Politologisk forskning med brug af vejr- og klimaobservationer

Præsentation på DMI

Kasper M. Hansen

Professor, ph.d.
Institut for statskundskab
Københavns Universitet

Søren Damsbo-Svendsen

Ph.d.-stipendiat
Institut for statskundskab
Københavns Universitet

16. januar 2023

Hvordan vi bruger vejr- og klimaobservationer i politologisk forskning

Plan

1. Vejr og valgdeltagelse
2. Temperatur og klimaholdninger
3. Stormflod og vælgeradfærd
4. Opsamling

1: Vejr og valgdeltagelse

1: Vejr og valgdeltagelse

Hvordan påvirker vejret på valgdagen vælgernes beslutning om at stemme?

• Fokus på nedbør: reducerer det valgdeltagelsen?

• Stemmeprocenten er en nøgleindikator på demokratiers sundhed og legitimitet

• Mekanisme: omkostninger (og potentielt også humør mv.)

Metaanalyse af nedbør og valgdeltagelse

• N = 27 (34 i alt)

• Forskellige data, designs og resultater

• De fleste negative:

  • Simpelt gns.: -0.82 %-point per cm
  • Vægtet gns.: -0.42 %-point per cm

Selvstændigt studie med bedre data

Data: valgdeltagelse

• Registerdata med valgdeltagelse for 4.549.145 vælgere (18-80 år)
  • Næsten komplet befolkning
  • KV2013 og KV2017 → vælgerpanel

Data: vejrobservationer

• Vælgernes hjemmekoordinater matchet med 166 vejrstationer

• Individuel vejroplevelse trianguleret fra tre nærmeste vejrstationer

Forskningsdesign og modeller

To typer modeller:

• 1. en “pooled model”
• 2. en “panel model” (primær)

• Regression af valgdeltagelse (0/1) på nedbør, solskin, temperatur med vælger og valg fixed effects

  • stærk basis for kausal inferens

  • supplerende kontrolvariable: alder, køn, tæthed på kyst, urbanisering, kommunestørrelse, kommuneindvandrere, valgets tæthed

Resultater 1: Negativ effekt af nedbør

• Nedbør — “dårligt vejr” — mindsker sandsynligheden for at stemme

• -0.95 %-point per centimeter regn, i gns.

• Nedbør effect vokser med mere regn (ikke-lineær)

• Ingen effekt ved 0 regn, næsten -4 %-point ved maks.

• Solskin (stråling) — “godt vejr” — øger også deltagelsen

Resultater 2: Stærk effekt på marginale vælgere

• Vigtigt spørgsmål: Er marginale vælgergrupper mere påvirkelige?

• Potentielt problem for demokratisk repræsentation

• Vi fokuserer på unge vælgere i 20erne (bor ikke hjemme, har ikke solid vane om at stemme)

• Unge vælgere påvirkes op til seks gange mere af vejret

  • Ingen effekt på førstgangsvælgere (første kohorte)

  • Meget stærke effekter på 2. og 3. kohorte

2: “Lokal opvarmningseffekten”

2: “Lokal opvarmningseffekten”

RQ: How do unusual temperatures affect Europeans’ opinions towards climate change?

• Direct experiences with the weather, independent of elite discourse (?)

• The local warming effect is when “deviations in the day’s temperature affect individuals’ global warming beliefs”

• Recent weather experiences as a heuristic

• Inspired by Egan & Mullin (2012) and others in the US

Data

• European Social Survey round 8 (2016)

• Dependent variable: climate opinion index (0-100, based on 6 questions)

• Independent variable: temperature over the past 7 days compared with “normal” in last 30 years

• Temperatures observered at country-level matched with respondents in each country

  • I exclude extremely large/mountainous countries → 12 remaining countries

Resultater (1)

• Regression analysis of relationship between temperatures and opinions

• → Result: clear positive association

• Identifying assumption: as-if random allocation of temperature

• Experiences with unusually hot weather causes strengthened climate opinions

• Fairly weak effect - but everyone is constantly affected by weather

• Weather experiences as “cheap”, readily available information about climate change

Resultater (2)

Who is affected by the weather? Who relies on the weather heuristic?

• Two potential moderators: education (political awareness, sophistication) and partisanship (motivation)

→ Result: some suggestive evidence, but no; everyone is affected

Resultater (3)

Is it exclusively a direct effect of personal experience?

