Waking Up the Golden Dawn
Does Exposure to Refugees Increase Support for Extreme-Right Parties?
Theoretical Framework
Causal Framework (DAG)
Show the code
# ------------------------------------------------------------
# Goal: Draw a simple Directed Acyclic Graph (DAG) that represents
# a theory of how exposure to refugees affects Golden Dawn (GD) vote share.
# This DAG uses ggdag + dagitty + ggplot2.
#
# Key idea: We draw *nodes* (circles) for variables and *arrows* for
# hypothesized causal effects. We'll also show how to make
# arrows shorter so they don't touch the nodes.
# ------------------------------------------------------------
# --- 0) Playground: tweak these safely -------------------------------
node_size <- 25 # how big each node (circle) is
node_alpha <- 0.5 # transparency of node fill (0 = invisible, 1 = solid)
arrowhead_pt <- 10 # arrowhead size in points (this is the *tip* triangle)
trim <- 0.20 # how much of the arrow *shaft* to chop off from BOTH ends
# e.g., 0.20 = remove 20% at the start AND 20% at the end
# so the arrow in the middle is 60% of the original length
# --- 1) Load required packages ----------------------------------------------
# ggdag → helper functions to draw DAGs in ggplot2
# dagitty → defines DAG objects and relationships between variables
# ggplot2 → plotting system used to visualize the DAG
library(ggdag)
library(dagitty)
library(ggplot2)
# --- 2) Define the DAG structure --------------------------------------------
# dagify() creates a DAG object where you specify arrows (causal paths).
# The notation A ~ B means "A is caused by B".
# Here:
# GDvote ← threat (Golden Dawn vote share caused by perceived threat)
# threat ← exposure (Threat is caused by exposure to refugees)
# exposure ← distance (Exposure caused by proximity to Turkey)
# Labels: provide readable names for the DAG nodes (with line breaks).
# Coords: x and y positions of each node (for nice plotting layout).
theory_dag <- dagify(
GDvote ~ threat, # DV (Golden Dawn vote share) depends on threat
threat ~ exposure, # Threat depends on exposure
exposure ~ distance, # Exposure depends on distance
labels = c(
GDvote = "Golden Dawn\nVote Share (DV)",
threat = "Perceived Threat\n(Mediator)",
exposure = "Refugee Arrivals",
distance = "Proximity to Turkey"
),
coords = list(
x = c(distance = 0, exposure = 1, threat = 2, GDvote = 3),
y = c(distance = 1, exposure = 1, threat = 1, GDvote = 1)
)
)
# --- 3) Convert DAG into a data frame for ggplot ----------------------------
# tidy_dagitty() extracts coordinates, labels, and edges for plotting.
# as.data.frame() makes it into a tibble/data frame usable by ggplot.
dag_df <- as.data.frame(tidy_dagitty(theory_dag, layout = "auto"))
# Extract plotting limits (so the nodes don’t touch the plot borders).
min_x <- min(dag_df$x); max_x <- max(dag_df$x)
min_y <- min(dag_df$y); max_y <- max(dag_df$y)
error <- 0.2 # small padding around the plot
# --- 3.5) Make arrow *segments* shorter and floating between nodes ----------
# Important: In ggplot, the "arrow()" only controls the arrowhead (the little
# triangle tip). It does NOT control the length of the shaft. Shaft length is
# determined by the start (x, y) and end (xend, yend) coordinates.
#
# To make arrow shafts shorter AND not start at the node centers, we:
# 1) Find all edge rows (they have xend/yend values).
# 2) Move the start point forward by 'trim' proportion along the edge.
# 3) Move the end point backward by 'trim' proportion along the edge.
# The result: the arrow appears between nodes without touching them.
edges_df <- subset(dag_df, !is.na(xend) & !is.na(yend))
# Safety checks: ensure trim is between 0 and 0.49
# (If trim >= 0.5 the start could pass the end!)
trim <- max(min(trim, 0.49), 0.00)
# Compute trimmed start (xstart_trim, ystart_trim) and end (xend_trim, yend_trim)
edges_df$xstart_trim <- edges_df$x + trim * (edges_df$xend - edges_df$x)
edges_df$ystart_trim <- edges_df$y + trim * (edges_df$yend - edges_df$y)
edges_df$xend_trim <- edges_df$xend - trim * (edges_df$xend - edges_df$x)
edges_df$yend_trim <- edges_df$yend - trim * (edges_df$yend - edges_df$y)
# --- 4) Plot the DAG --------------------------------------------------------
# Each node is a big semi-transparent point with a label on top.
# Edges (arrows) are drawn with geom_dag_edges_arc using the *trimmed* endpoints.
ggplot(data = dag_df) +
# Draw big transparent points for each node
geom_dag_point(aes(x = x, y = y), size = node_size, alpha = node_alpha) +
# Add text labels (only once per node → !duplicated(label))
geom_label(
data = subset(dag_df, !duplicated(label)),
aes(x = x, y = y, label = label),
fill = alpha("white", 0.8)
) +
# Draw arrows between nodes (edges of the DAG)
# NOTE: arrow(length = ...) controls ONLY the arrowhead size (the tip).
# The shaft length is controlled by the start/end coordinates we just trimmed.
geom_dag_edges_arc(
data = edges_df,
aes(x = xstart_trim, y = ystart_trim, xend = xend_trim, yend = yend_trim),
curvature = 0.0,
arrow = grid::arrow(length = grid::unit(arrowhead_pt, "pt"), type = "closed")
) +
# Set x- and y-limits to keep everything inside the plot
coord_sf(
xlim = c(min_x - error, max_x + error),
ylim = c(min_y - error, max_y + error)
) +
# Minimal background → removes axes, gridlines, etc.
theme_void()
Figure 2
Findings
Exposure and Electoral Divergence
Show the code
# ------------------------------------------------------------
# Goal: Make a simple, readable line chart showing average
# Golden Dawn (GD) vote share over time for Exposed vs Not Exposed
# municipalities. This is a *descriptive* plot, not a test.
# ------------------------------------------------------------
# 0) Load packages ------------------------------------------------------------
# ggplot2 → plotting
# dplyr → "verbs" for data wrangling (mutate, group_by, summarize)
library(ggplot2)
library(dplyr)
# 2) Create a clean, small summary table for plotting -------------------------
# - gdvote is assumed to be a proportion (0–1), so multiply by 100.
# - "group" becomes a readable label for the legend.
df_muni_avg <- datamuni %>%
mutate(
gdper = gdvote * 100,
group = ifelse(treatment_group == 1, "Exposed", "Not Exposed")
) %>%
group_by(group, year) %>%
summarize(
gd_mean = mean(gdper, na.rm = TRUE), # average GD% for that group×year
n = dplyr::n() # how many municipalities in this cell
)
# 3) Make the plot ------------------------------------------------------------
ggplot(df_muni_avg, aes(x = year, y = gd_mean, color = group, group = group)) +
geom_line(linewidth = 1.2) + # thicker line for readability
geom_point(size = 3) + # add dots so each year is clear
scale_color_manual(values = c("Exposed" = "red", "Not Exposed" = "blue")) +
labs(
x = "Election Year",
y = "Golden Dawn Vote Share (%)",
color = "Municipalities",
title = "Golden Dawn Support in Exposed vs. Non-Exposed Municipalities",
subtitle = "A simple descriptive comparison over time"
) +
# theme_bw = clean base theme; base_size nudges fonts larger for slides
theme_bw(base_size = 13) +
# Small style tweaks: move legend to bottom, lighten gridlines
theme(
legend.position = "bottom",
panel.grid.minor = element_blank()
)
Figure 3