Climate Adaptation of Democracy

DAC Fall 2026

Søren Damsbo-Svendsen

University of Copenhagen

Kasper M. Hansen

University of Copenhagen

Morten A. D. Larsen

Danish Meteorological Institute

September 23, 2026

The CAD project

  • Climate Adaptation of Democracy:
    • Climate impacts on environmental
      and democratic attitudes
    • 2026–2029
  • Funded by the Independent Research Fund Denmark (DFF)
  • Partners:
    • DMI: Morten A. D. Larsen from the Danish Meteorological Institute
    • Also: The Natural Hazards Council, Kystdirektoratet, Operate, Epinion, Scalgo
    • Academics: Pat Egan, Matto Mildenberger, Peter Howe
  • Three project legs – presentation is about the main leg:
    • a ‘preplanned natural experiment’ on storm surge…

Motivation

  • Growing but not yet mature literature on climate-related events
    and their effect on climate attitudes     and political opinions/behavior

  • Methodologically: increasing awareness of the difficulties in designing
    these post-hoc ‘natural experiments’

    • e.g., measurement, definition of events/exposure, power

  • Theoretically: no one has focused on the stress that the broader political system is put under when its citizens react to flooding and other climate events

    • e.g., do citizens still expect the authorities to protect them?
    • do they trust politicians past the next election?
    • would the local community step in?

  • Stakes: if floods raise citizens’ needs while lowering their belief that politics can help them, democracy is put under stress

Question

How do adverse local climate events affect citizens’ attitudes towards democracy and the green transition [part 1], and what strategies for communication can help create effects that are environmentally and democratically sustainable [part 2]?

  • Data: one survey wave fielded shortly after a storm surge (repeated)
    • leverage is in the sampling and high-resolution flood measurement
      (all Danish addresses in flood risk zones, event-driven CPR registry draw)
  • New:
    • design, sampling, outcomes, estimation fixed before the storm
    • flooding modelled at address level rather than self-reported
    • outcomes about the broader political system, not just climate attitudes

Sampling frame


DMI’s 100-year risk zone and qualifying cells

  • 1 km grid cells = communities
  • At least one address in the 100-year risk area
  • Permanent residents only
  • 2,576 cells nationally
  • Based on:
    • DAR (addresses)
    • BBR (buildings)
    • Danish national grid
    • DMI’s flood risk polygons

Treatment = wet declared cells


  • (A) Declared storm surge
    • rarer than a 20-year event
  • and (B) modelled inundation (water)
    • at least one wet address
  • Depth thresholds TBD
    • does “treated” require
      XX mm of water
    • does “control” allow
      YY mm of water

Example: Declared storm surge

Example: (Modelled) wet

Example: Declared AND wet = treated, wet-only = excluded from control

Storm Babet, 21 October 2023 — Oddermose, Møn. Photo: Bdarg, Wikimedia Commons, CC BY-SA 4.0

Controls = dry cells


Table 1

Controls = dry cells


  • Modelled dry = eligible control
  • Undeclared wet → excluded
  • 1 km buffer around treated → excluded
  • Two types of control cells:
    local and distant
  • Combined in main specification (both are matched on elevation, coast distance, density)

Example: Treated and excluded cells

Example: Local controls — same warning district, or near treated

Example: Distant controls — everywhere else in the frame

Sampling of people

  • All treated cells
  • Control cells matched bin by bin on elevation, pop density, distance to coast
  • 1 to 50 households (cap) sampled randomly from each cell — controls thinner
  • 1 person aged 18+ sampled randomly from each household
  • n is decided by the event: from a few thousand to ~20-30k (treated-n = local-n = distant-n)
  • Study cells → addresses → CPR registry (people) → invitation via Digital Post (~4 weeks after storm)

DMI’s nine coastal warning districts

We design for the “community channel”

