The Political Impact of Natural Disasters: Evaluating the Impact of Natural Shocks on Election Outcomes
\(\color{darkblue}{\text{Motivation}}\)
Our project investigates how natural disasters influence voting behavior in Spain, focusing on Catalunya and the Comunitat Valenciana. Specifically, we examine whether exposure to wildfires, an increasingly frequent natural shock, affects political preferences, turnout, and support for incumbent governments.
Recent studies suggest that voters affected by disasters may punish the incumbent government if the response is inadequate. Others find that such events can increase support for conservative parties, particularly when linked to climate change skepticism. Our goal is to test these theories empirically using geospatial data on fire perimeters and granular election outcomes disaggregated at the sección censal level, the highest level of territorial granularity available in Spain. By leveraging spatial analysis techniques, we integrate fire perimeters with administrative boundaries to precisely measure the extent of fire exposure at the local level.
By identifying treated areas as those where at least 20 percent of their territory was affected by fire within an election period and comparing them to unaffected areas, we aim to uncover whether natural shocks drive shifts in political engagement and party preferences.
\(\color{darkblue}{\text{Research Question}}\)
The key questions focus on whether voters in affected areas shift their political preferences.
We examine whether:
exposure to wildfires leads to increased support for conservative parties, particularly in regions where climate change is already shaping economic and environmental concerns.
affected voters are more likely to vote for parties outside the incumbent government.
natural disasters increase voter turnout, as previous research suggests that crisis events can mobilize the electorate.
\(\color{darkblue}{\text{Data}}\)
To analyze the relationship between natural disasters and electoral outcomes, we rely on three primary sources of data.
- Election data
We use detailed electoral outcomes fromCatalunya, and the Comunitat Valenciana, covering general (national), regional, and municipal elections. This data is disaggregated at the sección censal level, the smallest territorial subdivision available, which allows for a precise examination of political shifts within affected areas. Data sources include:- Spain: General elections and local elections from the Spanish Ministry of the Interior. These records require cleaning and aggregation, as results are reported at the mesa level, meaning multiple tables exist within each sección censal.
- Catalunya: Electoral data is structured in a more accessible format, available in clean CSV files with key variables such as turnout and vote shares.
- Comunitat Valenciana: Regional and general election results from 1987 to 2023, available in structured datasets that can be merged with geospatial data.
Geospatial data on administrative boundaries
We incorporate polygon shapefiles defining all secciones censales in Spain from 2001 to 2024. Since the number and shape of secciones have changed over time, we ensure that we match each election to the correct territorial boundaries. This allows us to merge election data with fire perimeters to define treated and control areas.- Geospatial data on wildfires
Our main treatment variable is derived from georeferenced fire perimeter data, available as polygon and multipolygon vector files.- Catalunya: Annual fire perimeter shapefiles from 1986 to 2023, mapped using satellite imagery.
- Comunitat Valenciana: A consolidated dataset covering 1993 to 2022, sourced from GPS records and satellite images. However, it is noted that not all fires are mapped, which may introduce selection bias.
- Catalunya: Annual fire perimeter shapefiles from 1986 to 2023, mapped using satellite imagery.
The limiting factor in our dataset is the availability of census section polygons only from 2001 onward, meaning our analysis is restricted to elections held within this period.
\(\color{darkblue}{\text{Visualizations}}\)
Visualizing Wildfires and Electoral Boundaries
To better understand the spatial distribution of wildfires and their overlap with electoral districts, we use geospatial visualizations that map fire perimeters onto municipal and census section boundaries. The figure below illustrates the affected areas in the Comunitat Valenciana, with fire perimeters in red overlaid on administrative divisions.
These visualizations help identify treated areas where a significant portion of a district has been burned. By combining this with electoral data, we can assess whether exposure to wildfires influences voting behavior.
