CID: V-Dem Democracy Report — India coverage, longitudinal series

Document: V-Dem Democracy Report (annual), India as featured case — scored as the full global report Organization: V-Dem Institute, University of Gothenburg (Varieties of Democracy) CID ID: CID-0046 (provisional; longitudinal series) Document type: TYPE 4, Composite Index Rubric version: v0.3.2 Scored: 2026-05-29 Series span: Democracy Report 2017 – 2026, 10 annual editions (4 in Era A, 6 in Era B) Scoring boundary: Companion apparatus credited (codebook v14/v15 + public V-Dem dataset + measurement model), per pre-registration decision. A conservative “PDF-stands-alone” reading is given in the sensitivity section.

---

Verdict

The V-Dem Democracy Report scores Adequate, at the very top of the band (7.87–7.94), and is the highest-scoring institutional report in the CID corpus to date — behind only Pew’s 2021 survey of religion in India (8.3, Research-Grade), the corpus’s lone Research-Grade study and a one-off primary survey rather than a recurring index. It sits more than two full points above the Amnesty India entries (4.60–4.87, Deficient), above the estimated USCIRF India chapters (~5.2–5.4), and above the Freedom House global report (6.08, borderline Adequate) — the institution it most resembles.

The reason is structural and it cuts the opposite way from every prior series the CID has scored. Amnesty, USCIRF, and the Freedom House country chapters were all held in or near the Deficient band by the same gap: no codebook, no sampling frame, no inter-coder reliability, no in-document verification pathway. V-Dem closes all four. It publishes a 200-plus-indicator codebook with operational definitions and ordinal scale anchors; it codes the entire universe of states (202 countries, 1789–present) on a fixed framework; it uses roughly five independent expert coders per indicator-country-year run through a Bayesian item-response measurement model that estimates coder reliability and emits confidence intervals; and it releases the full disaggregated dataset — 31 million data points — as Tier-1 public, replicable data. The annual report reports those confidence intervals inline and names the dataset and codebook versions it draws on.

The same finding the CID documented for Amnesty — structural invariance across a long institutional series — holds here too. The grade barely moves across ten editions because the measurement apparatus does not change. The difference is altitude. For Amnesty the invariance was the absence of method; for V-Dem the invariance is the presence of method. The score is stable at the top of Adequate, the D6 Research-Grade gate is open (D6 = 8 ≥ 7), and the series lands just 0.06 short of the 8.0 Research-Grade threshold.

Why this matters for India. V-Dem is the original source of the most widely-cited claim about Indian democracy: that India became an “electoral autocracy” in 2018–19, announced in Democracy Report 2021 (“the world’s largest democracy turned into an electoral autocracy”). That designation is repeated by journalists, scholars, and advocacy groups worldwide, frequently without attribution to its origin. The CID’s job is to grade how defensibly that claim is produced. The answer: it rests on the most rigorous, most transparent, most replicable measurement apparatus of any document in this corpus. A reader who distrusts the “electoral autocracy” label can do something they cannot do with an Amnesty entry — download the underlying indicator scores, read the coder confidence intervals, and check the classification threshold themselves.

---

Type classification rationale

The V-Dem Democracy Report aggregates scored indicators into country-level indices (the Liberal Democracy Index, Electoral Democracy Index, and the four-category Regimes of the World classification: closed autocracy, electoral autocracy, electoral democracy, liberal democracy). That is the defining marker of TYPE 4 (Composite Index) — the same classification assigned to the Freedom House Freedom in the World CompleteBooks (CID-0040–0044). All eight dimensions apply in Full, and the Conditional Module (Index Construction) activates at 10%, reducing the eight base-dimension weights proportionally to 90% of their standard values.

This is the correct and consistent comparator. Both V-Dem and Freedom House produce a global, universalist, indicator-based regime classification published annually. Scoring V-Dem against the same type and weights as Freedom House is what makes the two-point gap between them meaningful rather than an artifact of differing rubric paths.

