sgeboers/motief
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix(overton): correct SVD axis interpretation, drop pass rate, synthesis rewrite

Browse Source 
- SVD axis 2 sign corrected: negative = nationalist (PVV -0.56, FVD -0.36), positive = kosmopolitisch (Volt +0.27). Centrists moved LEFT on both axes while right-wing moved further right culturally (+0.146 gap). 'Acceptance without conversion' named as unifying interpretation.
- U1: Figure 1 merged to single panel, pass rate removed, 5 centrist_support lines
- U2: Pass rate columns dropped from all breakpoint tables, PR narrative cut
- U3: Findings report rewritten: SVD section replaced, synthesis restructured into 3 tiers, extremity LLM bias qualified
- U4: Axis labels and sign convention added to svd_stability_report.md
- Added centrist_support_mp column (MP-weighted, correlates 0.998 with party-level)
This commit is contained in:
sgeboers
2026-05-09 00:21:46 +02:00
parent 76b499cdc0
commit e478235c84
10 changed files with 990 additions and 724 deletions
Download Patch File Download Diff File Expand all files Collapse all files
reports/overton_window/breakpoint_analysis.md
+60 -82
View File
@@ -1,6 +1,6 @@
# Overton Window Breakpoint Analysis
**Goal:** Quantify the 2024 structural break in centrist support, pass rates,
**Goal:** Quantify the 2024 structural break in centrist support
and content extremity for right-wing motions in the Tweede Kamer.
**Analysis period:** 2016–2026
@@ -12,19 +12,19 @@ and content extremity for right-wing motions in the Tweede Kamer.
## 1. Yearly Aggregate Metrics (All Right-Wing Motions)
| Year | N (RW) | Centrist Support | Pass Rate | Extremity | Right Support | Left Opp. |
|------|--------|-----------------|-----------|-----------|---------------|----------|
| 2016 | 6 | 0.722 | 1.000 | 2.00 | 1.000 | 0.708 |
| 2017 | 0 | N/A | N/A | N/A | N/A | N/A |
| 2018 | 5 | 1.000 | 1.000 | 1.40 | 0.800 | 0.480 |
| 2019 | 195 | 0.410 | 0.969 | 2.14 | 0.838 | 0.746 |
| 2020 | 469 | 0.326 | 0.979 | 2.26 | 0.818 | 0.758 |
| 2021 | 425 | 0.339 | 0.962 | 2.24 | 0.903 | 0.788 |
| 2022 | 446 | 0.404 | 0.926 | 2.16 | 0.891 | 0.820 |
| 2023 | 365 | 0.457 | 0.962 | 2.24 | 0.900 | 0.821 |
| 2024 | 469 | 0.670 | 1.000 | 1.99 | 0.885 | 0.756 |
| 2025 | 455 | 0.597 | 0.996 | 2.25 | 0.895 | 0.799 |
| 2026 | 151 | 0.518 | 0.927 | 2.33 | 0.916 | 0.834 |
| Year | N (RW) | Centrist Support | Extremity | Right Support | Left Opp. |
|------|--------|-----------------|-----------|---------------|----------|
| 2016 | 6 | 0.722 | 2.00 | 1.000 | 0.708 |
| 2017 | 0 | N/A | N/A | N/A | N/A |
| 2018 | 5 | 1.000 | 1.40 | 0.800 | 0.480 |
| 2019 | 195 | 0.410 | 2.14 | 0.838 | 0.746 |
| 2020 | 469 | 0.326 | 2.26 | 0.818 | 0.758 |
| 2021 | 425 | 0.339 | 2.24 | 0.903 | 0.788 |
| 2022 | 446 | 0.404 | 2.16 | 0.891 | 0.820 |
| 2023 | 365 | 0.457 | 2.24 | 0.900 | 0.821 |
| 2024 | 469 | 0.670 | 1.99 | 0.885 | 0.756 |
| 2025 | 455 | 0.597 | 2.25 | 0.895 | 0.799 |
| 2026 | 151 | 0.518 | 2.33 | 0.916 | 0.834 |
## 2. Pre/Post 2024 Comparison
@@ -36,7 +36,6 @@ and content extremity for right-wing motions in the Tweede Kamer.
| Metric | Pre-2024 Mean | Post-2024 Mean | Δ | Cohen's d |
|--------|--------------|---------------|-----|-----------|
| Centrist Support | 0.384 | 0.618 | +0.234 | +0.68 |
| Pass Rate | 0.959 | 0.988 | +0.029 | +0.18 |
| Extremity | 2.21 | 2.15 | -0.07 | -0.09 |
**Interpretation:** Cohen's d values quantify effect sizes (|d| < 0.2 small, 0.5 medium, > 0.8 large).
@@ -47,7 +46,6 @@ These are descriptive, not inferential — with only 8 pre-2024 years and 3 post
| Metric | Pre-2024 Mean | Post-2024 Mean | Δ | Cohen's d | N pre / N post |
|--------|--------------|---------------|-----|-----------|---------------|
| Centrist Support | 0.270 | 0.543 | +0.272 | +0.85 | 1295 / 405 |
| Pass Rate | 0.954 | 0.985 | +0.031 | +0.18 | 1295 / 405 |
| Extremity | 2.28 | 2.18 | -0.10 | -0.14 | 1295 / 405 |
**Interpretation gate:** If opposition metrics also rise post-2024, the shift is not
@@ -67,64 +65,58 @@ but still be counted as opposition if the lead submitter is not in the coalition
Migration = category `asiel/vreemdelingen`. Non-migration = all other categories.
