Resonant vs World Values Survey Is the Wrong First Question

A valid dataset comparison must examine agency, context, and decision provenance—not just topics, reach, or outcome accuracy.

The consequential difference between two human-data projects is not how many values they cover, but whether each record preserves the structure of a decision: the alternatives considered, the context supplied, the person’s authority, and the eventual choice. The numbered evidence supports criteria for making that distinction, but it provides no documented schema, sampling method, questionnaire, or provenance for either Resonant or the data associated with The Economist and World Values Survey . A direct claim that one dataset is broader, more representative, or more predictive would therefore exceed the evidence. What the sources do permit is a sharper comparison standard: determine whether each project records positions people report or decisions they singularly make under specified constraints. That distinction matters because research can preserve outcome quality while overlooking internal search structure, including how attention and exploration were allocated before the result appeared . The honest first question is consequently not “Which dataset is better?” but “Which aspects of human judgment remain recoverable after a response becomes a row?”

A dataset can contain the same final answer while representing radically different cognitive events. Research on algorithmic business decisions distinguishes structural balance in a search process from the quality of its outcome, finding that improved symmetry and convergence stability can coexist with preserved accuracy and bias reduction . This means an apparently good choice does not reveal whether alternatives received balanced consideration, whether one option dominated prematurely, or whether the search process was distorted and merely arrived at the right destination . Applied to a Resonant–World Values Survey comparison, both sides would need to disclose whether they retain only selected responses or also capture the organization of the choice that produced them. A record of agreement with a proposition, a forced choice between alternatives, and a revision after reflection may all end at the same coded position while supporting different inferences about the person. Outcome-only comparison would collapse those distinctions, even though the cited optimization research treats internal allocation and final quality as analytically separable .

Agency creates a second dividing line because a decision is not fully described by its content; it also has an owner and a moment of commitment. One theoretical framework separates substantive agency into coherent plurality, causal openness, the absence of a uniquely pointed-to answer before action, act-level singularisation, and endogenous sourcehood . On that account, alternatives must be more than decorative: the act itself resolves a live plurality rather than merely reporting a selection fixed elsewhere . The framework argues that exact passive pre-act selection conflicts with retaining both the priority of the singularising act and invariance under causally irrelevant informational additions . Yet the author expressly does not prove that human decisions satisfy the bridge and certification premises required for the paper’s computational result, so its framework should be used as a diagnostic rather than as evidence that any particular dataset contains irreducible agency . A valid comparison should therefore ask whether responses are treated as static attributes of respondents or as authored resolutions whose timing, alternatives, and conditions are part of the data.

Context is not nuisance metadata when the object of study is judgment; it can determine what a response means. Research on digital financial decisions reports that literacy and perception of crisis-related risks vary with generational and educational factors, while also arguing that conventional knowledge about saving, borrowing, and budgeting is insufficient for an increasingly digital financial environment . In clinical decisions, professional role likewise moderates vulnerability to low-quality AI recommendations, with nurses and midlevel providers more susceptible than physicians and pharmacists in the reported experiments . These findings show why two identical answers cannot automatically be interpreted as equivalent evidence of preference, competence, or risk orientation . For both the Resonant side and the Economist/World Values Survey side, a credible comparison would need to identify which contextual variables are retained, which are deliberately excluded, and which populations the records can support conclusions about. Breadth without contextual resolution can enlarge a dataset while weakening the meaning of each comparison, particularly when education, generation, and domain knowledge alter how people recognize or respond to risk .

Decision provenance also depends on the workflow surrounding a response. In a Propose–Verify–Decide system tested on virtual surgical-instrument sorting lines, 18 of 28 workflow cases met expectations, while autonomous strategy-workflow success was only 3 of 10 and invalid inputs were correctly rejected in 7 of 8 cases . All four cases that passed preceding checks, produced complete evidence, and reached final engineering review passed that review, whereas failures involved semantic distortion, incomplete evidence, and missed invalid inputs . The small number reaching final review does not establish broad reliability, but it reveals a useful mechanism: evidence gates can make the provenance of accepted decisions more important than the raw number of generated outputs . A dataset comparison should accordingly disclose whether participants can reconsider answers, whether invalid or inattentive responses are rejected, whether the presentation changes between stages, and whether a final record retains its revision history. Without those details, response volume mixes considered decisions, misunderstood prompts, incomplete evidence, and accidental selections into a category that the workflow study shows should not be presumed homogeneous .

Trust and perceived quality provide an especially dangerous basis for declaring one human-decision dataset superior. A study of 33 experienced professionals found a moderate positive association between AI use and reported decision-making effectiveness, including accuracy, speed, and operational efficiency . Another collaborative framework reported 91% decision accuracy, mean trust of 6.3 out of 7, and a 52% reduction in decision time across collaboration cycles, with AI ownership concentrated more heavily in analytical than ethical or judgment stages . Yet clinical experiments found that recommendation quality and domain knowledge mattered more than explanation style, while local explanations increased trust without improving decision quality and trust itself was largely unrelated to clinical quality . A separate scenario experiment found that analytic structuring increased psychological empowerment and self-reported trust regulation, which were associated with perceived decision quality . These results are not a clean contradiction because one study measures clinical decision quality while another examines perceived quality in a scenario, but together they warn against treating trust, confidence, and correctness as interchangeable . A Resonant–World Values Survey comparison should therefore separate how convincing or empowering a prompt feels from whether its response is stable, informed, contextually valid, or behaviorally consequential.

The practical comparison should be built as an audit of inferential reach rather than a contest between brand names. Each side should document its unit of observation, live alternatives, prompt context, revision opportunities, respondent authority, exclusion rules, population coverage, and the claims that can legitimately be made from the resulting record; these requirements follow from the demonstrated separability of process and outcome, the contextual variation in risk perception, and the workflow dependence of accepted decisions . It should also distinguish augmented, hybrid, and autonomous distributions of cognitive authority, because managerial research identifies these as different patterns rather than interchangeable forms of decision support . Where machines participate, recommendation quality and domain expertise should be reported independently of explanation and trust, since greater trust did not rescue low-quality clinical advice . Where human agency is the intended object, the comparison should state whether it captures genuine resolution among alternatives or only a response after the operative selection may already have been shaped elsewhere . Until the schemas and collection protocols for Resonant and the Economist/World Values Survey data are supplied, the defensible conclusion is not that one is richer, but that richness must be judged by how much of a decision survives its conversion into data.

Decisions · Articles · The Trolley Problem · Game Theory · Human–AI Alignment · Resonant