Scope note — “India” vs. the global report. Like the Freedom House CompleteBook, the unit scored is the full annual report, not an India-only excerpt. India is the report’s marquee case across the series (the largest country to be downgraded, and by the population-weighted metric the single largest driver of the measured global democratic decline). Scoring the whole report is the honest unit: V-Dem’s India claim is inseparable from the global apparatus that produces it. The India relevance is the application; the methodology score is a property of the apparatus.

---

The two structural eras

The series carries one genuine methodological inflection — the adoption of the peer-review-vetted Episodes of Regime Transformation (ERT) method for identifying democratizers and autocratizers, which replaced the cruder “difference between the score ten years ago and now” approach used in the earlier reports. The 2024 report describes the older method plainly and explains why ERT is “more scientific” and gives “more precision and reliability.” This inflection, plus a steady densification of the report’s external citation apparatus, defines two eras. As with Amnesty, the inflection moves only one dimension (here D8, counter-evidence) and only by one point; every other dimension is invariant across the full series.

Era	Years	Editions	Change-detection method	Report-level citations (URLs / domains)	D8
A — Pre-ERT	2017 – 2020	4	10-year difference	8–33 URLs / 6–22 domains, moderate concentration	6
B — ERT era	2021 – 2026	6	ERT (peer-review vetted)	110–279 URLs / 42–99 domains, low concentration (diverse)	7

The India downgrade to electoral autocracy is announced at the Era A → B boundary (Democracy Report 2021), and it is announced with a transparent statement that the prior year’s classification had been uncertain and that more and better data now placed India in the electoral-autocracy category “with a higher degree of certainty.” That is a textbook example of the verification behavior that lifts this series above the rest of the corpus.

---

Dimension scores

Scores are invariant across the series except D8. Each is grounded in the per-year analysis and the source report text, not in general reputation. The companion apparatus (codebook, public dataset, measurement model) is credited per the pre-registration decision; the conservative alternative is in the sensitivity section.

D1: Definitional precision — 8 / 10

Effective weight: 10.8%. Invariant across series.

V-Dem’s codebook (v14/v15) provides operational definitions for every one of its 200-plus indicators, each on an ordinal scale (typically 0–4) with written anchors for each scale point and coding guidance for hard cases. The four Regimes of the World categories are defined by explicit, published thresholds on the v2x_regime measure, with a documented uncertainty variant (v2x_regime_amb) for “grey zone” cases near a threshold. The report names these definitions and version numbers inline. This is materially more granular than the Freedom House 25-question framework (scored 6), which embeds definitions in question guidance without per-scale-point anchors. Held below 9–10 because the annual report PDF itself reprints the definitions only partially — the full operational apparatus lives in the codebook a reader must fetch separately (an in-report definitions section appears in only 3 of 10 editions). Score of 8 reflects a genuinely codebook-grade definitional system, credited as referenced, docked one notch for not being self-contained in the report.

D2: Classification rigor — 8 / 10

Effective weight: 16.2% (tied heaviest). Invariant across series.

This is the dimension on which V-Dem decisively separates from Freedom House (which scored 5 here, explicitly because it publishes “zero inter-coder reliability data” and uses “no blind coding”). Inter-coder reliability is not an afterthought for V-Dem — it is the center of the method. Roughly five independent country experts code each indicator, and their ratings are run through a Bayesian item-response-theory measurement model that estimates each coder’s reliability and systematic bias, reconciles disagreement, and outputs a point estimate with a confidence interval. Anchoring vignettes and bridge/lateral coding link coders to a common scale. The report carries the markers throughout: “expert coders born in 185 different countries,” 4,200 contributing scholars, and confidence intervals on the headline figures. Held below 9–10 because the report PDF does not reprint per-indicator reliability statistics (ICR is absent from the in-report text in all 10 editions); they reside in the methodology papers and dataset. Score of 8 reflects a measurement model built around inter-coder reliability — the single biggest upgrade over every prior document in the corpus.

D3: Case capture and sampling — 8 / 10

Effective weight: 13.5%. Invariant across series.