# Valencia
# Ensure same crs as secciones data:
secciones_v <- st_transform(secciones_v, st_crs(fires_v))
# Create bounding box:
bbox_v <- st_bbox(fires_v)
bbox_polygon_v <- st_as_sfc(bbox_v)
# Plot:
ggplot() +
geom_sf(data = secciones_v, fill = "grey95", color = "grey60", size = 0.1) +
geom_sf(data = fires_v, fill = "red", color = NA) +
coord_sf(
xlim = c(bbox_v["xmin"], bbox_v["xmax"]),
ylim = c(bbox_v["ymin"], bbox_v["ymax"]),
expand = FALSE
) +
labs(title = "Fires in Comunitat Valenciana (1993–2022)") +
theme_minimal() +
theme(
plot.title = element_text(hjust = 0.5, size = 14, face = "bold")
)
Figure 1: Fire perimeters overlaid on administrative boundaries for Comunitat Valenciana
# Catalynua :
# Ensure same crs as secciones data:
secciones_c <- st_transform(secciones_c, st_crs(fires_c))
# Create bounding box:
bbox_c <- st_bbox(fires_c)
bbox_polygon_c <- st_as_sfc(bbox_c)
# Plot:
ggplot() +
geom_sf(data = secciones_c, fill = "grey95", color = "grey60", size = 0.1) +
geom_sf(data = fires_c, fill = "red", color = NA) +
coord_sf(
xlim = c(bbox_c["xmin"], bbox_c["xmax"]),
ylim = c(bbox_c["ymin"], bbox_c["ymax"]),
expand = FALSE
) +
labs(title = "Fires in Catalunya (2001–2023)") +
theme_minimal() +
theme(
plot.title = element_text(hjust = 0.5, size = 14, face = "bold")
)
Figure 2: Fire perimeters overlaid on administrative boundaries for Catalunya
The plots show the geographic distribution of wildfire activity in the Comunitat Valenciana (1993–2022) and Catalunya (2001–2023), with fire perimeters overlaid on municipal boundaries. We can see that Comunitat Valenciana experienced a lot more fires given the data we have. In both regions, fires are not evenly distributed but tend to cluster in specific inland and forested areas, while coastal and urban zones appear less affected. This spatial variation is important for our analysis, as it allows us to compare electoral outcomes in treated areas, those directly affected by fire, with nearby control areas that experienced no exposure. The maps provide an initial sense of the extent and location of treatment across space and time.
\(\color{darkblue}{\text{Analysis & Results}}\)
Empirical Methodology:
We follow closely the structure used in Masiero and Santarossa (2021), who analyze the political effects of earthquakes in Italy using a similar identification strategy. In their study, they compare municipalities affected by seismic events to those not affected, exploiting variation in exposure across time and space while controlling for municipality and time fixed effects. They show that voters in affected areas tend to reward the incumbent, especially when the government responds quickly and effectively. Their empirical approach provides a useful template for assessing the impact of natural shocks on electoral behavior in a different context.
Following this, to estimate the effect of wildfires on electoral outcomes, we adopt a difference-in-differences strategy with time and sección censal fixed effects. For each electoral period, we identify treated areas as those in which more than 20 percent of their area was affected by a wildfire between two elections. Control areas are those not meeting this threshold.
Our identification strategy relies on the assumption that fire occurrence is exogenous with respect to political preferences, and that treated and control areas would have followed similar electoral trends in the absence of fire exposure. The treatment variable is a dummy equal to one for secciones censales that were exposed to fire in the period prior to an election, and zero otherwise.
We estimate the following baseline equation:
\[ Y_{it} = \alpha_i + \gamma_t + \beta \cdot \text{Fire}_{it} + \varepsilon_{it} \]
where \(Y_{it}\) denotes the electoral outcome of interest (such as vote share for incumbent parties, turnout, or a left-right index) in sección \(i\) at time \(t\). The variable \(\text{Fire}_{it}\) is a dummy equal to 1 if sección \(i\) experienced significant fire exposure between elections. \(\alpha_i\) are sección fixed effects, and \(\gamma_t\) are election year fixed effects.
We estimate a difference-in-differences specification:
\[ Y_{it} = \beta_1 (\text{treated}_i \cdot \text{post}_{it}) + \beta_2 (\text{treated}_i \cdot \text{post}_{it} \cdot \text{monthsSinceFire}_i) + X'_{it} \cdot \theta_t + \gamma_t + \alpha_i + \varepsilon_{it} \]
Where:
- \(treated_i\) is a binary indicator equal to 1 if sección \(i\) had a fire in the election period, 0 otherwise.
- \(post_{it}\) is equal to 1 in the period after the fire occurrence. For example, if we are considering the election period 2016–2019 and there was a fire in sección \(i\) in 2017, then \(post_{it}\) equals 0 in 2016 and 1 in 2018 and 2019.
- \(months\_since\_fire_i\) is the number of months between the fire event and the election.
- \(\beta_1\) captures the average treatment effect on the treated secciones censales.
- \(\beta_2\) captures whether the treatment effect varies depending on how recent the fire was.
By including sección fixed effects, we account for time-invariant unobserved characteristics of each unit, such as long-term political leanings or geographic exposure to climate shocks. Year fixed effects absorb common shocks across all areas, such as national political trends or macroeconomic fluctuations. In extended models, we can also explore specifications that control for pre-treatment turnout or socio-demographic characteristics where data allows.