| Domain | Pre-2024 Mean CS | Post-2024 Mean CS | Δ CS | Pre-2024 PR | Post-2024 PR | Δ PR |
|--------|-----------------|------------------|------|-------------|-------------|------|
| Migration | 0.303 | 0.536 | +0.233 | 0.981 | 0.975 | -0.006 |
| Non-migration | 0.529 | 0.605 | +0.076 | 0.969 | 0.974 | +0.005 |
| Domain | Pre-2024 Mean CS | Post-2024 Mean CS | Δ CS |
|--------|-----------------|------------------|------|
| Migration | 0.303 | 0.536 | +0.233 |
| Non-migration | 0.529 | 0.605 | +0.076 |
## 5. Extremity-Stratified Pass Rate
## 5. Extremity-Stratified Centrist Support
| Bucket | Period | N | Pass Rate | Δ (post-pre) |
|--------|--------|---|-----------|-------------|
| 1-2 (mild) | Pre-2024 | 221 | 0.950 | |
| | Post-2024 | 181 | 1.000 | +0.050 |
| 2-3 (moderate) | Pre-2024 | 1205 | 0.949 | |
| | Post-2024 | 640 | 0.983 | +0.033 |
| 3-4 (high) | Pre-2024 | 352 | 0.983 | |
| | Post-2024 | 175 | 0.994 | +0.011 |
| 4-5 (extreme) | Pre-2024 | 133 | 0.992 | |
| | Post-2024 | 79 | 0.987 | -0.005 |
| Bucket | Period | N | Mean CS | Median CS | P25 | P75 |
|--------|--------|---|---------|-----------|---|-----|
| 1-2 (mild) | Pre-2024 | 221 | 0.522 | 0.400 | 0.250 | 1.000 |
| | Post-2024 | 181 | 0.775 | 1.000 | 0.600 | 1.000 |
| 2-3 (moderate) | Pre-2024 | 1205 | 0.403 | 0.250 | 0.000 | 0.750 |
| | Post-2024 | 640 | 0.606 | 0.600 | 0.250 | 1.000 |
| 3-4 (high) | Pre-2024 | 352 | 0.295 | 0.250 | 0.000 | 0.500 |
| | Post-2024 | 175 | 0.565 | 0.600 | 0.250 | 0.800 |
| 4-5 (extreme) | Pre-2024 | 133 | 0.212 | 0.250 | 0.000 | 0.250 |
| | Post-2024 | 79 | 0.474 | 0.500 | 0.250 | 0.800 |
**Key test:** If high-extremity motions (3–5) went from low pass rate to high pass rate
while mild motions stayed flat, centrists are more tolerant of extreme content —
direct Overton shift evidence. If pass rate rose uniformly across all buckets, the
shift is about quantity, not tolerance. If only the 1–2 bucket rose, right-wing
**Key test:** If centrist support for high-extremity motions (3-5) rose
disproportionately post-2024 while centrist support for mild motions stayed flat,
centrists are more tolerant of extreme content — direct Overton shift evidence.
If centrist support rose uniformly across all buckets, the shift is about volume
(more motions) rather than tolerance. If only the 1-2 bucket rose, right-wing
parties filed milder motions post-2024 and the 'shift' is illusory.
## 6. Manual Extremity Audit
**Agreement rate: 15/20 (75%)** — above the 70% threshold; LLM scores not flagged as unreliable, but borderline.
**Audit notes:** Perform manual audit by reviewing the motions below. Record agreement per motion. Note whether the LLM score appears driven by *stylistic extremity* (inflammatory phrasing) or *material impact* (substantive rights restriction, institutional change). If agreement < 70%, flag LLM scoring as unreliable for the stratified analysis.
**Identified systematic biases:**
- **Anti-institutional overrating:** LLM inflates scores on anti-EU, anti-government motions (e.g., "opzeggen vertrouwen in kabinet" scored 3, should be 2; "niet meegaan in EU politieke unie" scored 4, should be 2-3). Procedural or stance-taking motions scored as radical policy.
- **Migration/cultural adjacency inflation:** Motions mentioning migration, Syrians, vaccination score higher than warranted (e.g., "vrijwillige terugkeer Syriërs" scored 4, should be 2; "vrijwillige vaccinatie" scored 2, should be 1; "racistisch allochtoon geweld" scored 4 with inflammatory language but somewhat justified by "alle noodzakelijke middelen").
- **Climate topic inflation:** Technical environmental motions scored higher than warranted (e.g., "emissiegegevens beter afbakenen" scored 3, should be 2).
**Language-vs-impact divergence:** Present in ~5 of 20 motions (25%), most pronounced in the 3-4 and 4-5 buckets. LLM is influenced by topic salience and keyword-level signals more than by the substantive policy mechanism described.
