Measuring Democracy with V-Dem: Demonstrations
Jan Rovny
24 September 2026
Data
We use two packages from the V-Dem Institute: vdemdata,
which holds the country-year dataset, and ERT, which
identifies Episodes of Regime Transformation (get_eps(),
formerly part of vdemdata). Both are large, so everything
needed below is computed once and cached in vdem_cache.rds.
Delete that file to rebuild from a newer release.
library(dplyr)
library(tidyr)
library(ggplot2)cache_file <- "vdem_cache.rds"
# Czechoslovakia (1918-1992) and the Czech Republic form one continuous
# V-Dem series under the code "CZE".
countries <- c(CZE = "Czechoslovakia / Czech Republic",
HUN = "Hungary",
POL = "Poland",
FRA = "France",
USA = "United States")
# Threshold grid for the episode sensitivity analysis in section 5
eps_grid <- expand_grid(start_incl = c(0.01, 0.02),
cum_incl = c(0.05, 0.10, 0.15, 0.20))
if (!file.exists(cache_file)) {
if (!requireNamespace("remotes", quietly = TRUE)) install.packages("remotes")
if (!requireNamespace("vdemdata", quietly = TRUE))
remotes::install_github("vdeminstitute/vdemdata")
if (!requireNamespace("ERT", quietly = TRUE))
remotes::install_github("vdeminstitute/ERT")
vdem_full <- vdemdata::vdem
# Five focal countries, all variables
vdem_sub <- vdem_full |>
filter(country_text_id %in% names(countries))
# All countries, a few variables, for the comparison with Polity / Freedom House
vdem_global <- vdem_full |>
select(country_name, country_text_id, year,
v2x_polyarchy, e_p_polity, e_fh_pr, e_fh_cl)
# Episodes of regime transformation: default thresholds + sensitivity grid
eps_keep <- function(e) filter(e, country_text_id %in% names(countries))
eps_default <- eps_keep(ERT::get_eps(data = vdem_full))
eps_sens <- eps_grid |>
rowwise() |>
mutate(eps = list(eps_keep(ERT::get_eps(data = vdem_full,
start_incl = start_incl,
cum_incl = cum_incl)))) |>
ungroup()
saveRDS(list(vdem_sub = vdem_sub, vdem_global = vdem_global,
eps_default = eps_default, eps_sens = eps_sens), cache_file)
rm(vdem_full)
}
cache <- readRDS(cache_file)
vdem_sub <- cache$vdem_sub |>
mutate(country = factor(countries[country_text_id], levels = countries))
vdem_global <- cache$vdem_global
eps_default <- cache$eps_default
eps_sens <- cache$eps_sens
vdem_sub |>
group_by(country) |>
summarise(first_year = min(year), last_year = max(year), n_years = n())## # A tibble: 5 x 4
## country first_year last_year n_years
## <fct> <dbl> <dbl> <int>
## 1 Czechoslovakia / Czech Republic 1918 2025 108
## 2 Hungary 1789 2025 237
## 3 Poland 1789 2025 171
## 4 France 1789 2025 237
## 5 United States 1789 2025 237Shared plotting objects:
indices <- c(v2x_polyarchy = "Electoral",
v2x_libdem = "Liberal",
v2x_partipdem = "Participatory",
v2x_delibdem = "Deliberative",
v2x_egaldem = "Egalitarian")
country_cols <- c("Czechoslovakia / Czech Republic" = "#1b9e77",
"Hungary" = "#d95f02",
"Poland" = "#7570b3",
"France" = "#e7298a",
"United States" = "#66a61e")
theme_set(theme_minimal(base_size = 12) +
theme(panel.grid.minor = element_blank(),
strip.text = element_text(face = "bold")))
# Long format of the five indices. complete() inserts empty years so that
# lines break across gaps in coding (e.g. Poland 1796-1808) instead of
# being drawn straight through them.
hli <- vdem_sub |>
select(country, year, all_of(names(indices))) |>
group_by(country) |>
complete(year = full_seq(year, 1)) |>
ungroup() |>
pivot_longer(all_of(names(indices)), names_to = "index", values_to = "score") |>
mutate(index = factor(indices[index], levels = indices))1. The five high-level indices
V-Dem does not offer a single measure of democracy. It measures five conceptions, each scaled from 0 to 1.
Electoral democracy (v2x_polyarchy).
This is Dahl’s polyarchy and the core of all the other indices. It
combines five components: elected officials (v2x_elecoff),
clean elections (v2xel_frefair), freedom of association
(v2x_frassoc_thick), suffrage (v2x_suffr) and
freedom of expression and alternative sources of information
(v2x_freexp_altinf). The index is the average of a
multiplicative term, in which a zero on any component sinks the whole
score, and a weighted additive term, in which strengths compensate for
weaknesses.
Liberal democracy (v2x_libdem). Adds
the protection of individual and minority rights against the state and
the majority: equality before the law and individual liberties, judicial
constraints on the executive, and legislative constraints on the
executive (v2x_liberal).
Participatory democracy
(v2x_partipdem). Adds active citizen participation
beyond voting: civil society participation, direct popular votes, and
elected local and regional government (v2x_partip).
Deliberative democracy (v2xdl_delib component,
v2x_delibdem index). Adds the quality of public
reasoning: whether elites justify positions and refer to the common
good, respect counterarguments, consult widely, and whether society is
engaged in public debate.
Egalitarian democracy (v2x_egaldem).
