SCB Partisympatiundersökningen (PSU) Support-Voting Dataset

This document provides descriptive documentation, data provenance, and quality assurance diagnostics for the historical SCB PSU support-voting panel covering 2010M11 through 2026M05 (Step 1 of the support-voting research plan).

[!IMPORTANT] Conceptual Distinction Disclaimer Partisympati (bästa parti) and Röstningssympati / Val idag are fundamentally different SCB survey concepts:

• Partisympati measures general ideological and emotional sympathy or party identification among respondents.
• Röstningssympati / Val idag measures intended vote in a hypothetical parliamentary election held today, calibrated to expected turnout and reweighted by SCB’s demographic and register-based weighting model.

The aggregate difference between overall party sympathy and overall vote intention must not be interpreted directly as tactical voting.

1. Available PSU Waves

The dataset covers 29 distinct waves spanning from November 2010 to May 2026:

2010M11, 2011M05, 2011M11, 2012M05, 2012M11, 2013M05, 2013M11, 2014M05, 2014M11,
2015M05, 2015M11, 2016M05, 2016M11, 2017M05, 2017M11, 2018M05, 2018M11, 2019M05,
2019M11, 2020M05, 2020M11, 2021M05, 2021M11, 2022M05, 2022M11, 2023M05, 2024M05,
2025M05, 2026M05

Survey Frequency Timeline

• 2010M11 – 2022M11 (25 waves): PSU was fielded semiannually in May (M05) and November (M11).
• 2023M05 – 2026M05 (4 waves): Starting in 2023, SCB changed the survey schedule to an annual survey fielded exclusively in May (M05).

2. Table Coverage and Provenance

All raw datasets were retrieved directly from SCB Statistikdatabasen (SSD API). For every table, the API schema metadata, submitted query payload, and data response are archived under data/raw/scb_support_voting/ with SHA-256 integrity hashes recorded in manifest.json.

Both percentage point estimates (estimate_pct) and margins of error (margin_error_pp) are preserved separately across all tables.

3. Missing and Suppressed Cell Rates

In official SCB publications, cells with insufficient sample sizes are suppressed using the string .. to prevent unreliable estimates and protect respondent privacy.

In this pipeline:

• Suppressed cells (..) are parsed as NaN (float) and flagged with value_status = 'suppressed'.
• Missing cells are flagged with value_status = 'missing'.
• Observed cells are flagged with value_status = 'observed'.
• Under no circumstances are suppressed or missing values coerced to zero.

Summary of Cell Statuses across 29 Waves

Suppression in Table A occurs predominantly in cross-ideological, low-probability transfers (e.g. V sympathizers voting for KD or M sympathizers voting for V). Transfers between natural coalition partners (e.g. M $\to$ KD, M $\to$ L, S $\to$ MP, S $\to$ V) are largely observed.

4. Canonical Party Mapping and Category Handling

The pipeline maps raw SCB codes and labels into canonical party identifiers while preserving original raw codes and text in separate columns.

Canonical Classifications

• Parliamentary Parties (8): M, C, L, KD, MP, S, V, SD
  • Historical mapping: FP / Folkpartiet $\to$ L (Liberalerna).
• Historical / Other Named Parties:
  • NYD $\to$ NYD (historical party 1991–1994, retained as historical_party).
  • övr / övriga $\to$ OTHER (other_party).
• Non-Party Response Categories (kept distinct from political parties):
  • ingen sympati/vet ej $\to$ NO_SYMPATHY_OR_DONT_KNOW (no_sympathy)
  • hela väljarkåren $\to$ TOTAL_ELECTORATE (total_electorate)
  • blankt $\to$ BLANK_VOTE (blank_vote)
  • vet ej $\to$ DONT_KNOW (dont_know)
  • inget parti $\to$ NO_SECOND_CHOICE (no_second_choice)

5. Example Donor $\to$ Recipient Matrices

Below are illustrative empirical cross-tabulations for selected waves from the processed panel.