  • Water reaches cells far more often than people
  • At most 5–15% of a treated cell’s addresses in the flood zone
  • Hard to identify flooded individuals
  • Cell effect is community effect + personal effect, mostly the first
  • Community effects are policy-relevant, not just a proxy
  • (Note: we’ll also pursue estimation of individual-level estimands
    and of moderation by communities’ ‘water dose’)
  • So: we select outcomes that may be affected in someone who witness a local flood

Outcomes

Five short indices, 3–5 items each:

  • P1 — expected protection and support — will myndighederne be there for me? A flood is the state failing to keep people dry, and the state turning up afterwards
  • P2 — demand for climate adaptation policy — should the state build the defences, and would you pay for them? Water plausibly increases support
  • P3 — can politicians protect? — perceived capacity and will. Coastal defence pays off long after the next election, so this asks whether people believe democratic politics can act on that timescale
  • P4 — expected support from local community (social capital) — if the state or municipality is doubted, do neighbours etc. take its place
  • C1 — climate concern and importance — do floods move climate attitudes at all? Three DNES items, directly comparable to existing research

Only these five will be preregistered.

Question examples

If a serious flood hit my area, I could count on help from the authorities.

P1 — expected protection and support

The government should invest heavily in climate adaptation — coastal protection, flood barriers, drainage.

P2 — demand for climate adaptation policy

Politicians would rather spend money on something voters can see before the next election than on protection against flooding twenty years from now.

P3 — can democratically elected politicians protect?

If a serious flood hit my area, neighbours and the local community would help each other.

P4 — expected support from community and civil society

Also in the questionnaire


  1. Checks — perceived flood likelihood and preparedness, self-reported exposure, summer-house ownership
  2. Supplementary outcomes
    • vote intention, retrospective evaluation of government and municipality
    • deservingness of compensation, willingness to help
    • democratic satisfaction and institutional trust, responsibility attribution
  3. Covariates — age, gender, origin, education, income, savings, housing tenure
  4. Simple embedded framing experiment(s)
    • e.g., community-orientation vs individual-responsibility

Identification

  • Post-event, cross-sectional
    • design-stage matching + regression ladder
    • identification by as-if-random inundation within the risk area
    • given similar elevation, distance to coast, pop density, and surrounding water (fetch)
  • Quasi-random physics decide where water actually goes
    (storm track, wind, micro-elevation)
  • Threat: residential sorting, e.g., onto slightly higher ground or to a better view
    • wealth, education
  • Covariate ladder with 9 rungs (all reported), 3-17 cell-level covariates
    • physical core (5) → district SES (share owners, household wealth, share tertiary education) → cell’s respondent composition [primary]→ outcome predictors (e.g., lagged vote, origin)

Conclusions

  • RQ: How do adverse local climate events affect citizens’ attitudes towards democracy and the green transition […]?
  • The bigger question: whether democracy can adapt as climate risks keep rising — whether being flooded changes what citizens trust and expect: the state protecting them, politicians acting ahead of disaster, neighbours coming together
  • Core will be preregistered before the 2026/27 storm season
    • an addendum covers storm-specifics: sample etc.
  • Honest concern: low power (as in most similar research)
  • All comments are welcome:
    • sampling/design, outcomes, identification, power
    • paper-writing/publishing perspectives, ideas for later

Thank you

Appendix

P1 — expected protection and support

  • Q3.5 If a serious flood hit my area, I trust that the emergency response would reach everyone who needed it.
  • Q3.6 If a serious flood hit my area, I trust that help would be distributed fairly among those affected.
  • Q3.6b If a serious flood hit my area, I expect that residents would largely be left to cope on their own. (R)
  • Q3.6c If a serious flood hit my area, I could count on help from the authorities.

P2 — demand for climate adaptation policy

  • Q4.10 The government should invest heavily in climate adaptation measures (e.g., coastal defences, flood barriers, improved drainage).
  • Q4.21 I would be willing to pay higher taxes if the money went to large-scale climate adaptation in Denmark, such as coastal protection and sea barriers.
  • Q4.10b Climate adaptation should have lower priority than other public tasks in the coming years. (R)
  • Q4.7 More public resources should be allocated to protecting the climate.

