Voting Records

All data comes from the OHCHR Search Library — Voting collection, harvested in full (… records — matching the collection's own reported total) via its MARCXML export. It covers every voting record catalogued for the Commission on Human Rights (1946–2006) and its successor, the Human Rights Council (2006–present).
• The 2006 institutional break — the Council did not merely rename the Commission: it introduced new membership, election rules, the Universal Periodic Review and a different sessional rhythm. Because voting dynamics before and after differ substantially, every time-axis chart carries a dashed HRC reference line at 2006; treat trends read straight across it with care.

The Blocs & alignment tab uses three standard measures from the study of legislative and UN voting, all computed client-side from the recorded roll-calls (no external data):
• Alignment map — a reference-pole scatter: each state is placed by its voting agreement with two anchor states you choose (horizontal = agreement with the first, vertical = agreement with the second, each 0–100% over shared recorded votes). Positions are directly interpretable; the dashed diagonal marks equal agreement with both anchors. The axes are drawn to an identical scale on a square plot over the full, uncropped 0–100% domain — so the diagonal is exactly 45° and distance from it reads directly as a state's net leaning toward one anchor.
• Regional-group cohesion — the Agreement Index (Hix, Noury & Roland 2005): for each resolution, (largest bloc − ½·the rest) ÷ total among a group's members, averaged over the window; 1 = unanimous, 0 = evenly split.
• Most often outvoted — the share of a state's votes cast against the side that prevailed, over states with at least 20 votes in the window. The prevailing side is whichever of Yes or No is larger in the official whole-chamber totals: abstention is not a side that can win, so a plurality of abstentions is not treated as a verdict, and a Yes/No tie leaves no prevailing position (those resolutions are skipped rather than tie-broken by column order). Excluding abstentions from the winning side is not an arbitrary choice — it mirrors UN practice itself: under General Assembly rule 86, a member that abstains is "considered as not voting" and only Yes/No votes count toward the majority required (1986 UN Office of Legal Affairs opinion, reproduced in UNITAR NYO, Guidelines for United Nations Resolutions, 2020, Annex I). CHR/HRC operate under their own, smaller-body rules rather than the GA's, but the same "abstention ≠ a position that can win" convention holds — see the caveat below on transferring GA-specific findings. Because a 62-vote record and a 953-vote record are not equal evidence, every bar carries its 95% Wilson interval and its vote count. It measures being on the losing side, which is not the same as being isolated — a state can be in a large, cohesive minority bloc.

The Consensus tab reads the mode of adoption recorded in the source catalogue (MARC 591) as a diplomatic signal in its own right — a recorded vote signals that consensus could not be reached, and a withdrawal marks a text that never reached a decision.
• Modes — adopted without a vote (blue; commonly called consensus), recorded vote (red), non-recorded vote / non-recorded unanimous / no information (grey, mostly Commission-era), withdrawn (amber) and not considered. "Consensus share" uses considered texts only — withdrawn and not-considered drafts are excluded from the denominator.
• Caveat — adoption without a vote is not unanimity: delegations may dissociate from consensus in explanations of position after adoption, which the catalogue (and therefore this tab) does not capture.
• Flip heuristic — subjects with ≥6 resolutions in both modes, comparing the consensus share of the first vs the second half of the subject's chronology; |Δ| ≥ 40 percentage points is flagged as a breakdown (consensus→votes) or rapprochement (votes→consensus). Small series (n≈6–9) should be read as candidates for investigation, not findings.
• Withdrawn — some withdrawals are procedural (texts merged or resubmitted), not defeats; the panel counts them without attributing motive.
• Amendments & the Scope filter — 731 catalogued items are amendments to draft texts; the 462 that went to a recorded vote make up 27% of all roll-calls (almost all post-2010; every withdrawn or not-considered item is an amendment). A Yes on a hostile amendment frequently opposes the parent resolution's aim — e.g. votes supporting amendments to women's-rights texts are typically votes to weaken them — so mixing amendments with resolutions inverts meaning (final-passage-only datasets are known to bias inference: Fjelstul, Hug & Kilby 2025). The global Scope control (top right) therefore defaults to resolutions only; amendments can be analysed separately or together, are tagged AMD in tables, and detection uses two agreeing catalogue signals (the title marker ": amendment /" and the 993 draft note).