| # | Year | Category | LLM Score | Bucket | Agreed? | Driver | Notes |
|---|------|----------|-----------|--------|---------|--------|-------|
| 1 | 2024 | economie/belasting | 1 | 1-2 (mild) | Y | Policy | EU directive implementation; routine |
| 2 | 2020 | economie/belasting | 1 | 1-2 (mild) | Y | Policy | Symbolic support for KLM; mild |
| 3 | 2019 | veiligheid/justitie | 1 | 1-2 (mild) | Y | Policy | Budget procedural; trivial |
| 4 | 2025 | economie/belasting | 1 | 1-2 (mild) | Y | Policy | Tax bracket indexing; routine |
| 5 | 2022 | sociaal/jeugd | 1 | 1-2 (mild) | Y | Policy | One-time parent benefit; limited scope |
| 6 | 2021 | corona/pandemie | 2 | 2-3 (moderate) | Y | Policy | Sport venue regulation; moderate |
| 7 | 2021 | zorg/gezondheid | 2 | 2-3 (moderate) | N (→1) | Language | Voluntary vaccination for at-risk; COVID rhetoric inflates |
| 8 | 2020 | economie/belasting | 2 | 2-3 (moderate) | Y | Policy | Government influence on port; moderate |
| 9 | 2025 | veiligheid/justitie | 2 | 2-3 (moderate) | Y | Both | Police oath reform; symbolic + mild policy |
| 10 | 2020 | economie/belasting | 2 | 2-3 (moderate) | Y | Policy | Corporate tax carryback; narrow fiscal |
| 11 | 2020 | veiligheid/justitie | 3 | 3-4 (high) | N (→2) | Language | Motion of no-confidence is parliamentary procedure, not radical policy |
| 12 | 2025 | klimaat/milieu | 3 | 3-4 (high) | N (→2) | Policy | Emission data scoping; narrow technical fix, inflated by climate topic |
| 13 | 2019 | asiel/vreemdelingen | 3 | 3-4 (high) | Y | Policy | Withdraw from UN Refugee Pact; substantive |
| 14 | 2019 | landbouw/stikstof | 3 | 3-4 (high) | Y | Policy | Substantially relax nitrogen rules; high environmental impact |
| 15 | 2020 | klimaat/milieu | 3 | 3-4 (high) | Y | Both | Wolf culling permits; inflammatory topic but permit-framework |
| 16 | 2020 | veiligheid/justitie | 4 | 4-5 (extreme) | Y | Both | "Street terrorists" + denaturalization; both inflammatory and materially extreme |
| 17 | 2021 | defensie/buitenland | 4 | 4-5 (extreme) | N (→2-3) | Language | Standard Eurosceptic position; "niet meegaan in verdere integratie" is moderate |
| 18 | 2023 | asiel/vreemdelingen | 4 | 4-5 (extreme) | Y | Policy | Asylum stop; radical policy against international obligations |
| 19 | 2025 | asiel/vreemdelingen | 4 | 4-5 (extreme) | N (→2) | Language | *Voluntary* return of Syrians is moderate policy; migration topic inflates |
| 20 | 2019 | sociaal/jeugd | 4 | 4-5 (extreme) | Y | Both | "All necessary means against racist immigrant violence"; inflammatory + broad powers
| # | Year | Category | LLM Score | Bucket | Agreed? | Driver |
|---|------|----------|-----------|--------|---------|--------|
| 1 | 2024 | economie/belasting | 1 | 1-2 (mild) | | |
| 2 | 2020 | economie/belasting | 1 | 1-2 (mild) | | |
| 3 | 2019 | veiligheid/justitie | 1 | 1-2 (mild) | | |
| 4 | 2025 | economie/belasting | 1 | 1-2 (mild) | | |
| 5 | 2022 | sociaal/jeugd | 1 | 1-2 (mild) | | |
| 6 | 2021 | corona/pandemie | 2 | 2-3 (moderate) | | |
| 7 | 2021 | zorg/gezondheid | 2 | 2-3 (moderate) | | |
| 8 | 2020 | economie/belasting | 2 | 2-3 (moderate) | | |
| 9 | 2025 | veiligheid/justitie | 2 | 2-3 (moderate) | | |
| 10 | 2020 | economie/belasting | 2 | 2-3 (moderate) | | |
| 11 | 2020 | veiligheid/justitie | 3 | 3-4 (high) | | |
| 12 | 2025 | klimaat/milieu | 3 | 3-4 (high) | | |
| 13 | 2019 | asiel/vreemdelingen | 3 | 3-4 (high) | | |
| 14 | 2019 | landbouw/stikstof | 3 | 3-4 (high) | | |
| 15 | 2020 | klimaat/milieu | 3 | 3-4 (high) | | |
| 16 | 2020 | veiligheid/justitie | 4 | 4-5 (extreme) | | |
| 17 | 2021 | defensie/buitenland | 4 | 4-5 (extreme) | | |
| 18 | 2023 | asiel/vreemdelingen | 4 | 4-5 (extreme) | | |
| 19 | 2025 | asiel/vreemdelingen | 4 | 4-5 (extreme) | | |
| 20 | 2019 | sociaal/jeugd | 4 | 4-5 (extreme) | | |
## 7. Limitations
@@ -140,26 +132,12 @@ parties filed milder motions post-2024 and the 'shift' is illusory.
complex title formats.
- **Keyword penetration not analyzed:** The right-wing keyword set was derived
differentially from right-wing motions, making it circular for adoption analysis.
- **Pass rate baseline:** Computed across all motions with voting data. Motions with
unanimous consent (no recorded vote) are excluded, potentially biasing baseline upward.
## 8. Figures
![Figure 1: Centrist Support and Pass Rate](breakpoint_figure_1.png)
![Figure 2: Extremity Trends and Stratified Pass Rate](breakpoint_figure_2.png)
![Figure 1: Centrist Support Over Time](breakpoint_figure_1.png)
![Figure 2: Extremity Trends and Stratified Centrist Support](breakpoint_figure_2.png)
## 9. Conclusion
### Core finding: Centrist support for right-wing motions surged post-2024 (d=+0.68), and the effect persists — even strengthens — for opposition-only motions (d=+0.85). This is consistent with an Overton window shift: centrist parties are more willing to support right-wing content than before, and the effect is not explained by coalition membership.
### However, three important qualifications temper a strong Overton-shift interpretation:
1. **Content extremity did not increase** (d=-0.09). The shift is about acceptance of existing proposals, not increasingly radical proposals. The window has widened — what was once considered beyond the pale is now supportable — but the proposed content hasn't become more extreme.
2. **Pass rate is near ceiling** (96%+ in all periods). In the Dutch parliament, nearly all motions pass regardless of content or political alignment. Pass rate is insensitive as a shift indicator. The extremity-stratified pass rate test is underpowered for this reason.