Universal coverage: 202 countries from 1789 to the present, every one coded on the identical fixed framework, enabling genuine longitudinal and cross-national comparability. This is broader than Freedom House (195 countries, modern era; scored 7) on both the country and the time axis. The indicator set is transparent and stable year to year. Held below 9 because the within-country evidence-gathering of individual experts is not documented at the per-claim level, and the deep historical coverage necessarily rests on thinner source material than the contemporary years. Score of 8 reflects universal, fixed-framework coverage with documented indicator selection.

D4: Coverage symmetry — 8 / 10

Effective weight: 13.5%. Invariant across series.

The framework is structurally neutral and universalist: it scores electoral process, freedom of expression, civil society, rule of law, and constraints on the executive for every state, with no identity-specific criteria. It passes the Swap Test — removing identity markers does not change how the indicators function. India’s prominence in the report is a product of the population-weighted metric (a large country moves the global average more than a small one), not of any India-specific or identity-specific framing, and the report says so explicitly. Identity directionality is minimal across the series (the only edition with any identity-target co-occurrence is 2021: Muslim-as-target 4, all others ≤1). Held below 9 because narrative emphasis naturally tracks decline over improvement, and the editorial choice to foreground population-weighted figures structurally elevates large countries like India — a choice V-Dem makes defensibly but which is still a choice (see D8). Score of 8 reflects strong structural neutrality and accurate, non-overstated scope claims.

D5: Source independence — 8 / 10

Effective weight: 9.0%. Invariant across series.

V-Dem is an academic project housed at the University of Gothenburg, funded through research councils and foundations rather than by the government of any country it assesses — a cleaner independence position than Freedom House (scored 7, docked for structural dependence on US-government funding). Its evidence base is its own network of independent country experts, not self-citation to prior advocacy. The report’s external citation profile is diverse and broadens sharply over the series: from 6 unique domains and a moderate Herfindahl concentration of 0.22 in 2017 to 99 unique domains and a low (diverse) concentration of 0.04 by 2026, mixing academic, media, and advocacy sources. Held below 9 because the report draws on secondary journalism (BBC, Reuters) and advocacy reporting (HRW) for narrative color around the index findings, and because research-council/foundation funding, while disclosed and independent of assessed subjects, is still external funding. Score of 8 reflects academic independence and an increasingly diverse evidence base.

D6: Verification standards — 8 / 10

Effective weight: 16.2% (tied heaviest). Invariant across series. THE difference-maker.

This is the dimension that gates Research-Grade, and it is where crediting the companion apparatus matters most. V-Dem publishes the full disaggregated dataset — every indicator, every country, every year, with point estimates and confidence intervals — as Tier-1 public, machine-readable, replicable data (31 million data points, downloaded over 360,000 times). The measurement model is documented and reproducible. The report reports confidence intervals inline and names the exact dataset and codebook versions and variables it uses (e.g., v2x_regime, v2x_regime_amb, v14/v15). The 2021 India passage demonstrates the verification standard in action: it states that the prior year’s India classification was “highly uncertain” and that re-estimation with more and better data raised the certainty — a documented, data-driven self-correction. This is verification down to the indicator level, far beyond Freedom House (scored 5, with Tier-1 aggregate scores but Tier-3 underlying worksheets). Held below 9 because the individual expert coders’ raw responses are confidentialized to protect coders (a reader gets the modeled estimates and intervals, not the individual coder identities), and per-claim sourcing in the report’s narrative prose remains uneven. Score of 8.

D6 ≥ 7: Research-Grade gate is OPEN. Unlike every prior document in the corpus, the verification floor does not block Research-Grade for V-Dem. The series falls short of the 8.0 total by arithmetic, not by the gate — see Weighted totals.

D7: Transparency and governance — 8 / 10

Effective weight: 4.5%. Invariant across series.

V-Dem is governed as an academic institution with a steering committee and scientific oversight, names its 4,200 contributing scholars, publishes its data, codebook, and graphing tools openly, and discloses funding in every edition (funding disclosure is present in all 10 editions; a conflict-of-interest statement appears by 2025). This edges out Freedom House (scored 7). Held below 9 because per-edition itemization of funders inside the report PDF is partial. Score of 8.

