several items measuring the same concept agree, often summarized by Cronbach’s alpha.
• Measurement invariance. The property that an item or scale measures the same concept in the same way across groups or time, so that comparisons are meaningful.
• Behavioral trust measure. An observed action that reflects trust, such as using a service or checking a published audit, rather than a stated attitude.

Evidence

The long-running U.S. series and how it is built

Pew Research Center’s fact sheet “Public Trust in Government: 1958-2025,” dated December 4, 2025, reports that “just 17% of Americans now say they trust the government in Washington to do what is right ‘just about always’ (2%) or ‘most of the time’ (15%),” and calls it “one of the lowest in the nearly seven decades since the question was first asked by the National Election Study”.⁠¹ Its data table lists 73% for the 1958 National Election Study reading and 77% for 1964.⁠¹

How the series is assembled is itself a methods lesson. Pew’s chart note states that sources include “Pew Research Center, National Election Studies, Gallup, ABC/Washington Post, CBS/New York Times, and CNN surveys,” and that “from 1976-February 2025, the smoothed trend line represents a three-survey moving average,” while data before 1976 and the most recent number “are from individual polls”.⁠¹ A long trend thus combines different organizations, sampling methods and interview modes, smoothed for readability. Differences between survey houses and between telephone and online administration can shift levels. The trend’s direction over decades is robust to this; a change of a few points between two adjacent readings may not be.

Trust moves with who holds power

The same fact sheet reports that in 2025 “just 9%” of Democrats said they trust the federal government just about always or most of the time, while 26% of Republicans and Republican-leaning independents did, and that a year earlier, under the previous administration, “only 11% of Republicans” said so while “Democratic trust was higher (35%)”.⁠¹ Pew summarizes: “Since the 1970s, trust in government has been consistently higher among members of the party that controls the White House than among the opposition party”.⁠¹

This pattern means a single trust item measures partly specific support for current officeholders. A rise or fall after an election may reflect who won more than any change in institutions.

What drives trust, and what trust drives

Arthur Miller’s 1974 study in the American Political Science Review found that “support of the federal government decreased substantially between 1964 and 1970,” that “increased distrust in government, or cynicism, was associated with reactions to the issues of racial integration and U.S. involvement in the Vietnam war,” and that a “curvilinear relationship” held between policy preference and cynicism, with “cynics of the left” and “cynics of the right” both distrustful.⁠⁴ Distrust, in this account, can come from opposite directions at once.

Marc Hetherington’s 1998 article argued that scholars had “systematically underestimated” trust’s relevance by treating it only as an outcome, and showed that trust “is simultaneously related to measures of both specific and diffuse support,” with trust’s effect on feelings about the incumbent president “even stronger than the reverse”.⁠² His conclusion: “low trust helps create a political environment in which it is more difficult for leaders to succeed”.⁠² If trust is both cause and effect of performance evaluations, simple correlations between trust and anything else are hard to interpret causally.

Levi and Stoker’s 2000 review in the Annual Review of Political Science covers survey-based research on trust together with historical and comparative work on government trustworthiness, organized around trust and participation, trust and the vote, trust and citizen compliance, and political and social trust.⁠³ Their distinction between trust and trustworthiness frames a design question for any new institution: measure both, separately.

International guidance

The OECD’s 2017 Guidelines on Measuring Trust aim “to assist data producers in collecting and reporting trust measures,” “describe best practices in trust measurement, propose a core set of measures to form the basis for international comparisons, and encourage national statistical offices (NSOs) to include measures in their regular household surveys”.⁠⁵ The move toward standard core items reflects the comparability problem described above: if every survey words trust differently, levels cannot be compared.

Trust in elections specifically

Trust in how elections are run is a narrower concept than trust in government. The Survey of the Performance of American Elections, which the MIT Election Data and Science Lab describes as dedicated “to understanding how voters themselves experience the election process,” has been “conducted after each federal election since 2008, when it was launched by the Pew Charitable Trusts,” interviewing “200 registered voters in each of the 50 states (as well as Washington, DC),” with management passing to MIT in 2017.⁠⁶ State-level samples allow comparisons of voter experience and confidence across different election laws, though with 200 per state, state estimates carry wide uncertainty; see Survey sampling.

