{"id":"7c73b1dc-760b-45c9-a3fd-d126c783385c","arxiv_id":"2607.21732","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A 20-expert interview study yields a taxonomy of 68 values in nine virtue clusters for ethical data-visualization practice.","lead":"20 interviews with data-visualization researchers, journalists, and artists surface 68 values grouped into nine virtue clusters, offering a shared vocabulary for ethics in visualization. The framework maps tensions like persuasion vs. neutrality that practitioners actually navigate, giving educators and designers more than a don't-lie rulebook.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Self-report gap is acknowledged in §5.2, but the conclusion's 'in practice' claim overreaches; a behavioral check would settle whether the taxonomy is descriptive or aspirational.","rationale":"The reader's weakest assumption identifies the same core risk: retrospective self-reports may not capture enacted practice. I agree that this is the most load-bearing assumption. However, the reader's ACCEPT overlooks that the conclusion (§5.3) states the descriptive claim without the caveat the authors themselves give in §5.2. Since the paper's contribution is a vocabulary for ethical reflection, the concern does not warrant rejection; but the claim should be scoped to 'espoused values' unless a behavioral check is provided. A conditional acceptance with a wording/scoping revision is the appropriate outcome. The proposed observational study is a concrete way to settle whether 'in practice' is accurate; until then, the conclusion should not assert it.","tokens_in":21714,"tokens_out":12539,"duration_ms":123500,"concrete_test":"Run a small observational or think-aloud study: give 8–10 visualization practitioners a realistic design brief with competing pressures (tight deadline, a client request to truncate a y-axis or omit uncertainty, a low-data-literacy audience). Independently code their design decisions and final artifacts against the 68-value framework. If enacted values diverge systematically from the interview-derived pattern — e.g., efficiency and persuasion dominate while transparency and honesty rarely appear — the §5.3 'in practice' claim is not supported and should be revised to 'espoused by experienced practitioners.' If the enacted values align with the interview pattern, the concern is resolved.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that the 68 values and 9 clusters 'reflect matters of concern that visualization experts consider in practice' (§5.3). The supporting evidence is 20 retrospective interviews (§3.2). The load-bearing assumption is that what experts say they value in a one-hour interview corresponds to the values that actually operate in real design work. The authors themselves flag this: 'what people (even experts) say they value is one thing: it remains to be seen to what extent these values are embodied or prioritized in practice' (§5.2). The interview protocol includes the direct question 'Do you think your ethical values ever influence the visualizations that you make?' — this invites espoused ideals, social-desirability responses, and hindsight bias rather than observed trade-offs. If the gap is large — e.g., under deadline or client pressure practitioners routinely sacrifice the honesty/transparency values they espouse — then §5.3 overstates the result. The generative vocabulary survives, but the descriptive 'in practice' framing fails. Secondary: 14/20 transcripts were single-coded without inter-rater reliability (§3.3), so the frequency counts in Fig. 3 (e.g., N=19) are illustrative, not robust.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a qualitative interview study with 20 experienced data visualization researchers, practitioners, journalists, designers, and educators. Through semi-structured interviews and thematic analysis, the authors derive 68 values organized into nine 'virtue clusters,' and identify four recurring tensions (subjectivity vs. objectivity, data vs. truth, persuasion vs. neutrality, engagement vs. efficiency). The stated contribution is a generative descriptive vocabulary for ethical reflection in data visualization, intended to broaden the field's focus beyond avoiding deception. The paper includes an explicit limitations section acknowledging the subjectivity of the analysis, the Western-centric sample, and the gap between espoused and enacted values.","tokens_in":21968,"tokens_out":8950,"duration_ms":87295,"significance":"If accepted as a descriptive framework, this paper would provide a valuable resource for visualization ethics pedagogy and practice. Its strengths include a transparent interview protocol, a coding procedure with an initial dual-coded calibration phase, extensive use of participant quotes, and a supplemental table of value definitions. The paper is careful to frame the taxonomy as non-exhaustive and generative rather than prescriptive. The identification of recurring tensions experienced by practitioners is a useful contribution that goes beyond simple lists of dos and don'ts. The explicit acknowledgment of the authors' own positionality and the self-report limitation adds credibility. However, the central claim as stated in the conclusion slightly overreaches the evidence, and some quantitative presentations of qualitative coding results are not fully supported by the stated methodology.","major_comments":[],"minor_comments":[{"comment":"The conclusion states that the framework 'reflects matters of concern that visualization experts consider in practice.' This phrasing is stronger than the evidence