D8: Counter-evidence — 6 / 7 by era

Effective weight: 6.3%. Era-dependent — the only moving score.

V-Dem’s defining counter-evidence practice is structural: every estimate it publishes carries its own confidence interval, so uncertainty is engaged at the level of each number rather than relegated to a disclaimer. Beyond that:

• Era A (2017–2020): 6. Uncertainty reporting is present and limitations/counter-evidence sections appear in some editions (2018, 2019), but the change-detection method is the cruder 10-year difference, and the report’s most pointed methodological self-examination is not yet present. This still matches or beats Freedom House’s 6.
• Era B (2021–2026): 7. The peer-review-vetted ERT method is adopted and explained against its predecessor. Critically, the 2024 report contains a striking self-critique of its own headline finding: it warns that the population-weighted metric makes it look as though “global democracy has declined in a dramatic fashion largely because India is backsliding,” states plainly that this framing “may give misleading impressions,” and shows the arithmetic both with and without India (the equal-weighted decline is 0.03; population-weighted 0.13; population-weighted excluding India only 0.05). A report engaging this directly with a counter-reading of its own most-quoted number is rare in the genre. Conflict-of-interest disclosure also appears by 2025.

Held below 8 in both eras because systematic engagement with the substantial external academic literature critiquing expert-coding and index aggregation is not carried inside the report, and not every edition includes a formal limitations section.

Conditional Module: Index Construction — 8 / 10

Effective weight: 10.0%. Invariant across series.

The aggregation method is fully published and peer-reviewed: indicators → Bayesian measurement model → indices (LDI, EDI, and components) → Regimes-of-the-World thresholds. Crucially, V-Dem supplies the two things the Freedom House index lacked (which held the Freedom House module to 6): confidence intervals on the index scores, and an explicit uncertainty/sensitivity treatment in the form of the v2x_regime_amb “grey zone” variant that flags countries whose intervals straddle a classification threshold. Held below 9–10 because a formal sensitivity analysis of how alternative indicator weightings would change the headline regime classifications is not foregrounded in the report itself. Score of 8.

---

Weighted totals

Dimension	Score (A / B)	Effective weight	Weighted (A)	Weighted (B)
D1	8	10.8%	0.86	0.86
D2	8	16.2%	1.30	1.30
D3	8	13.5%	1.08	1.08
D4	8	13.5%	1.08	1.08
D5	8	9.0%	0.72	0.72
D6	8	16.2%	1.30	1.30
D7	8	4.5%	0.36	0.36
D8	6 / 7	6.3%	0.38	0.44
CM (Index Construction)	8	10.0%	0.80	0.80
Total			7.87	7.94

• Era A (2017–2020): 7.87 — Adequate (top of band)
• Era B (2021–2026): 7.94 — Adequate (top of band)

Non-compensatory checks

• D3 sampling cap: D3 = 8, well above the threshold of 3. Cap does not apply.
• D6 Research-Grade gate: D6 = 8 ≥ 7. Gate is open — Research-Grade is not blocked. The series falls 0.06 short of the 8.0 Research-Grade total on the weighted arithmetic alone. A single one-point lift on any heavily-weighted dimension (D2 or D6 to 9) would tip Era B into Research-Grade, which would make V-Dem the only document in the corpus to cross that line. The conservative call here keeps every credited dimension at 8, holding the series at the top of Adequate.

---

Sensitivity analysis

Scheme	Era A	Era B	Grade
Standard (TYPE 4 + CM)	7.87	7.94	Adequate
Equal weights (9 × 11.1%)	7.78	7.89	Adequate
Verification-heavy (D6 @ 25%)	7.89	7.94	Adequate
PDF-stands-alone (companion not credited)	6.22	6.28	Adequate (mid-band)

Grade stability: STABLE. All three credited-apparatus schemes place both eras firmly at the top of Adequate, just under the Research-Grade line. The standard scheme already weights D6 at 16.2%, so the verification-heavy scheme barely moves it.

