Parliamentary activity

Compare speeches, questions, and initiatives by party across 2019–2026. Adjust for seats or sitting MPs to compare parties of different sizes.

Speech figures describe the selected analytical sample, not every parliamentary speech. See dataset scope.

Speeches come from plenary puheenvuoro records; the type buckets collapse the Finnish labels into six groups (Regular, Reply, Fast, Group, Intro, Other). Questions & initiatives are first-signer matters: KK written, SKT oral, VK interpellation, TPA/TAA/KA motions. Seat counts are from the official Vaalit.fi results (2019, 2023). "Sitting MPs" counts MPs who gave at least one speech in the term. Note the systematic government/opposition effect: governing parties file far fewer questions because they don't interrogate themselves. In the all-years view, per-seat figures divide by the combined seats across both electoral terms (seat-terms), while per-MP figures use distinct speakers across the whole period.

Which party talks about what?

Compare the topic shares of each party’s speeches with parliament overall.

Which topics do parties emphasise?

Read one party row at a time. A value of 1 means the same topic share as parliament overall; 2 means twice that share. Darker cells mean greater relative emphasis, not more speeches.

The baseline stays fixed when filtering government/opposition. Topics use the selected model’s classifications of the shared speech sample.

Ideological compass

Compare parties’ economic and values positions across datasets. Each point shows one party’s position in one dataset.

Party positions on the compass

Horizontal axis: economic policy

−2 = left, +2 = right.

Vertical axis: values (GAL–TAN)

−2 = green and liberal (GAL), +2 = traditional and nationalist (TAN).

Select at least one dataset to show party positions.

Clear and neutral scores (−2…+2) are averaged by party. Mixed and unscorable positions are excluded from the means.

How do position and agenda shift over time?

Party positions and parliamentary topics by year (2019–2026). Choose the speech analysis model separately for each chart. All models evaluated the same sample of 22,167 speeches.

The documentation below describes the original dataset. Model comparison scope and sample counts appear beside the charts’ dataset controls.

About

• Speeches
  All three language models evaluated the same sample of 22,167 plenary speeches, selected from records collected through the Eduskunta open-data speech search. It accumulated from capped API result pages in API return order; historical run settings were not persisted. The collected dataset’s exact IDs are in a machine-readable manifest (SHA-256 5e2dcbbabe9d869abd5b206be6b3a23b5401ffe51a130219562970468e46f248; yearly counts: 2019: 3,450, 2020: 6,180, 2021: 6,100, 2022: 6,095, 2023: 3,280, 2024: 6,290, 2025: 5,990, 2026: 4,143).
• MPs
  281 distinct speakers with recorded speeches.
• Questions & initiatives
  8,266 party-attributed matters from parliamentary years 2019, 2020, 2021, 2022, 2023, 2024, 2025, 2026: written and oral questions (KK, SKT), interpellations (VK), and motions (TPA, TAA, KA).
• Stance scores
  The number of speeches with accepted scores varies by model and filters. The MP map and speech-position comparison use the selected speech analysis model. Counts of speeches with scores differ from the total evaluated sample. One speech can contribute a score on each axis; axis scores are not a count of speeches.
• Area tags
  The original topic dataset uses title-based keyword rules. The speech-model comparison uses model classifications of the same speech sample.
• Parliamentary years
  2019, 2020, 2021, 2022, 2023, 2024, 2025, 2026

Area classification

There are 13 policy areas. In the speech-model comparison, each model classifies the speech text; the party is withheld from the prompt. The original topic dataset uses keyword rules over the title of the matter being debated. Questions and initiatives are also classified by title keywords. When no rule matches, the category is “muu” (other). Producing a classification does not establish its accuracy; both methods can make mistakes.

Stance scoring

Every non-empty speech is eligible for scoring, with no minimum length. Speeches in the selected sample are evaluated separately on both axes using a scale from −2 to +2. If a speech contains no explicit position on an axis, that axis is marked unscorable and excluded from the mean:

The prompt instructs the model to reason from speech substance, not speaker identity, and marks speeches without an explicit position as unscorable. Party is withheld to prevent speaker-prior bias. The published scores used deepseek-v4-flash. The prompt is available below.

Olet suomalaisen politiikan asenneanalyytikko. Sinulle annetaan yksi
eduskuntapuheenvuoro. Arvioi puhujan oma kanta ERIKSEEN kahdella akselilla.