Short method notes under individual charts link to the literature they draw on. One scope warning first, then the caveats sorted by how far they travel.
• Most of the quantitative literature studies the UN General Assembly — a different body. The GA has universal membership (~193 states) voting in plenary; the Commission (53 members elected by ECOSOC) and the Council (47, elected by the GA under regional quotas and membership criteria) are small elected chambers where membership itself is selected, absence is marginal (1.7% of roll-call cells in this dataset), and the agenda is set differently. GA findings are therefore reported below as hypotheses for this body, not established facts — they are never used here to interpret CHR/HRC statistics directly.

Properties of the measures — hold for any voting body by construction:
• Chance agreement — with lopsided majorities, any two states "agree" often by chance; raw coincidence and cohesion overstate alignment. Häge 2011, Political Analysis 19(3).
• Consensus inclusion — similarity/cohesion measures change substantially depending on whether consensus adoptions count as unanimous pseudo-votes (demonstrated on UNGA data; the point is general). Häge & Hug 2016, International Interactions 42(3).
• Agreement Index — the cohesion measure used here; developed for the European Parliament. Hix, Noury & Roland 2005, BJPS 35(2).
• Selection into roll-calls — visible directly in this dataset: 73% of catalogued texts passed without any vote, so roll-call statistics describe contested politics only (the same problem is documented for the UNGA: Fjelstul, Hug & Kilby 2025, RIO).

Evidence from this body (CHR / HRC):
• Hug & Lukács 2014, RIO 9 — Bayesian analysis of HRC roll-calls 2006–2010: state preferences, not regional-bloc membership, dominate voting; contested texts are disproportionately tabled by states with poor rights records.
• Lebovic & Voeten 2006, ISQ 50(4) — CHR country-condemnation politics 1977–2001: after the Cold War, targeting tracked actual violations more than partisan ties.

Findings from the UN General Assembly — related body; transfer with caution:
• Boockmann & Dreher 2011, Public Choice — on ~13,000 GA voting decisions 1980–2002, a state's own human-rights record is irrelevant to its vote once regional dependence is controlled; simple vote-aggregation rules can mislead. Whether this holds in an elected 47-member Council is an open question.
• Bailey, Strezhnev & Voeten 2017, JCR 61(2) — agenda composition confounds voting trends; dynamic ideal points separate position change from agenda change.
• Morse & Coggins 2024, RIO 19 — strategic absence: weak GA states abstain by staying away. Marginal here (1.7% absence) but worth remembering for the CHR era.
• Panke 2014, International Politics 51(6) — capacity, not politics, drives much GA absenteeism.
• Brazys & Panke 2017, IPSR 38(1) — why states switch positions between iterations of recurring GA texts.

The Language tab reads each resolution's operative verbs as a signal of how strongly the organ commits. Every operative paragraph opens, by drafting convention, with a controlled present-tense verb; we match it against the UNITAR glossary Terms Used in Resolutions (JG-1/02/10), cross-checked term-by-term against a second, publicly downloadable UNITAR source — UNITAR NYO, Guidelines for United Nations Resolutions (2020), Annex VII "Resolutions: frequently used terms" — over the clause-segmented texts of the Texts tab — . Those texts begin in 1993, so this tab covers a shorter period than the rest of the dashboard; amendments (edit instructions, not commitments) and the articles of annexed declarations and protocols (treaty-style text, not resolutions of the organ) are excluded from scoring — unlike the other seven tabs, the global Scope filter has nothing to switch between here, so it is disabled while this tab is open rather than implying a control it cannot exercise.
• Two axes, not one — the vocabulary mixes two illocutionary types, so a single strength scale is not defensible. Axis 1, directive force ranks verbs that direct an addressee to act: D5 decisional (Decides, Demands, Establishes) > D4 strong exhortation (Urges, Calls upon) > D3 tasking (Requests, Recommends) > D2 invitation (Invites, Encourages). Axis 2, sentiment ranks expressive verbs from condemnation (−2 Condemns) through neutral registering (0 Notes, Reaffirms) to commendation (+2 Commends). A third flag, delegation, marks clauses that create or task machinery: a directive clause that establishes, renews or extends a mechanism, or that names a UN actor and gives it something to do. A passing reference to a rapporteur or a report is not delegation — scoring mentions put 40% of all operative clauses on this axis and imported a spurious trend, since OHCHR and the special-procedures system only expanded after 1993.
• Grounding — the framework is the "legalization" triad of obligation, precision and delegation (Abbott, Keohane, Moravcsik, Slaughter & Snidal 2000, International Organization 54(3); Abbott & Snidal 2000); the directive-vs-sentiment split adapts a content-analysis approach developed for UN Security Council resolutions (Benson & Tucker 2022, JCR 66(3) — a different, treaty-binding body; borrowed here only as a coding method); the graded naming-and-shaming reading extends Lebovic & Voeten 2006. The tiers and their order are scholarly-grounded; the exact rank within a tier and the numeric spacing are an author-defined coding — a practitioner gloss notes there are "no official UN guidelines" fixing precise meanings.
• Why "Notes" / "Takes note" / "Expresses" / "Recognizes" are context-scored — these four verbs carry no fixed valence; the score is resolved per clause from the object ("expresses grave concern" → −1, "expresses appreciation" → +1, bare "notes" → neutral). This is not an ad hoc reading: a 2001 UN Office of Legal Affairs opinion, endorsed by the General Assembly in decision 55/488 and reproduced in the same UNITAR handbook (Annex VI), holds that the bare terms "takes note of" and "notes" are neutral, constituting neither approval nor disapproval — while UNITAR's own glossary separately catalogues the valenced compound forms ("notes with appreciation", "notes with concern") as distinct entries. The per-clause split is shown inline on the vocabulary ladder below wherever it applies.
• Critical caveat — this measures rhetorical / directive hardness, not legal bindingness. Even "Decides" binds only under Security-Council Chapter VII; CHR and HRC resolutions are formally non-binding whatever the verb. Tagging only the leading verb also omits the addressee and the precision of the ask — read the score as a first-order signal, not a legal determination.
• New terms — the corpus uses operative verbs absent from both UNITAR sources, above all Encourages (~3,100 clauses), plus low-volume Applauds, Thanks, Highlights, Observes — a small measure of drafting drift toward hortatory language. (Four terms once flagged here — Appreciates, Underscores, Calls for, and the unused Discourages — turned out to be attested in Annex VII of the second UNITAR source and were moved into the canonical count.)