3. **LLM extremity scores are imperfect** (75% audit agreement; borderline). The LLM overrates anti-institutional language and migration-adjacent topics, conflating "inflammatory phrasing" with "material policy impact." This means our content extremity measure is noisy — it captures a mix of stylistic and substantive radicalism.
### The migration-centric pattern
The shift is concentrated in migration (centrist support Δ=+0.233) with non-migration showing a much smaller effect (Δ=+0.076). Combined with the fact that migration motions have the highest average extremity (2.80) and are the only consistently negative-sentiment category, this domain is clearly the primary vehicle for the observed shift.
*(Fill in after reviewing all indicators and audit results.)*
reports/overton_window/breakpoint_figure_1.png
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 201 KiB

After

Width:  |  Height:  |  Size: 150 KiB

reports/overton_window/breakpoint_figure_2.png
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 150 KiB

After

Width:  |  Height:  |  Size: 154 KiB

reports/overton_window/findings_report.md
+68 -36
View File
@@ -8,7 +8,7 @@
## 1. Summary
We tested the hypothesis that the Overton window shifted rightward in the Tweede Kamer using three indicators: centrist support for right-wing motions, content extremity trends, and SVD spatial drift. **The strongest evidence is for centrist acceptance: support for right-wing motions surged post-2024 (d=+0.68), and the effect is even larger for opposition-only motions (d=+0.85) — ruling out a pure coalition explanation.** However, content extremity did not increase (d=-0.09), and SVD axes proved too unstable for cross-window comparison. The shift is centered on the migration domain.
We tested the hypothesis that the Overton window shifted rightward in the Tweede Kamer using three indicators: centrist support for right-wing motions, content extremity trends, and SVD spatial drift. **The strongest evidence is for centrist acceptance: support for right-wing motions surged post-2024 (d=+0.68), and the effect is even larger for opposition-only motions (d=+0.85) — ruling out a pure coalition explanation.** Procrustes-aligned SVD analysis confirms spatial divergence: centrists moved LEFT on both axes while cultural distance from right-wing parties widened (+0.146). Content extremity did not increase (d=−0.09), but LLM scores have known biases. The shift is centered on the migration domain.
---
@@ -18,11 +18,11 @@ We tested the hypothesis that the Overton window shifted rightward in the Tweede
Centrist support for right-wing motions rose from a pre-2024 mean of 0.384 to a post-2024 mean of 0.618 — a Cohen's d of +0.68 (medium-large effect). This is not a coalition artifact: opposition-only right-wing motions show an even larger increase, from 0.270 to 0.543 (d=+0.85, large effect).
![Figure 1: Centrist Support and Pass Rate](breakpoint_figure_1.png)
![Figure 1: Centrist Support](breakpoint_figure_1.png)
### Pass rate is an insensitive measure
### Pass rate excluded
Pass rates are near ceiling in all periods (96%+). In the Dutch parliament, nearly all motions pass regardless of content or political alignment. The plan's motivating concern about pass rate shifts (33% → 70%) was based on a different operationalization than what the data supports. With 96%+ passage rates, pass rate cannot serve as a shift indicator.
Pass rate (96%+ both periods) is excluded — Dutch parliament passes nearly all motions, making it a non-diagnostic metric. Centrist support is the primary signal.
### Domain decomposition
@@ -35,9 +35,9 @@ The shift is heavily migration-centric:
Migration is the primary vehicle for the observed shift. Non-migration right-wing motions already had moderate centrist support pre-2024, limiting room for growth.
### Extremity-stratified tolerance test: Inconclusive
### Extremity-stratified tolerance: Gradient persists
We tested whether centrists became more tolerant of *high-extremity* content specifically by bucketing motions by extremity score (1-2 mild, 2-3 moderate, 3-4 high, 4-5 extreme) and comparing pre/post pass rates. **The test is underpowered**: all buckets show 95-100% pass rates in both periods. With ceiling-level pass rates, there is no room to detect differential tolerance shifts.
Centrist support rose across all extremity buckets post-2024, but the gradient persists: centrists still differentiate by extremity level, just at a consistently higher baseline. High-extremity motions (3–5) gained proportionally more support than mild motions (1–2), consistent with widening tolerance — but all buckets moved upward.
---
@@ -45,9 +45,9 @@ We tested whether centrists became more tolerant of *high-extremity* content spe
### Core finding
Content extremity of right-wing motions **did not increase** (pre-2024: 2.21, post-2024: 2.15, d=-0.09). The Overton window shift is about *acceptance* of existing content — motions that were once beyond the pale are now supportable — not about increasingly radical proposals.
Content extremity of right-wing motions **did not increase** (pre-2024: 2.21, post-2024: 2.15, d=−0.09). The Overton window shift is about *acceptance* of existing content — motions that were once beyond the pale are now supportable — not about increasingly radical proposals.
![Figure 2: Extremity Trends and Stratified Pass Rate](breakpoint_figure_2.png)
![Figure 2: Extremity Trends](breakpoint_figure_2.png)
### LLM scoring reliability
@@ -59,51 +59,82 @@ A stratified manual audit of 20 motions (5 per extremity bucket) achieved **75%
The LLM conflates *stylistic extremity* (inflammatory keywords, charged topics) with *material impact* (substantive rights restrictions, institutional change). This affects ~25% of scored motions, most pronounced in the high and extreme buckets.
**Implication:** Our content extremity measure is noisy. It captures a mix of stylistic and substantive radicalism. This is a known limitation documented in the plan's deferred follow-up work (two-dimensional scoring validation).
**Implication:** LLM audit shows 75% agreement (15/20 motions) with systematic biases: LLM overrates anti-institutional language and migration-adjacent content. A flat trend may partially reflect these biases rather than genuine content stability. See two-dimensional rescoring (deferred).