Reliability: single items are noisy

Attitude items are less reliable than factual ones. Tourangeau, Yan and Sun, analyzing reinterview data from 524 respondents interviewed twice “five to twenty-four days apart,” found that “factual questions (especially demographic questions), shorter questions, questions that did not use scales, those with fewer response options, and those that asked about a noncentral topic produced more reliable answers than attitudinal questions, longer questions, questions using ordinal scales, those with more response options, and those asking about a central topic”.⁠⁷ Trust items are attitudinal and usually use ordinal scales, so they sit on the less reliable side of every contrast.

The standard response is to combine several items into a scale. For k items with variances σ_i² and total score variance σ_X², Cronbach’s alpha is:

alpha = ( k / (k-1) ) * ( 1-sum(sigma_i^2) / sigma_X^2 )

Alpha rises when items covary strongly. It is a property of a set of items in a particular population, not of a concept, and a high alpha does not show that the items measure the intended concept, only that they measure something in common.

Self-reports and social desirability

Ansolabehere and Hersh’s fifty-state vote validation found that “nonvoters who are politically engaged and equipped with politically relevant resources consistently misreport that they voted,” and that “respondents are found to misreport only on survey items associated with socially desirable outcomes”.⁠⁸ Trust questions may carry their own social pressures, such as appearing loyal to one’s side, which is one reason behavioral measures are valuable alongside stated attitudes.

Comparability: measurement invariance

Comparing trust across parties, age groups or years assumes that each item means the same thing to each group. The following is a method description, not a finding. Measurement invariance is tested in stages with multi-group factor models: configural (same items load on the same factor), metric (equal loadings, so a one-unit change means the same thing), and scalar (equal intercepts, so group means can be compared). If scalar invariance fails, a gap in average trust between two groups may reflect different interpretations of the questions rather than different levels of trust. Given the partisan swings reported above,⁠¹ invariance across parties and across administrations is a live question for any trust measure.

Competing interpretations

“Low trust is a crisis.” Long-run decline,⁠¹ and Hetherington’s finding that low trust makes governing harder⁠² support treating trust as a resource in short supply.

“Low trust is often rational.” If institutions have performed poorly or behaved untrustworthily, distrust is an accurate judgment. On Levi and Stoker’s distinction⁠³ the problem to fix may be trustworthiness, not trust.

“Trust items mostly measure partisanship.” The flip in partisan trust with control of the White House.⁠¹ supports this view for general government trust.

“Some trust is diffuse and stable.” Hetherington’s link between trust and diffuse support⁠² suggests trust items also capture system-level attachment that does not reset with each election.

Application to this framework

These are design proposals for My Vote Forward, not results.

1. Measure trust in specific, checkable behaviors. Instead of “Do you trust My Vote Forward?”, ask whether participants believe their ranking was counted as entered, whether they could verify it, and whether the published release matches what they saw. Pair each with a behavioral check: how many participants actually used verification tools. See Audits and verification.
2. Measure trustworthiness independently. Track corrections issued, time to correction, audit discrepancies and incidents, published in Published releases. Trust and trustworthiness are reported side by side so a gap is visible.
3. Use multi-item scales, report reliability. Any trust scale publishes its items, alpha or a comparable statistic, and test-retest results where available, given Tourangeau and colleagues’ evidence on attitude reliability⁠⁷
4. Check invariance across political groups before comparing them. If the framework reports that one group trusts its releases more than another, it first shows that the items behave comparably across those groups.
5. Never report rising trust as success without a comparison. Trust among self-selected participants will be higher than among non-participants. Changes should be compared with a control or benchmark, as in Evaluation.
6. Keep election trust and platform trust separate. The framework is not an official voting service (see Is this an official voting service?), so trust in it is not trust in elections and must not be reported as such.

Unresolved questions

• Which behaviors of a priority platform matter most for earned trust: accuracy of counting, transparency of method, speed of correction, or independence of governance?
• Can trust in a new civic institution be measured without its own partisan coding, when any political topic invites partisan readings?
• Should trust items be asked before and after people see a release, and would asking change their behavior?
• How should the framework respond if participants trust it more than its trustworthiness indicators justify?
• Are there behavioral trust measures, such as voluntary use of verification tools, that are not themselves driven by the most engaged participants?