Latest Wave: 2026M05

Intended Vote Conditional on Best Party: $P(\text{vote } p \mid \text{best } j)$

| Donor \ Recipient | M | C | L | KD | MP | S | V | SD | | :— | :—: | :—: | :—: | :—: | :—: | :—: | :—: | :—: | | M | 79.1% | 1.1% | 1.0% | 1.5% | .. | 1.1% | .. | 3.5% | | C | 2.0% | 77.1% | 0.0% | .. | .. | 8.9% | .. | 0.0% | | L | 8.5% | 5.2% | 68.8% | .. | 0.0% | 5.5% | 0.0% | .. | | KD | .. | .. | .. | 79.5% | 0.0% | 0.0% | 0.0% | .. | | MP | 0.0% | .. | 0.0% | 0.0% | 81.4% | 6.7% | 3.3% | .. | | S | 1.0% | 0.5% | .. | .. | 1.1% | 85.2% | 0.6% | 0.7% | | V | 0.0% | .. | .. | 0.0% | 3.6% | 4.9% | 78.5% | .. | | SD | 0.9% | 0.0% | 1.2% | 1.0% | 0.0% | .. | 0.0% | 90.5% |

Second-Best Party Conditional on Best Party: $P(\text{second } p \mid \text{best } j)$

| Donor \ Recipient | M | C | L | KD | MP | S | V | SD | | :— | :—: | :—: | :—: | :—: | :—: | :—: | :—: | :—: | | M | 0.0% | 6.9% | 12.5% | 22.7% | 0.9% | 12.0% | 1.2% | 26.1% | | C | 18.3% | 0.0% | 6.9% | 3.7% | 11.8% | 38.3% | 3.3% | .. | | L | 41.6% | 20.4% | 0.0% | 4.8% | .. | 15.5% | .. | .. | | KD | 43.8% | 8.2% | 4.7% | 0.0% | 0.0% | 6.2% | .. | 20.2% | | MP | 1.1% | 10.3% | .. | .. | 0.0% | 39.0% | 34.7% | .. | | S | 9.1% | 16.4% | 2.1% | 1.2% | 15.7% | 0.0% | 22.9% | 3.6% | | V | .. | .. | .. | .. | 30.3% | 47.9% | 0.0% | 4.4% | | SD | 44.0% | .. | 1.7% | 21.2% | 1.0% | 6.0% | 1.3% | 0.0% |

Historical Wave: 2018M05 (Pre-Election Wave)

Intended Vote Conditional on Best Party: $P(\text{vote } p \mid \text{best } j)$

| Donor \ Recipient | M | C | L | KD | MP | S | V | SD | | :— | :—: | :—: | :—: | :—: | :—: | :—: | :—: | :—: | | M | 79.9% | 1.6% | .. | 0.5% | .. | .. | 0.0% | 4.2% | | C | 2.7% | 79.8% | 1.0% | .. | 0.0% | .. | .. | 2.3% | | L | 3.8% | 3.0% | 76.5% | .. | 0.0% | .. | .. | .. | | KD | 4.3% | .. | .. | 75.9% | .. | .. | 0.0% | .. | | MP | 4.4% | 3.0% | .. | .. | 67.7% | 3.8% | .. | .. | | S | 1.5% | 0.4% | 0.6% | .. | .. | 81.3% | 1.4% | 1.6% | | V | 0.0% | .. | 0.0% | 0.0% | .. | 3.4% | 84.5% | 1.6% | | SD | 0.9% | 0.0% | 0.0% | .. | 0.0% | .. | .. | 93.8% |

6. Stability of Second-Choice Relationships Over Time

Over the 2010M11–2026M05 window, voters’ declared second-choice parties exhibit persistent bloc structures with quantifiable variation:

• Moderate Party (M) voters:
  • Second-choice KD: Mean = 15.3% (range: 5.6% – 39.1%, std = 8.1%, observed in 29/29 waves).
  • Second-choice L: Mean = 23.5% (range: 12.5% – 36.5%, std = 6.4%, observed in 29/29 waves).
  • Second-choice SD: Mean = 13.3% (range: 3.2% – 26.1%, std = 6.7%, observed in 29/29 waves).
• Social Democrat (S) voters:
  • Second-choice V: Mean = 21.5% (range: 12.9% – 26.2%, std = 3.6%, observed in 29/29 waves).
  • Second-choice MP: Mean = 17.7% (range: 9.1% – 32.8%, std = 6.0%, observed in 29/29 waves).
  • Second-choice C: Mean = 10.2% (range: 2.0% – 22.9%, std = 5.2%, observed in 29/29 waves).
• Christian Democrat (KD) & Liberal (L) voters:
  • KD $\to$ M second choice: Mean = 40.5% (range: 24.3% – 53.9%, std = 7.7%, observed in 29/29 waves).
  • L $\to$ M second choice: Mean = 38.5% (range: 27.9% – 50.0%, std = 6.1%, observed in 29/29 waves).
• Left Party (V) & Green Party (MP) voters:
  • V $\to$ S second choice: Mean = 46.3% (range: 35.8% – 55.3%, std = 5.3%, observed in 29/29 waves).
  • MP $\to$ S second choice: Mean = 38.4% (range: 28.9% – 49.6%, std = 5.5%, observed in 29/29 waves).