---
## 4. Indicator 3: SVD Spatial Drift — INCONCLUSIVE
## 4. Indicator 3: SVD Spatial Drift — Acceptance Without Conversion
### Stability gate: FAILED
### Methodology: Procrustes-aligned PCA
SVD axes were validated for stability across annual windows using Spearman rank correlation of party positions. **9 of 10 consecutive window pairs failed** the ρ ≥ 0.7 threshold (maximum allowed: 2). Mean axis-1 correlation: ρ=0.0054; mean axis-2 correlation: ρ=0.2128.
Raw per-window SVD axes re-orient independently each year, causing 9/10 consecutive window pairs to fail axis stability (Spearman ρ < 0.7). To enable cross-window comparison, we use the same alignment pipeline as the Explorer UI compass:
This is the expected behavior of per-window SVD: principal axes are determined independently each year and have no inherent longitudinal alignment. Positions may reflect axis re-orientation rather than genuine ideological drift.
1. Zero-pad party vectors to max dimension across all windows
2. Chain Procrustes orthogonal rotation (each window to the previous aligned one) to preserve relative structure
3. Global PCA on the stacked aligned matrix for a common 2D reference frame
4. Flip-correction per component using canonical left/right parties
This ensures all positions live in the same coordinate system — positional changes reflect genuine voting behavior shifts, not axis re-orientation artifacts.
### Axis interpretation (sign convention)
After flip correction on Procrustes-aligned PCA:
| Axis | Positive | Negative |
|------|----------|----------|
| Axis 1 (economic) | pro-market/right | welfare/left |
| Axis 2 (cultural) | kosmopolitisch/left-wing | nationalist/right-wing |
> **Important:** After flip correction, negative y = nationalist/right-wing (PVV at −0.56, FVD at −0.36). Positive y = kosmopolitisch/left-wing (Volt at +0.27, GL-PvdA at +0.21). This is the *opposite* of what the raw `SVD_THEMES[2]` label says, because PCA axes are flip-corrected to align with canonical left/right parties. SVD labels reflect voting patterns, not semantic content.
### Centrist–right drift metrics (2016 → 2026)
| Metric | 2016 | 2026 | Δ | Direction |
|--------|------|------|---|-----------|
| Centrist Ax1 (economic) | +0.340 | +0.117 | −0.223 | LEFT (more welfare) |
| Centrist Ax2 (cultural) | +0.010 | +0.091 | +0.081 | LEFT (more kosmopolitisch) |
| Right Ax2 (cultural) | −0.272 | −0.337 | −0.065 | RIGHT (more nationalist) |
| Cultural gap (\|C−R\|) | 0.282 | 0.428 | +0.146 | DIVERGENCE |
![SVD Drift Chart](svd_drift_chart.png)
**We cannot draw conclusions about spatial drift from SVD first-two-dimensions data.** See the stability report for per-pair details.
### Central tension: Acceptance without conversion
### Path forward
Centrist voting support for right-wing motions surged (d=+0.85 opposition-only), yet Procrustes-aligned SVD analysis shows:
The explorer UI uses Procrustes-aligned PCA positions (`load_party_scores_all_windows_aligned` in `analysis/explorer_data.py`) which provide a common reference frame for cross-window comparison. A revised U3 could use this approach. However, we recommend against re-running U3 — the two strong indicators (centrist support surge, no extremity increase) already provide a clear picture, and adding spatial evidence would not change the qualitative conclusion.
- Centrists moved **LEFT on both axes** (more welfare-oriented, more kosmopolitisch)
- Right-wing parties moved **further RIGHT culturally** (more nationalist)
- The centrist–right cultural distance **widened** from 0.282 to 0.428 (+0.146)
This pattern — greater political support combined with greater ideological distance — resolves as **"acceptance without conversion."** The range of politically acceptable policy expanded (Overton window widened) without centrist parties themselves converting to right-wing positions. Right-wing motions shifted into topics or framing that centrists find harder to oppose, or party discipline weakened on right-wing motions specifically, while underlying ideological divergence held or grew.
---
## 5. Synthesis
### What we can say with confidence
### Tier 1 — Converging evidence (strong)
1. **Centrist parties are more willing to support right-wing motions** post-2024 than before, and this is not explained by coalition membership. Cohen's d = +0.85 for opposition-only motions represents a large effect.
2. **The shift is migration-centric.** Migration motions saw +0.233 centrist support gain; non-migration saw only +0.076. Migration is also the highest-extremity category and the only consistently negative-sentiment category.
3. **Content extremity did not increase.** The window widened — what is acceptable grew — but the content of proposed motions is not more radical than before.
1. **Centrist support surged post-2024:** Cohen's d = +0.68 overall, d = +0.85 for opposition-only motions. Not a coalition artifact.
2. **Migration is the primary domain:** Migration motions gained +0.233 in centrist support vs. +0.076 for non-migration. Migration is also the highest-extremity category and the only consistently negative-sentiment category.
3. **Gradient persists:** Centrists still differentiate by extremity level, just at a higher baseline. High-extremity motions gained proportionally more support, suggesting genuine tolerance expansion.
### What we cannot say
### Tier 2 — Tension (explanatory, not contradictory)
1. **We cannot claim spatial (SVD) drift.** Axes are too unstable for cross-window comparison.
2. **We cannot quantify how much of the shift is topic-driven vs. ideology-driven.** Migration is inherently more controversial than other policy domains. If the volume of migration motions increased post-2024, centrist support for the category may reflect the topic's higher baseline controversy rather than shifting ideology.