supports, since the data come from retrospective interviews rather than observations of practice. The authors themselves acknowledge in §5.2 that 'what people (even experts) say they value is one thing: it remains to be seen to what extent these values are embodied or prioritized in practice.' Please revise the abstract and conclusion to say 'report considering in their professional work' or 'espouse' rather than 'consider in practice,' to stay consistent with the stated limitation.","section":"§5.3 (Conclusion) and Abstract"},{"comment":"The coding procedure used dual-coded calibration for six transcripts and single coding for the remaining 14, and the authors state the codes are 'not strictly amenable to quantitative measures such as inter-rater reliability.' Yet Fig. 3 reports per-value transcript counts, and §5 highlights 'Nearly all (N=19) interviewees' for two values. Without any reliability check, these counts are difficult to interpret. I suggest either explicitly labeling the counts as illustrative of the diversity of responses rather than robust prevalence estimates, or softening the 'nearly all' claim.","section":"§3.3 and Fig. 3"},{"comment":"Minor typographical and style issues: the abstract contains 'V Y et' instead of 'Yet'; §1 has 'intension' where 'in tension' is meant; §5 has 'pedogogical' instead of 'pedagogical'; §5.1.4 has 'unambigious' instead of 'unambiguous.' Please proofread carefully.","section":"Throughout"},{"comment":"The participant table and Figure 2 report professional areas and engagement frequency, but the paper does not report the distribution of participants across the three continents mentioned in the text. A sentence or small table clarifying the geographic spread would help readers assess the diversity claim, especially given the acknowledged Western-centric limitation in §5.2.","section":"§3.1"}],"recommendation":"minor_revision","confidential_remarks":"This is a solid qualitative contribution. The main issue is that the conclusion's 'in practice' phrasing overstates what retrospective interviews can support; the authors already have the right caveat in §5.2 and simply need to align the abstract and conclusion with it. The coding-reliability concern is real but secondary, and can be addressed by softening the quantitative-sounding claims. I would not require new data for this revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's the short version: this is a solid, honest qualitative study that gives visualization ethics a useful working vocabulary. It reports 68 values in nine clusters from 20 expert interviews, and it does not oversell what it did. If you work in critical visualization or teach it, you'll want this on the reading list.\n\nWhat is actually new: there have been prescriptive calls (Correll 2019), principle cards (Wang et al.), and text analyses of assumptions (Saharan et al.), but this is the first interview-based elicitation of what practitioners, researchers, and artists say they value. The taxonomy is grounded in extensive participant excerpts, and the authors are careful to present the clusters as affinities, not an ontology. The recurring tensions (objectivity vs. subjectivity, data vs. truth, persuasion vs. neutrality, engagement vs. efficiency) come out of the interviews and will be genuinely useful for design critique.\n\nWhat is good: the methods are clearly described. They ran a focus group to seed codes, dual-coded a subset for calibration, then single-coded the rest. They explicitly say the codes are not amenable to inter-rater reliability, and they don't pretend otherwise. The limitations section is unusually candid: they flag their own positionality, the small Western-leaning sample, and the gap between what experts say and what they do. The stress-test concern about self-report validity is really just the authors' own caveat restated; they write that 'it remains to be seen to what extent these values are embodied or prioritized in practice.' The only wobble is the conclusion's phrase 'reflects matters of concern that visualization experts consider in practice' — that is a slightly stronger claim than the data support, since the interviews capture espoused values, not observed trade-offs. But it's a word-choice issue, not a load-bearing flaw; the study's value as a generative vocabulary does not depend on it.\n\nMinor soft spots: N=20 with three continents but still mostly Western; Fig. 3 frequencies are illustrative and should not be read as population estimates. None of this undermines the central contribution.\n\nBottom line: this deserves a serious referee and likely a place in the HCI/visualization ethics canon. I'd bring it to a reading group and would cite it.","headline":"A careful, honest qualitative study that delivers a genuinely useful vocabulary for visualization ethics; the only real wobble is the conclusion's 'in practice' phrasing, and that is a word-choice issue, not a load-bearing flaw.","tokens_in":22458,"tokens_out":2006,"would_cite":true,"duration_ms":21128,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Ethical data visualization is not just avoiding lies; it is a practiced craft of 68 values in nine clusters, and this paper maps them from 20 expert interviews.","keywords":["data visualization ethics","virtue ethics","values","expert interviews","qualitative coding","phronesis","ethical tensions","visualization practice"],"falsifier":"An observational study that follows visualization designers through real projects—recording decisions under deadline, client, and budget pressure—and then compares those decisions against the 68 listed