TALOUSAKSELI (economic), vasen ↔ oikea:
−2 = vaatii julkisen sektorin, uudelleenjaon tai työntekijän aseman LAAJENTAMISTA
     (lisää palveluita, korkeampi verotus, vahvempi sääntely/ay-asema)
−1 = puolustaa nykyisiä julkisia palveluita tai etuja leikkauksia vastaan
 0 = nimenomaisesti keskitie tai tasapaino talouskysymyksessä
+1 = kannattaa maltillista markkinaehtoisuutta, veronkevennystä tai menokuria
+2 = vaatii selvästi markkinaehtoisuutta, merkittäviä veronkevennyksiä tai
     julkisen sektorin supistamista

ARVOAKSELI (gal_tan), GAL ↔ TAN:
  GAL = vihreä, vaihtoehtoinen, liberaali, kosmopoliittinen
  TAN = perinteinen, autoritaarinen, nationalistinen
−2 = vahva GAL (yksilönvapaus, avoimuus, pluralismi, yli-rajainen yhteistyö)
−1 = maltillinen GAL
 0 = nimenomaisesti keskitie tai tasapaino arvokysymyksessä
+1 = maltillinen TAN
+2 = vahva TAN (kansallinen sulkeutuneisuus, järjestys, perinteiset normit)

Säännöt jokaiselle akselille:
- Perusta arvio VAIN puhujan omaan, eksplisiittiseen kantaan. Älä käytä puoluetta,
  henkilön tunnettuja mielipiteitä tai yksittäisiä avainsanoja oikoteinä.
- Hallituksen puolustaminen tai opposition arvostelu EI OLE oma kanta. Kysy:
  esittääkö puhuja oman vaihtoehdon, vai reagoiko hän vain toisten esityksiin?
  Jos jälkimmäinen, akseli on "unscorable".
- Useimmat puheet ottavat kantaa vain TOISEEN akseliin. Se on odotettua: merkitse
  toinen "unscorable". Älä pakota kantaa akselille, jota puhe ei käsittele.
- Aihe ei ratkaise akselia: talouspuhe voi ottaa arvokannan ja päinvastoin.

status jokaiselle akselille:
- "clear": selvä oma suunta; stance −2, −1, +1 tai +2.
- "neutral": puhuja kannattaa eksplisiittisesti keskitietä; stance 0.
- "mixed": vastakkaisia omia kantoja ilman hallitsevaa suuntaa; stance 0.
- "unscorable": tähän akseliin liittyvää omaa kantaa ei ole; stance 0.

Vastaa kentät tässä järjestyksessä jokaiselle akselille: poimi ensin SANATARKKA
lainaus puheesta ("evidence"), päätä sitten status ja stance. Älä tiivistä tai
muotoile lainausta uudelleen; jos sanatarkkaa lainausta ei ole (esim. unscorable),
jätä "evidence" tyhjäksi.

AIHE (taustatieto): {matter}

PUHEENVUORO:
{speech}

Party compass

A party's place on the compass is the mean of its stance-scored speeches on each axis. Each party therefore gets one point: (economic, GAL-TAN). The same coordinates are compared below to CHES, Manifesto, and EVA.

Activity

The activity view counts plenary speeches as well as questions and initiatives (by first signer). Figures are shown by party, as raw counts or relative to seat totals or sitting MPs. Without this adjustment, larger parties have higher counts simply because of their size.

Seat counts come from Vaalit.fi (2019 and 2023). Here, the Marin term means parliamentary years 2019–2022 and the Orpo term 2023–2026. The all-years per-seat figure uses the combined seat count across both terms.

The speech-compass party ordering is compared with external series using Pearson correlation. This is a directional sanity check, not independent accuracy validation: only eight parties overlap, the series cover different years, and no uncertainty intervals were calculated.

1. Eduskunta open-data API: api.eduskunta.fi. Valtiopaivaasia and Puheenvuoro categories.
2. Yle election questionnaire, Finnish parliamentary election 2023: vaalikone.yle.fi. The open data published by Yleisradio Oy is licensed under CC BY (license); this project processes and analyzes the material. YLE open data.
3. CHES: Chapel Hill Expert Survey. Expert judgments on party positions. 2014 Chapel Hill Expert Survey, Version 2015.1; 0–10 GAL-TAN → −2..+2 via (x-5)·0.4. GAL-TAN mapped with (x−5)×0.4; economic values are the project's normalized comparison series.
4. Manifesto: Manifesto Project (MARPOR). Coded party election manifestos. Manifesto Project Dataset 2025a; RILE -100..+100 → -2..+2 via x/50. RILE mapped with x/50; GAL-TAN is a project-defined approximation from cultural categories.
5. EVA: EVA value survey. Finnish citizen value survey. Suomen kuvitettu arvokartta 2023; Economic/social -1..+1 → -2..+2 via ×2. Coordinates were transcribed from the original publications into the comparison series and multiplied by two.
6. Ohjelmat: Party programmes (DeepSeek). DeepSeek-scored party principle & election programmes (Pohtiva archive), same axes. program-stance-v1; −2..+2 per-party mean across all programmes. Programme-level DeepSeek scores are averaged by party across the stored Pohtiva snapshot.
7. YSO, the Finnish General Ontology: finto.fi/yso.