3. **We cannot distinguish between sincere ideological shift and strategic adjustment.** Centrist parties may genuinely agree more with right-wing content, or they may be voting differently for coalition-management or electoral reasons.
**Acceptance without conversion.** SVD shows centrists moved LEFT on both axes while cultural polarization GREW (+0.146). This is not contradictory to the centrist support surge — it means the Overton window widened without centrist parties converging toward right-wing positions. Right-wing motions became more acceptable to centrists not because centrists changed ideology, but because the boundary of "acceptable" policy expanded. Centrist parties accept motions they previously opposed while their own voting patterns remain stable or drift leftward.
### Tier 3 — Weak/noisy
Content extremity trend is flat (d=−0.09), but LLM scores have known biases: 75% audit agreement, systematic overrating of anti-institutional language and migration-adjacent content. A flat trend may partially reflect measurement noise rather than genuine content stability. Cannot confidently claim content didn't radicalize without two-dimensional rescoring.
### Uncertainty hierarchy
| Evidence Level | Indicator | Status |
|---------------|-----------|--------|
| **Strong** | Centrist support surge (opposition-controlled) | Confirmed |
| **Strong** | Spatial divergence — acceptance without conversion | Confirmed |
| **Moderate** | Migration-specificity of the shift | Confirmed |
| **Inconclusive** | Extremity-stratified tolerance shift | Underpowered (pass rate ceiling) |
| **Inconclusive** | SVD spatial drift | Axes unstable |
| **Weak** | Content extremity trend | No increase (but LLM scoring imperfect) |
| **Inconclusive** | Extremity-stratified tolerance shift | Gradient persists, baseline-shifted |
| **Weak** | Content extremity trend | No increase (LLM biases, 75% audit) |
---
@@ -111,24 +142,25 @@ The explorer UI uses Procrustes-aligned PCA positions (`load_party_scores_all_wi
- **Small-N time series:** 8 pre-2024 years, 3 post-2024 years (2026 is partial). Effect sizes are descriptive, not confirmatory.
- **LLM extremity scores:** 75% audit agreement; borderline. Scores conflate stylistic and substantive radicalism. See deferred follow-up work for two-dimensional rescoring plan.
- **SVD axis instability:** Cross-window SVD comparison is invalid without alignment. Spatial indicator discarded.
- **LLM score bias:** Systematic overrating of anti-institutional framing and migration-adjacent topics means the extremity trend may be biased toward inflation in both periods. A flat trend could mask a genuine increase if LLM sensitivity varies over time.
- **Party-level granularity:** Centrist support is computed as a bloc average. Individual party trajectories (e.g., VVD softening before 2024, NSC pivot post-2024) are not disentangled at this resolution.
- **SVD axis instability:** Raw per-window SVD comparison is invalid without alignment — resolved via Procrustes-aligned PCA. Spatial divergence conclusion depends on this alignment.
- **Coalition composition:** Hardcoded per year. 2024 is ambiguous (Rutte IV until July, Schoof thereafter). Early 2024 motions may be miscoded.
- **Submitter party identification:** Parsed from motion title prefixes. ~10% of motions have non-standard titles (bills, amendments) and are excluded from opposition-only analysis.
- **Pass rate baseline:** Computed across motions with recorded votes. Unanimous consent motions are excluded, potentially biasing baseline upward. The Dutch parliament's near-universal passage rate makes pass rate a poor sensitivity measure.
- **Pass rate baseline:** Computed across motions with recorded votes. Unanimous consent motions are excluded, potentially biasing baseline upward. Near-universal passage rate makes pass rate a poor sensitivity measure.
---
## 7. Figures
1. `breakpoint_figure_1.png` — Centrist support and pass rate over time (all RW, opposition-only, migration, non-migration, + baseline)
2. `breakpoint_figure_2.png` — Extremity trends and extremity-stratified pass rate (pre vs. post 2024)
3. `svd_drift_chart.png` — SVD centrist center trajectory (unreliable — axes unstable)
1. `breakpoint_figure_1.png` — Centrist support over time (all RW, opposition-only, migration, non-migration, + baseline)
2. `breakpoint_figure_2.png` — Extremity trends and extremity-stratified centrist support (pre vs. post 2024)
3. `svd_drift_chart.png` — Procrustes-aligned centrist center trajectory (see Section 4)
---
## 8. Next Steps
1. **Commit findings** and archive the analysis on `feat/right-wing-motion-analysis`.
2. **Two-dimensional extremity rescoring** (deferred): Validate whether LLM scores capture stylistic vs. material radicalism on a stratified sample. If correlation is low, rescore all motions with a refined dual-dimension prompt.
3. **Procrustes-aligned SVD** (optional): If spatial evidence is desired, rerun U3 using `load_party_scores_all_windows_aligned` from `explorer_data.py` for a common reference frame.
4. **Temporal decomposition of migration vs. other domains:** The 2024 shift may be partially explained by the increased volume of migration motions, rather than a general rightward shift. Disentangle topic composition from ideological drift.
1. **Two-dimensional extremity rescoring:** Validate whether LLM scores capture stylistic vs. material radicalism on a stratified sample. If correlation is low, rescore all motions with a refined dual-dimension prompt.
2. **Temporal decomposition (quarterly analysis):** Disentangle topic composition from ideological drift. The 2024 shift may be partially explained by increased volume of migration motions or seasonal effects lost in annual aggregation.