• The media (elites) could play the role of framing/articulating weather experiences as politically relevant

  • connect weather and climate change — and provide “indirect experiences”

  • → the effect would be larger for people with higher news consumption

→ Result: suggestive evidence, but no: news exposure makes no difference

3: Stormflod og klimastemmer

Vikingeskibsmuseet

Theory: extreme weather experience → behavior

RQ: How does climate-related extreme weather experience affect pro-climate voting?

• Ample evidence that extreme weather experiences affects climate change perceptions and attitudes
• But limited evidence that weather experiences affect actual behavior:
  • environmental behavior: it does not increase private energy saving, recycling, etc. (e.g. Rüttenauer 2021)
  • information/communication behavior: it can increase Twitter climate discussion, community dicussion, and web searches (e.g. Boudet et al. 2019; Osberghaus & Demski 2019; Sisco et al. 2017)
• How about more collective, political behavior?
  • Pro-climate voting!

Theory: extreme weather experience → behavior

RQ: How does climate-related extreme weather experience affect pro-climate voting?

Pro-climate voting:

• “a sizeable effect for pro-climate voting [20 %-points] after experiencing a flood” (Baccini & Leemann 2021)
• Californian wildfires increase support for costly, pro-environment policies by 5-6 %-points (Hazlett & Mildenberger 2020)
• Studies of retrospective voting and natural disasters sometimes show electoral rewards for (relatively) pro-climate parties (e.g. Gasper & Reeves 2011; Bechtel & Hainmueller 2011)
• Some null findings (e.g. Hilbig & Riaz, 2022; Melville-Rea 2022)

The 2013 storm surge in Denmark

• A rarely harmful and costly weather event → “Hurricane Sandy-equivalent”
• Data: detailed and complete geocoded storm surge insurance cases from the Danish Storm Council
• N = 2,924

Study I: voting for pro-climate parties

• Level of analysis: 1,386 polling districts

• Context: Danish parliamentary elections 1994-2019 (focus on 2011-2015)

• Outcome: district vote share for pro-climate parties

  • the left-green bloc w/ strong issue ownership

• Matched with district flooding based on 3 treatment definitions (all binary):

  • Storm surge damages (USD) above average
  • 1+ cases of rehousing cases
  • 1+ total flooding cases

• Difference-in-differences (DID) design

Study I: pretreatment balance

→ flooded and unflooded areas are fairly well-balanced on most covariates

Study I: parallel trends

→ generally, parallel pretrends (with some imbalances); 2015 (post) stands out!

Study I: DID results - main

→ flooding effect on pro-climate voting approx. 2-4 %-points

Study I: DID robustness - propensity-score matching

• Following Abadie (2005)
• Covariates: ln(polling district area), ln(distance to the water), ln(historical average turnout), academic education (%), primary education (%), ln(median income), ln(total wealth), single-family homes (%)
• DID estimate (damages): 2.06 (bootstrapped 95% CI: [1.11, 3.02])
• No placebo effects

→ flooding effect robust to covariate matching

Study I: DID results - flooding intensity

→ flooding effect tends to grow with flooding intensity

Study II: voting for pro-climate candidates

• Level of analysis: 9,554 election candidates

• Context: Danish local elections 2009-2017 (focus on 2013-2017)

• Outcome: elected or not (binary)

• Data: candidate survey w/ climate-related policy responses from 73% of candidates

• Pro-climate candidate (binary): climate as key priority or climate score above 4/5

• Treatment: high economic damage in municipality where candidate is running

• Triple differences (DIDID) design

Study II: pro-climate candidates

Key expectations

1. If voters engage in pro-climate voting,
2. the gap in election chances of pro-climate candidates versus non-climate candidates
3. should increase in heavily flooded communities (because of the flood)
4. → positive DIDID

Pro-climate candidates within each party (%)

Study II: DIDID results

• \(Treatment\:group \times Post\): electoral punishment of -4.0 pp for non-climate candidates because of the flood (small reward for pro-climate)
• \(Climate \:candidate \times Post\): no change over time (1 pp) in election chances of non-climate candidates in non-flooded areas
• \(Treatment\:group \times Climate \:candidate\): election chances of pro-climate candidates in flooded areas were only slightly higher before the flood (2.3 pp)

Study II: DIDID results

• DIDID, all parties: approx. 5.5 %-points

• → pro-climate candidates benefit electorally in post-flooding elections

Opsummering

Opsummering

Nogle spørgsmål.

Tak

→ kmh@ifs.ku.dk, sdas@ifs.ku.dk

Marco Faria