values. If designers' choices repeatedly contradict their stated values (for example, accepting y-axis truncation when a client demands it), then the taxonomy describes espoused ideals rather than enacted ethics.","tokens_in":21602,"feed_emoji":"📊","tokens_out":4152,"duration_ms":37832,"temperature":0.7,"pith_summary":"The paper sets out to capture what experienced data visualization practitioners actually value and worry about, rather than relying only on rulebooks that say what to avoid. Through 20 semi-structured interviews with researchers, journalists, designers, and data artists, the authors identify 68 values and group them into nine 'virtue clusters'—from professional wisdom and honesty to fairness, care, and aesthetic appeal. They argue that ethical visualization is less about obeying a list of prohibitions and more about building practical judgment, a moral skill acquired through experience. If correct, the field gains a shared vocabulary for ethical reflection and a clearer picture of the tensions—such as persuasion versus neutrality, or engagement versus efficiency—that practitioners navigate daily. This matters because ethical guidance that only says 'don't deceive' leaves designers without language for the harder, more common choices.","feed_headline":"Interview study maps 68 ethical values in data visualization","feed_subtitle":"Nine virtue clusters and four recurring tensions give practitioners a richer language for ethical reflection.","key_machinery":"The load-bearing object is the interview-derived taxonomy: 68 values affinity-diagrammed into nine virtue clusters, with each value tagged as applying to the visualization object, the designer, or both. The authors define a 'virtue' as an ethical matter of concern and a 'value' as any concern a person considers important; the distinction lets them avoid judging which concerns are genuinely virtuous. The taxonomy is what carries the argument that ethical work spans character, craft, and care.","core_discovery":"The central claim is descriptive and generative: the ethical concerns of data visualization practitioners are far richer than the usual injunction to avoid misleading charts. Based on interviews with 20 experts, the paper reports 68 values, sorted into nine virtue clusters that apply either to visualizations themselves, to their designers, or to both. It also surfaces four recurring tensions—subjectivity versus objectivity, data versus truth, persuasion versus neutrality, and engagement versus efficiency—that practitioners negotiate without fixed answers. The paper concludes that ethical visualization is a matter of practical wisdom built through experience, not a matter of learning rules.","pith_inferences":["The same interview approach could be replicated in adjacent fields such as data journalism or AI interface design to test whether the nine clusters generalize across technical practices.","The nine clusters could be converted into a card deck or reflective tool for design teams, similar to existing ethics card methods, to make the values actionable during design.","The four tensions suggest a testable hypothesis: design decisions made under deadline or client pressure will systematically sacrifice the 'care' values, such as situatedness and empathy.","The near-universal mention of situatedness and mindfulness of bias suggests a generational shift in the field's self-conception, but whether this shift is real or only linguistic needs observational validation."],"forward_implications":["Ethics education in visualization can move from checklists to case-based discussion of virtues and tensions.","Researchers and evaluators gain a broader set of criteria for judging 'good' visualization beyond effectiveness and honesty.","The four tensions give instructors concrete material for teaching ethical judgment.","The taxonomy opens the door to empirical studies of whether espoused values are actually enacted in practice."],"fun_headline_variants":["Map of data viz ethics: 68 values, 9 virtue clusters","Data viz pros reveal 68 values and 4 ethical tensions","The full moral universe of data viz: 68 values","Interviewing 20 experts: 68 values for ethical viz"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The taxonomy rests on the assumption that what experienced practitioners say they value in retrospect accurately reflects the values that actually shape their visualization decisions in practice.","fun_headline_variants_meta":{"raw":{"variants":["Map of data viz ethics: 68 values, 9 virtue clusters","Data viz pros reveal 68 values and 4 ethical tensions","The full moral universe of data viz: 68 values","Interviewing 20 experts: 68 values for ethical viz"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000188,"raw_usage":{"total_tokens":1178,"prompt_tokens":765,"completion_tokens":413,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":509,"completion_tokens_details":{"reasoning_tokens":353}},"tokens_in":509,"tokens_out":413,"duration_ms":4166,"temperature":1.0,"reasoning_tokens":353,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T06:50:56.610744+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An observational study that follows visualization designers through real projects—recording decisions under deadline, client, and budget pressure—and then compares those decisions against the 68 listed values. If designers' choices repeatedly contradict their stated values (for example, accepting y-axis truncation when a client demands it), then the taxonomy describes espoused ideals rather than enacted ethics.","supporting_citations":[],"review_version":1}