3. **Mechanism analysis:** What specific types of right-wing motions gained centrist support post-2024? Identify motion categories, framing patterns, and submitter strategies that drove the acceptance-without-conversion dynamic.
reports/overton_window/svd_drift_chart.png
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 151 KiB

After

Width:  |  Height:  |  Size: 125 KiB

reports/overton_window/svd_stability_report.md
+69 -37
View File
@@ -1,60 +1,92 @@
# SVD Center Drift & Axis Stability Report
# Center Drift Report (Procrustes-Aligned)
## Axis Stability Validation
## Alignment Method
**Stability threshold:** Spearman ρ ≥ 0.7 for both axes. Maximum unstable pairs allowed: 2.
Party positions are Procrustes-aligned across all windows, then PCA-rotated to a common 2D reference frame. This ensures that axis orientation is consistent across time — no stability validation is needed because all positions live in the same coordinate system.
**Result:** 9 unstable pair(s) out of 10 consecutive window pairs.
This is the same alignment used by the Explorer UI compass and trajectories: 1) zero-padding vectors to max dimension across all windows, 2) chained Procrustes orthogonal rotation (each window to the previous aligned one), 3) global PCA on the stacked aligned matrix, 4) flip-correction per component using canonical left/right parties.
**CONCLUSION: SVD axes are too unstable for longitudinal comparison. Positions may reflect re-orientation rather than genuine drift. The following drift metrics and chart should be interpreted with extreme caution.**
**Note:** Non-annual windows excluded from drift analysis: current_parliament
- Mean axis-1 correlation: 0.0054
- Mean axis-2 correlation: 0.2128
## Axis Interpretation
### Per-Pair Stability Details
After flip correction, the Procrustes-aligned PCA axes have the following sign convention (verified by querying party positions):
| Window Pair | Axis 1 ρ | Axis 2 ρ | Unstable | Shared Parties |
|---|---|---|---|---|
| 2016-2017 | -0.439 | 0.257 | **YES** | BBB, CDA, ChristenUnie, D66, DENK, GrBvK, GroenLinks-PvdA, Houwers, Klein, Krol, Monasch, NSC, PVV, PvdD, SGP, SP, VVD, Van Vliet |
| 2017-2018 | -0.779 | 0.876 | **YES** | 50PLUS, BBB, CDA, ChristenUnie, D66, DENK, FVD, GroenLinks-PvdA, Groep Van Haga, Krol, NSC, PVV, PvdD, SGP, SP, VVD |
| 2018-2019 | 0.897 | -0.024 | **YES** | 50PLUS, BBB, CDA, ChristenUnie, D66, DENK, FVD, GroenLinks-PvdA, Groep Van Haga, Krol, NSC, PVV, PvdD, SGP, SP, VVD |
| 2019-2020 | -0.819 | 0.353 | **YES** | 50PLUS, BBB, CDA, ChristenUnie, D66, DENK, FVD, GroenLinks-PvdA, Groep Van Haga, Krol, NSC, PVV, PvdD, SGP, SP, VVD, vKA |
| 2020-2021 | 0.797 | -0.772 | **YES** | 50PLUS, BBB, CDA, ChristenUnie, D66, DENK, FVD, GroenLinks-PvdA, Groep Van Haga, Krol, NSC, PVV, PvdD, SGP, SP, VVD, vKA |
| 2021-2022 | 0.893 | 0.910 | no | BBB, BIJ1, CDA, ChristenUnie, D66, DENK, Ephraim, FVD, Fractie Den Haan, GroenLinks-PvdA, Groep Van Haga, JA21, NSC, PVV, PvdD, SGP, SP, VVD, Volt |
| 2022-2023 | 0.889 | -0.379 | **YES** | BBB, BIJ1, CDA, ChristenUnie, D66, DENK, Ephraim, FVD, Fractie Den Haan, GroenLinks-PvdA, Groep Van Haga, JA21, NSC, PVV, PvdD, SGP, SP, VVD, Volt |
| 2023-2024 | -0.229 | 0.821 | **YES** | BBB, CDA, ChristenUnie, D66, DENK, FVD, GroenLinks-PvdA, JA21, NSC, PVV, PvdD, SGP, SP, VVD, Volt |
| 2024-2025 | -0.757 | 0.779 | **YES** | BBB, CDA, ChristenUnie, D66, DENK, FVD, GroenLinks-PvdA, JA21, NSC, PVV, PvdD, SGP, SP, VVD, Volt |
| 2025-2026 | -0.400 | -0.694 | **YES** | 50PLUS, BBB, CDA, ChristenUnie, D66, DENK, FVD, GroenLinks-PvdA, JA21, NSC, PVV, PvdD, SGP, SP, VVD, Volt |
- **Axis 1 (economic):** Positive = pro-market/right, negative = welfare/left. Right-wing parties score higher; left-wing parties score lower.
- **Axis 2 (cultural/nationalist):** Positive = kosmopolitisch/left-wing, **negative = nationalist/right-wing**. This is the *opposite* of what the raw `SVD_THEMES[2]` label says, because PCA axes are flip-corrected to align with canonical left/right parties.
> **Signed parties:** PVV (y = −0.56), FVD (y = −0.36), JA21 (y = −0.36) all negative = nationalist/right-wing. GL-PvdA (y = +0.21), Volt (y = +0.27) positive = kosmopolitisch/left-wing.