Note: These statistics quantify descriptive empirical stability and missingness rates across waves without drawing behavioral or causal inferences about tactical voting.

7. Cross-Party Flow Observability Near the 4% Threshold

For parties that historically or currently poll near the 4% parliamentary threshold (L, KD, MP, C):

1. Large Donor $\to$ Threshold Party Flows:
   • M $\to$ KD: Observed in 29/29 waves (100% coverage). Mean intended vote flow is 1.6% (range 0.4% – 8.3%).
   • M $\to$ L: Observed in 24/29 waves (82.8% coverage, 5 waves suppressed). Mean observed flow is 1.0% (range 0.3% – 1.8%).
   • S $\to$ MP: Observed in 28/29 waves (96.6% coverage, 1 wave suppressed). Mean observed flow is 1.2% (range 0.5% – 4.2%).
   • S $\to$ C: Observed in 24/29 waves (82.8% coverage, 5 waves suppressed). Mean observed flow is 0.7% (range 0.0% – 1.8%).
2. Threshold Party $\to$ Anchor Party Flows:
   • KD $\to$ M: Observed in 23/29 waves (mean 5.0%).
   • L $\to$ M: Observed in 29/29 waves (mean 6.4%).
   • MP $\to$ S: Observed in 29/29 waves (mean 6.6%).
   • V $\to$ S: Observed in 29/29 waves (mean 6.2%).

These high coverage rates indicate that the panel preserves sufficient detail for subsequent empirical analysis of cross-party intentions.

8. SCB Comparability Caveats and Methodology

1. 2020 Weighting Methodology Revision:
   • In 2020, SCB introduced an improved calibration and weighting methodology that incorporates updated register-based auxiliary variables and refined non-response adjustments.
   • SCB recalculated historical Vid10 (Val idag) estimates back to 2010 using this revised model.
   • Other PSU tables (such as Rostningssympati170 and Nastbastaparti190) reflect SCB’s standard survey weighting at publication.
2. Frequency Shift (2023):
   • Prior to 2023, PSU surveyed voters twice per year (May and November).
   • From 2023 onward, PSU is conducted only in May. Researchers using this panel should account for the change in time step ($\Delta t = 6\text{ months}$ pre-2023 vs $\Delta t = 12\text{ months}$ post-2023).
3. Response Universe & Reconciliation:
   • In Vid10, percentage shares are calculated over decided voters expressing a party preference (summing to ~100% across parties).
   • In Rostningssympati170, the universe for each donor category includes blankt and vet ej.
   • Reconstructed estimates from the conditional matrix will differ slightly from headline Vid10 due to turnout weighting, blank/undecided handling, and rounding. These differences are recorded as diagnostic metrics in validation_report.json and are never forced into equality.

9. Processed Dataset Summary

All processed files are stored under data/processed/scb_support_voting/:

• vote_by_sympathy.csv: 3,509 rows (wave × best party × intended vote party).
• second_choice_by_sympathy.csv: 3,509 rows (wave × best party × second-best party).
• overall_vote_intention.csv: 406 rows (wave × party headline Val idag).
• overall_party_sympathy.csv: 290 rows (wave × party overall sympathy for total electorate).
• scb_donor_recipient_panel.csv: 1,856 rows ($8 \times 8 \times 29$ parliamentary donor-recipient panel with separate uncertainty measures).
• validation_report.json: Comprehensive machine-readable report with row sums, reconciliation gaps, coverage metrics, and assertion passes.