When the election rains out and how bad weather excludes marginal voters from turning out

class: center, middle, inverse, title-slide

# When the election rains out and how bad weather excludes marginal voters from turning out
## MPSA 2022
### Søren Damsbo-Svendsen sdas@ifs.ku.dk
### Kasper Møller Hansen kmh@ifs.ku.dk
### Department of Political Science University of Copenhagen
### 08 April 2022

---

layout: true

.pull-left[
 
<img src="data:image/png;base64,#media/election-rainout_frontpage.png" width="55%" style="display: block; margin: auto;" />
]

---

.pull-right[ 
**RQ: How does Election Day weather affect voters' decision to turn out for election &mdash; and for whom?**

- Electoral turnout is a key __indicator of health and legitimacy__ of democracy

- Focus on rainfall: does it reduce turnout?

- *Mechanism:* **Cost of voting** (and possibly __mood__)
]

---

.pull-right[ 
First, we did a **meta-analysis:**

- Rainfall effect

- 33 studies

- Mixed designs and results

- (But) avg. rainfall effect: **-0.76 %-points per cm**
]

---

.pull-right[ 
Secondly, we did a new **stand-alone study** with great data:

1. Revisit **rainfall-turnout** thesis
  
2. Explore **other weather** variables
  
3. **Nonlinear** weather effects

4. **Heterogeneous** weather effects
]

---
layout: false

# Data: turnout records

Individual turnout records for 4,549,145* Danish voters age 18-80

- **Validated** registry data
    - &rarr; high quality 
    - [handful of previous studies]

- (Almost) **complete electorate** 
    - &rarr; huge N 
    - [first study]

- **Repeated measurements** at 2013 and 2017 local elections 
    - &rarr; voter panel 
    - [first study]

???

- Missing data on ~1% (2013) and ~9% (2017) of eligible voters because of non-participating municipalities.

- Exclusion of ~3% early voting and voters above age 80

---

# Data: objective weather observations

- Voters' home coordinates matched with **166 weather stations**

- __Individual weather__ *triangulated* from three nearest weather stations

---

# Design and models

.pull-left[
**(1) Pooled model**

- Regression of turnout on *rainfall*, *sunshine*, *temperature* with municipality and election FEs

- cross-sectional variation
    
    - assumption of (conditional) as-if randomness in assignment of weather 
]

.pull-right[
**(2) Panel model**

- Regression of turnout on *rainfall*, *sunshine*, *temperature* with voter and election FEs (TWFE)

- within-unit variation over time
 
 - assumption of no selection into treatment from *time-varying* individual-level factors that shape where one lives (i.e., local weather)
 
 - stronger basis for causal inference, first in the literature
]

???

- __Additional controls:__ (individual) age, gender, living close to coast, municipal population size, share of non-Western immigrants, and closeness of the election

---
class: title-slide, center, middle

# Results

---
class: middle

## Result I: Negative rainfall effect

- **Rainfall** &mdash; "bad weather" &mdash; reduces probability of voting

- **-0.84 %-points** per centimeter rain, on average

- Substantial difference between panel model &uarr; and pooled model

---
class: middle

## Result II: Positive sunshine effect

- __Sunshine__ &mdash; "nice weather" &mdash; also boosts turnout

- **1.52 %-points** diff. between max and min observed sunshine

---
class: middle

## Result III: Nonlinear rainfall effect

.left-column[

- **Rainfall effect** increases with more rain

- No effect at zero rain, approx. **-3.5 %-points** at max

- (Sunshine effect is not nonlinear)
]

.right-column[
<img src="data:image/png;base64,#media/nonlinear_panel.png" width="1000px" style="display: block; margin: auto;" />
]

---
class: middle

## Result IV: Stronger effects on marginal voters

- Important Q &rarr; **Are marginal voter groups more susceptible?**

- **Potentially major issue** for democratic representation

- We focus on **young voters** in their twenties who have
    - left their childhood home 
    - not yet established a solid voting habit

???

Marginal voters: "*those whose decisions to turn out are sensitive to exogenous factors*" (Fowler 2015: 205)

<div class="figure" style="text-align: center">
<img src="data:image/png;base64,#media/turnout_by_age.png" alt="'The rollercoaster ride of turnout and age' (descriptive turnout by age at KV13-KV17)" width="1000px" />
'The rollercoaster ride of turnout and age' (descriptive turnout by age at KV13-KV17)
</div>

---

## Result IV: Stronger effects on marginal voters

- Young voters are influenced up to **seven times more susceptible** by the weather

- **No effect** on first-time voters in *first cohort*

- **Very strong effect** on *second and third cohort*

---
background-color: #ECECF2

# In sum

We revisit the question of how Election Day weather affects turnout with (a) a **meta-analysis** of the rainfall-turnout thesis and (b) a **stand-alone study** &darr;

- __rainfall__ *does* reduce turnout 
 
 - effect comparable to avg in existing literature 
 
 - effect is **nonlinear**

- **sunshine** also increases turnout

- **marginal voters** &mdash; young voters without robust turnout habit &mdash; are *much more susceptible*

---
layout: true
class: center, middle
background-color: #ECECF2
---

# Thanks for the attention
 
<img src="data:image/png;base64,#media/rain-election.png" width="50%" style="display: block; margin: auto;" />
Edward Linsmier—Getty Images 2014, time.com/3554884/2014-election-weather-forecast

---
 
<img src="data:image/png;base64,#media/interaction_pool.png" width="1000px" style="display: block; margin: auto;" />
---
<img src="data:image/png;base64,#media/weather-turnout-table.png" width="80%" style="display: block; margin: auto;" />
---
<img src="data:image/png;base64,#media/election-rainout_meta.png" width="53%" style="display: block; margin: auto;" />