The boundary matters, but not the band. The fourth row models the conservative reading the pre-registration decision set aside: scoring only what the annual PDF reprints inline, refusing to credit the referenced codebook, dataset, and measurement model. Under that reading D1 falls to 6, D2 to 5, D6 to 5, and the Index module to 6 — and the series lands at 6.2–6.3, almost exactly on top of the Freedom House CompleteBook (6.08). This is the single most consequential scoring decision in the analysis: crediting V-Dem’s companion apparatus is worth roughly 1.6 points and is the difference between “best in the genre by a wide margin” and “tied with Freedom House.” The decision to credit it is defensible and consistent with how the Freedom House companion methodology document was credited — and V-Dem’s apparatus is incomparably more rigorous than that document. But the document a casual reader actually holds is the PDF, and the conservative reading is recorded here in full so the choice is transparent. Anang owns the final pre-registration boundary assignment.

---

Calibration context

The V-Dem Democracy Report (7.87–7.94) sits at the top of the institutional-report corpus, second only to Pew’s one-off 2021 survey (8.3, Research-Grade), the corpus’s single Research-Grade study. The ordering is methodologically coherent:

Document	Grade	Score	What it has
Pew — Religion in India (2021 survey)	Research-Grade	8.3	Probability sampling, fully documented methodology, published limitations (a one-off TYPE 1 survey)
V-Dem Democracy Report	Adequate (top)	7.87–7.94	Codebook + public disaggregated dataset + Bayesian measurement model with ICR + confidence intervals
Freedom House FitW (global)	Adequate (borderline)	6.08	25-question framework + Tier-1 aggregate scores; no ICR, no CIs
Freedom House India chapter	Deficient	5.93	Framework, but excerpted; D2/D3 thin
USCIRF India (est.)	Deficient	~5.2–5.4	Designation framework
Amnesty International — India entry	Deficient	4.60–4.87	No codebook, no sampling frame, no in-document verification
SASAC	Advocacy-grade	~4.3	—
CSOH / India Hate Lab (floor)	Unreliable	~2.1	—

V-Dem beats Freedom House — the institution it most resembles — by closing the exact two weaknesses that held Freedom House to the bottom of Adequate: inter-coder reliability (D2: 8 vs 5) and underlying-data verification (D6: 8 vs 5). It beats Amnesty by more than three points because Amnesty has none of the apparatus V-Dem is built on.

As with every prior series, the grade is a property of the document as presented, not a referendum on the institution or on whether India “really is” an electoral autocracy. V-Dem’s Adequate-top grade says the apparatus behind the India designation is the most checkable in the corpus. It does not say the designation is correct — only that a skeptic has been given the tools to test it, which is precisely what Amnesty, USCIRF, and the Freedom House country chapters do not provide.

---

Structural invariance note

Across all ten editions (Democracy Report 2017–2026), the dimensions that measure methodology — D1, D2, D3, D4, D5, D6, D7, and the Index Construction module — are invariant. The core apparatus (codebook, universal expert coding, Bayesian measurement model with inter-coder reliability, public disaggregated dataset with confidence intervals, published aggregation method) does not change across the series. Only D8 moves, and only by one point, tracking the adoption of the peer-review-vetted ERT method and the appearance of explicit in-report methodological self-critique from 2021 onward.

Year-to-year variation in word count, quantitative-claim count (22 in 2017 → 143 in 2026), citation density (8 URLs / 6 domains in 2017 → 279 URLs / 99 domains in 2026), and the prominence of India (a spike to 2.23 mentions per 1,000 words in the 2021 downgrade edition) reflects changing global events, a densifying citation apparatus, and editorial emphasis — not a change in how the underlying knowledge is produced.

This mirrors the CID’s USCIRF, Freedom House, and Amnesty findings precisely in form, and inverts them in substance: score variation across a long institutional series reflects citation-infrastructure and formatting changes, not a change in method — and here the unchanging method is the most rigorous of any institutional report in the corpus.

The per-year data (word count, claims, citations, directionality, era assignment, assigned scores) is consolidated in the per-year analysis.