This dashboard, the harvest and parsing pipeline, and the derived CSVs are a piece of research software with a versioned Zenodo DOI — cite it if you use it, quote a figure from it, or build on the underlying dataset.
• The footer's "Cite" button (bottom right, on every tab) copies the same line — useful mid-read, without navigating here. Both read from one place in the source (CITE_ACADEMIC in index.html) so they cannot drift apart.
• Machine-readable metadata — CITATION.cff in the repository carries the same record in Citation File Format; GitHub's own "Cite this repository" button on the repo page generates APA or BibTeX from it directly. The Zenodo record offers the same in a dozen further formats (BibTeX, RIS, DataCite XML, …).
• What the DOI covers — the concept DOI above always resolves to the latest release; cite a specific version (data snapshot + dashboard code as they stood on a given date) via the versioned DOI on the Zenodo record if reproducibility of an exact figure matters.
• Citing the underlying votes themselves, rather than this software: the source is the OHCHR Search Library's Voting collection — every export from this dashboard (CSV and PNG) stamps that citation plus an access date automatically, so a chart pulled from here always carries its own provenance without needing this section.
• Built and maintained by Łukasz Szoszkiewicz (Adam Mickiewicz University, Poznań) — ORCID 0000-0001-6671-2893. Source and issues on GitHub.

Nothing here is locked inside the interface. Pick whichever route matches what you are doing.
• One chart — every panel on this site carries a CSV button for the exact numbers behind it and a PNG button for the figure. Each export is stamped with the source citation, the active Scope and the access date, so a chart pulled from here carries its own provenance.
• The full tables — data/csv/ in the repository: resolutions.csv (one row per catalogued text) and votes_long.csv (one row per country per resolution), exactly as the pipeline emits them.
• A citable archive — the Zenodo record freezes code and data together at a version, for when an exact figure has to stay reproducible.
• Analysis-ready — the Hugging Face dataset packages the same records as parquet in five configs: The package ships the derived columns this Methodology argues for, so the same reasoning survives leaving the dashboard: prevailing_side and adopted (Yes vs No, never max(Y,N,A) — see §08), rollcall_reconciles (the ~9% in §06), is_amendment (§09), clause_type separating annexed instruments from operative text (§11), and subject_is_country_situation (§05). The clauses config also carries the experimental verb coding, with the same caveat: it measures rhetorical hardness, not legal bindingness.
• Licence — everything above is PolyForm Noncommercial 1.0.0: research, education, non-profit and personal use are free; commercial use needs a separate licence. The underlying voting records are public UN documents.