Public opinion — read-only context

The "Public opinion" section on the dashboard displays survey data as supplementary context alongside the pillar index. The key rule: these values do not feed into the score function or any automated decision about event severity. They serve a human reader for contextual comparison: how the public perceives democracy vs. where the index has actual institutional signals.

Why read-only and NOT an input to the score

When designing this section, three integration levels were considered:

Level	What	Why not / why yes
A. Read-only display ✅	Polls displayed on the dashboard but don’t affect any number.	Implemented — we see correlation, no causal claim, no feedback loop.
B. Quarterly validation	Polls as an external benchmark alongside V-Dem/EIU.	Possible in a future iteration. Clear hypothesis-test semantics.
C. Direct input into the score ❌	Polls as a sub-component of the civil pillar.	Rejected. Three reasons:

Reasons for rejecting direct integration (C):

1. Causality direction: polls measure perception, not institutional reality. Low trust in courts doesn’t necessarily mean courts function worse — it can be the consequence of media coverage, polarisation, or a single high-profile case. The index is meant to reflect institutional shifts, not the media mood.

2. Feedback loop: if polls counted into the score, bad publicity → low score → even worse coverage → even lower score. The index would turn into a mood ring instead of a democracy gauge.

3. Double-count: the structural baseline from V-Dem and EIU already contains polls indirectly (V-Dem synthesises expert input with the help of surveys, EIU has a "Political Culture" subscale based on WVS/Eurobarometer). Adding polls a second time would mean counting the same signal twice.

Sources and their profile

Source	Frequency	Type	Use
CVVM (Institute of Sociology, Czech Academy of Sciences)	monthly	academic, transparent methodology	the main time-series chart — trust in constitutional institutions
STEM	irregular	commercial with a public archive, century-long tradition (since 1990)	topical findings cards — ad-hoc surveys
Median	irregular	commercial	topical findings cards — ad-hoc surveys
Eurobarometer (EC)	2× per year	EU-standardised, comparable across the 27 members	deferred: not in yet, see "Known gaps"

Deliberately omitted:

• Voting intent / party preference — the democracy index isn’t an election forecast; political preferences are orthogonal to institutional health.
• Popularity of specific politicians — politicised, short-sighted, says nothing about the institutional state.
• Single-issue topical surveys (e.g. "do you support the pension reform?") — too topical.

The section sticks to trust in institutions + perceived corruption + satisfaction with democracy across all sources, on the same scale.

CVVM as the primary source

CVVM is the gold standard for Czech opinion research for these reasons:

• Academic affiliation (Institute of Sociology of the Czech Academy of Sciences) — no commercial client, no political loyalty
• Transparent methodology — sample size, fieldwork, weights are published
• Long-term series — monthly measurement of "trust in constitutional institutions" has been running since the 1990s
• Standardised question — wording stable across decades, comparability preserved
• Publicly available — press releases + microdata for free, citation requirement

Institutions measured: president, government, Chamber of Deputies, Senate, regional assemblies, regional governors, municipal councils, mayors.

Important note (2025-11): in November/December 2025, CVVM changed its data-collection methodology (transition CAPI → online panel). The dashboard curve is visibly broken by a red reference line — datapoints before and after this boundary are not directly comparable. For trend analysis, evaluate each era separately.

STEM and Median as supplementary context

Commercial pollsters were included deliberately, despite initial worries about bias. Reasons for inclusion:

• Topical agility — commercial pollsters react to events much faster than academic ones (CVVM has a 1–3 month lag between fieldwork and publication; STEM/Median sometimes publish within 2 weeks of an event)
• Cross-source transparency — displaying polls side by side is methodologically more transparent than cherry-picking one source. If STEM and Median publish different numbers for the same question, that’s a feature (= the user sees that polls aren’t unanimous), not a bug.
• Read-only safety — commercial-pollster bias matters mainly when the data drives a serious decision (election forecast, policy). For supplementary dashboard context with no impact on the score, the risk is minimal.

Commercial pollsters appear in the dashboard as "Recent findings" — cards with a headline finding + a link to the original report, not as data points in a chart. Reason: their publication is irregular (no stable time series), the output format varies (percentages, percent changes, qualitative categories).

Deliberately omitted commercial pollsters:

• Kantar CZ, NMS Market Research, Ipsos CZ — they don’t have a publicly available archive of running surveys on trust in institutions; their data is published primarily through media (secondary citation), unsuitable for direct ingest.

Known gaps (TODO future iterations)

Eurobarometer (EC)

Eurobarometer publishes Standard EB twice a year (spring + autumn) with a "Trust in national institutions" + "Perceived corruption" section for all member states. The Czech fact sheet (~20-page PDF) has all the relevant values.

Why it isn’t in the dashboard yet:

• The europa.eu/eurobarometer site is an SPA with JS-rendered content — WebFetch (LLM-friendly content extraction) doesn’t reach it
• The EU open data portal data.europa.eu has dataset entries, but the search is also JS-driven
• Manual ingest from PDF country fact sheets is possible (~10 minutes per survey) but requires a regular human-in-the-loop

Planned workflow:

1. When a new Standard EB is published (May + November) manually download the PDF Czechia country fact sheet
2. Extract "Trust in [parliament, government, courts, police, EU]" + "Corruption perception"
3. Add to data/public_opinion/eurobarometer.json with the same shape as the CVVM file

It can be implemented, just waiting for the bandwidth for manual ingest.

GLOBSEC Trends

Annual report (May) focused on V4 (CZ, SK, PL, HU). Measures democratic attitudes, perceptions of threat, attitudes towards the EU/NATO. Available as PDF + press release.

Why it’s missing: the annual cadence + ad-hoc questions mean a time series cannot be built from GLOBSEC alone — it would always be 3–4 datapoints. More suitable as a qualitative topical card; deferred for now.

Microdata for re-analysis

CVVM publishes raw microdata with a median delay of ~1 year after fieldwork. A full re-analysis with custom aggregation (different cohorts, different weighting) is possible, but well beyond the dashboard scope. Deferred.

Update workflow

CVVM publishes a new report monthly (typically 2–3 weeks after fieldwork ends). Workflow for adding a new data point:

1. Track CVVM category "Institutions and politicians"
2. Find the new press release "Trust in constitutional institutions – [period]"
3. Open it, extract trust % for each institution
4. Add a new object to data/public_opinion/cvvm-trust.json in the data array (sorted ascending by period)
5. Commit + push — Vercel redeploys, the dashboard shows the new point

For STEM/Median (topical findings):

1. On a significant publication, add an object to data/public_opinion/topical.json in the items array
2. Recommended retention: the most recent ~6–10 findings; move older ones to archive or delete
3. Commit + push

No cron, no automation — manual ingest with human review ensures quality and prevents false positives when source structure changes.