P3 — whether politicians can and will protect

  • Q3.2 Politicians take account of what citizens want when they make decisions about the climate.
  • Q4.16 Even though coastal protection and other flood defences only pay off long after the next election, I expect politicians to carry them through.
  • Q4.16b Politicians are not able to protect us against sea-level rise, flooding and other consequences of climate change. (R)
  • Q4.16c Politicians would rather spend money on something voters can see before the next election than on protection against flooding twenty years from now. (R)

P4 — expected support from community and civil society

  • Q3.6d If a serious flood hit my area, neighbours and the local community would help each other.
  • Q3.6e If a serious flood hit my area, volunteers and local associations would step in.
  • Q3.6f If a serious flood hit my area, I would get help from people I do not know.
  • Q3.6g If something serious happened where I live, people would be able to manage it together.
  • Q3.16b If a serious flood hit my area, most people would mainly look after themselves. (R)

C1 — climate concern and importance

  • Q2.1 Climate change is the greatest challenge of our time.
  • Q2.2 I am afraid that climate change will have consequences for me and my loved ones.
  • Q2.2c I am willing to accept a lower standard of living in order to protect the climate.

Sampling, step by step

  1. Treated = cells that are declared and wet.
  2. Drop from control eligibility: every wet cell, and the 1 km ring around the treated set.
  3. Split what remains into two arms: local — same warning district, or within 20 km of a treated cell — and distant, the rest of the frame.
  4. Pick the matching grid by rule: 4 x 4 x 3 = 48 bins where support allows, else 3 x 3 x 2 or 2 x 2 x 2.
  5. Cut the bins at quantiles of the treated distribution (elevation, coast distance, address density) and place treated and eligible cells in a bin.
  6. Treated invitations: up to 50 addresses per treated cell, sampled at random — all of them if the cell has fewer than 50.
  7. Each bin’s target n, per control arm = the treated invitations in that bin — so the treated share is 1/3 with both arms (1/2 with one of them).
  8. Distribute that target n across the bin’s eligible cells:
    1. evenly, in rounds, while remaining invitations outnumber eligible cells;
    2. one each to a random subset when they do not (report under-fill if the cells run out of addresses)
  9. Within each selected cell: draw the allocated number of addresses at random.
  10. Send the address list to CPR and request the extract — residents at those addresses.
  11. Before any outcome data are seen:
    1. describe the realised sample — cells per arm, bin fill, under-fill, and balance on cell-level BBR variables — and recompute power on the realised numbers;
    2. file the event addendum to the preregistration: the storm and the sample.
  12. Hand the extract to Epinion, who invite by Digital Post — about 4 weeks after the surge.

Controls — local, distant, and pooled

  • Local — same warning district as a treated cell, or within 20 km. Above all the same latent probability of flooding: same water, same surge regime, cells that could have been treated. With it the same warning, news and municipal response — and closer socio-geographically: settlement type, labour market, political culture. Nearer on what we cannot measure.
  • Distant — the rest of the frame: (relatively) far from actual exposure to this surge. No warning, no local coverage — still coastal, still at risk in principle.
  • Within local, declared-dry against undeclared-dry asks whether the declaration alone moves attitudes on dry land.
  • Why both arms: each buys off a different problem. Local controls may have been moved by the storm themselves — spillover — so treated-vs-local runs small. Distant controls carry whatever regional confounding the matching missed, in either direction. Assume one away and the other bounds it. Never clean, but a diagnostic: agreement means both are small, divergence tells us which to argue about.
  • The headline pools them — for power. Either arm alone is around half the control invitations, too much precision to spend on a diagnostic. Pooling holds because both arms are matched on the same cell specifications — elevation, coast distance, address density

Covariate ladder

Table 2