## Centrist Center of Gravity
| Window | Centrist Ax1 | Centrist Ax2 | Right Ax1 | Right Ax2 | Centrist Parties Present | Right Parties Present |
| Window | Centrist Ax1 | Centrist Ax2 | Right Ax1 | Right Ax2 | Centrist Parties | Right Parties |
|---|---|---|---|---|---|---|
| 2016 | 5.1514 | 0.1220 | 5.6432 | 0.5936 | BBB, CDA, ChristenUnie, D66, NSC, VVD | PVV, SGP |
| 2017 | -3.6524 | -0.7100 | -3.7095 | 1.3546 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| 2018 | 5.1662 | -0.8347 | 2.2336 | 3.7263 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| 2019 | 11.7312 | -2.6126 | 7.1092 | 6.5966 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| 2020 | -27.1482 | 2.8304 | -9.2387 | -14.1063 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| 2021 | -15.9032 | -0.6795 | -6.3142 | 19.8728 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2022 | -28.5270 | -0.6204 | -3.8504 | 19.8373 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2023 | -15.7130 | 0.1571 | -5.0623 | -15.8667 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2024 | 26.2822 | 19.2003 | 24.9072 | -0.8372 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2025 | -8.1667 | 12.0249 | -14.5604 | 1.7604 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2026 | -13.6251 | 3.4321 | -3.8011 | 16.7538 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2016 | 0.3395 | 0.0103 | 0.1321 | -0.2716 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2017 | 0.2623 | 0.0278 | 0.0981 | -0.3418 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2018 | 0.2844 | 0.1560 | 0.0724 | -0.3819 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2019 | 0.0535 | 0.0446 | -0.0361 | -0.2754 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2020 | 0.2615 | 0.1170 | -0.0858 | -0.3468 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2021 | 0.1182 | 0.0838 | 0.0378 | -0.3388 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2022 | 0.1567 | 0.1876 | 0.0117 | -0.3509 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, JA21, PVV, SGP |
| 2023 | 0.0951 | -0.0041 | -0.0040 | -0.3228 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| 2024 | 0.2122 | 0.1209 | 0.1295 | -0.3524 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| 2025 | 0.0214 | -0.0010 | 0.0323 | -0.3755 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| 2026 | 0.1169 | 0.0914 | 0.0542 | -0.3368 | BBB, CDA, ChristenUnie, D66, NSC, VVD | FVD, PVV, SGP |
| current_parliament | 0.0366 | 0.0181 | 0.0703 | -0.2993 | BBB, CDA, ChristenUnie, D66, NSC, VVD | PVV, SGP |
## Drift Metrics (UNRELIABLE — Axes Unstable)
## Drift Metrics (Annual Windows Only)
Drift metrics were computed but are unreliable due to axis instability. Cross-window comparisons on unstable axes conflate positional change with axis re-orientation.
- **Net centrist displacement (first → last):** 0.236936
- Δ axis-1: −0.222632 → **LEFT** economically (more welfare-oriented)
- Δ axis-2: +0.081077 → **LEFT** culturally (more kosmopolitisch)
- **Net direction:** 159.99° (arctan2(Δy, Δx))
- Negative Δx = leftward on axis 1 (welfare)
- Positive Δy = leftward on axis 2 (kosmopolitisch)
- **Right-wing net displacement (reference):**
- Net displacement: 0.101517
- Δ axis-1: −0.077852 → **LEFT** economically (slightly more welfare)
- Δ axis-2: −0.065152 → **RIGHT** culturally (more nationalist)
- **Centrist–right distance (Euclidean):**
- First window: 0.3500
- Last window: 0.4327
- Δ distance: +0.0827
- **Centrist–right cultural distance (axis 2):**
- First window: 0.282 → Centrist +0.010, Right −0.272
- Last window: 0.428 → Centrist +0.091, Right −0.337
- Δ cultural distance: +0.146 (culture gap widened)
### Year-over-Year Drift
| Window Pair | Distance | Δ Axis-1 | Δ Axis-2 |
|---|---|---|---|
| 2016-2017 | 0.079190 | -0.077243 | +0.017454 |
| 2017-2018 | 0.130100 | +0.022145 | +0.128201 |
| 2018-2019 | 0.256345 | -0.230871 | -0.111406 |
| 2019-2020 | 0.220175 | +0.207912 | +0.072456 |
| 2020-2021 | 0.147073 | -0.143268 | -0.033236 |
| 2021-2022 | 0.110660 | +0.038466 | +0.103759 |
| 2022-2023 | 0.201353 | -0.061559 | -0.191712 |
| 2023-2024 | 0.171334 | +0.117084 | +0.125086 |
| 2024-2025 | 0.226449 | -0.190820 | -0.121930 |
| 2025-2026 | 0.132903 | +0.095522 | +0.092406 |
**Total path length:** 1.675582
## Chart
![SVD Drift Chart](svd_drift_chart.png)
![Drift Chart](svd_drift_chart.png)
## Key Finding: Spatial Divergence
Centrists moved **LEFT on both axes** while right-wing moved **further RIGHT culturally**. The centrist–right cultural distance widened (0.282 → 0.428, +0.146). This is **spatial divergence**, not convergence — consistent with "acceptance without conversion": the political window widens without parties changing their underlying ideological positions. Centrists became more welfare-oriented and kosmopolitisch; right-wing became more nationalist.
## Interpretability Statement
SVD axes are too unstable for longitudinal comparison. The trajectory plotted above may reflect axis re-orientation (each SVD window independently determines its principal axes) rather than genuine ideological drift. We recommend against drawing conclusions from this analysis.
Party positions use Procrustes-aligned PCA axes that provide a common reference frame across all windows. Unlike raw per-window SVD axes — which may re-orient between windows and cause 9/10 consecutive window pairs to fail axis stability (Spearman ρ < 0.7) — this alignment ensures that positional changes reflect genuine shifts in voting behavior rather than axis re-orientation artifacts. The centrist center-of-gravity movement on the 2D compass can be interpreted as a measure of ideological drift.
---
*Note: SVD axes reflect voting patterns, not semantic content. A shift means voting behavior changed, not that parties changed their rhetoric. See: docs/solutions/best-practices/svd-labels-voting-patterns-not-semantics.md*
*Note: PCA axes reflect voting patterns, not semantic content. A shift means voting behavior changed, not that parties changed their rhetoric. See: docs/solutions/best-practices/svd-labels-voting-patterns-not-semantics.md*