Adds equality in the exercise of rights and freedoms: equal protection
across social groups, equal access to power, and equal distribution of
the resources needed to participate (v2x_egal).
Each of the four “thick” indices combines the electoral index with its own component using the same formula, for example for the liberal index:
\[\text{LDI} = 0.25 \cdot \text{EDI}^{1.585} + 0.25 \cdot \text{LCI} + 0.5 \cdot \text{EDI}^{1.585} \cdot \text{LCI}\]
The exponent penalises weak electoral democracy, so a country cannot score highly on any conception of democracy without holding meaningful elections.
france_events <- tibble(
year = c(1792, 1799, 1815, 1830, 1848, 1852, 1870, 1940, 1944, 1958),
label = c("First Republic", "Consulate / Empire", "Restoration", "July Monarchy",
"Second Republic", "Second Empire", "Third Republic", "Vichy",
"Liberation", "Fifth Republic"))
ggplot(filter(hli, country == "France"), aes(year, score, colour = index)) +
geom_vline(data = france_events, aes(xintercept = year),
linetype = "dotted", colour = "grey60") +
geom_text(data = france_events, aes(x = year, y = 1.02, label = label),
inherit.aes = FALSE, angle = 90, hjust = 1, vjust = -0.3,
size = 2.8, colour = "grey40") +
geom_line(linewidth = 0.7, na.rm = TRUE) +
scale_y_continuous(limits = c(0, 1.02), breaks = seq(0, 1, 0.25)) +
scale_x_continuous(breaks = seq(1800, 2025, 25)) +
scale_colour_brewer(palette = "Dark2") +
labs(title = "France: V-Dem's five democracy indices, 1789 to present",
x = NULL, y = "Index score (0-1)", colour = NULL,
caption = "Source: V-Dem Country-Year dataset.") +
theme(legend.position = "bottom")
Points for discussion:
- The deliberative and egalitarian indices start only in 1900 (1918 for the Central European countries), because some of their components are not coded in V-Dem’s historical data. Longer series come at the cost of narrower concepts.
- The indices share their big movements (the Second Republic, the Third Republic, Vichy), because every thick index is built on the electoral one.
- The Third Republic sits in the middle of the scale until 1944 largely because of the exclusion of women: suffrage is one of the five electoral components, and the jump at the Liberation is mostly women’s enfranchisement.
- The ordering of the indices is itself a finding. France scores consistently lowest on participatory democracy, reflecting a centralised state with limited direct democracy, a gap that an electoral measure alone would hide.
2. Five countries over time
The same five indices, now with one panel per index to compare countries. Each series starts at the country’s earliest V-Dem observation: 1789 for France, Hungary, Poland and the United States, 1918 for Czechoslovakia.
ggplot(hli, aes(year, score, colour = country)) +
geom_line(linewidth = 0.55, na.rm = TRUE) +
facet_wrap(~ index, ncol = 2) +
scale_colour_manual(values = country_cols) +
scale_y_continuous(limits = c(0, 1), breaks = seq(0, 1, 0.25)) +
scale_x_continuous(breaks = seq(1800, 2025, 50)) +
labs(title = "V-Dem democracy indices, earliest observation to present",
x = NULL, y = "Index score (0-1)", colour = NULL,
caption = "Source: V-Dem Country-Year dataset. Gaps indicate years without coding.") +
theme(legend.position = c(0.75, 0.13))
Points for discussion:
- Poland’s series has gaps (1796-1808, 1868-1917 and 1939-1943). V-Dem codes political units that are not always sovereign states, such as the Duchy of Warsaw and Congress Poland, and stops where no such unit exists. What is the unit of observation for a nation without a state?
- Under communism the egalitarian index for the three East-Central European countries sits well above their liberal, participatory and deliberative scores. The egalitarian component rewards equal access to resources and power, which these regimes partly delivered. Is that part of democracy, or a separate outcome?
- The United States leads France on the electoral index until 1870 but trails it from the Third Republic until the 1980s, because suffrage restrictions and unclean elections in the South cap its score until the mid-1960s.
3. Measurement uncertainty
V-Dem’s expert-coded indicators are aggregated with a Bayesian item
response theory model, which estimates each coder’s reliability and
scale use. The indices therefore come with uncertainty bounds,
_codelow and _codehigh, which mark the
interval of one standard deviation around the point
estimate (roughly 68%). A 95% interval would be about twice as wide, so
overlap at 68% is a conservative test of distinguishability.
cee <- vdem_sub |>
filter(country_text_id %in% c("CZE", "HUN", "POL"), year >= 2005) |>
mutate(country = droplevels(country))
ggplot(cee, aes(year, v2x_libdem,
ymin = v2x_libdem_codelow, ymax = v2x_libdem_codehigh,
colour = country, fill = country)) +
geom_ribbon(alpha = 0.2, colour = NA) +
geom_line(linewidth = 0.8) +
geom_point(size = 1.2) +
scale_colour_manual(values = country_cols) +
scale_fill_manual(values = country_cols) +
scale_x_continuous(breaks = seq(2005, 2025, 5)) +
labs(title = "Liberal democracy index with uncertainty bounds, 2005 to present",
subtitle = "Lines are point estimates; bands span codelow to codehigh (one SD)",
x = NULL, y = "Liberal democracy index", colour = NULL, fill = NULL,
caption = "Source: V-Dem Country-Year dataset.") +
theme(legend.position = "bottom")
Which pairs are statistically distinguishable in each year? The table marks a pair as overlapping when its intervals intersect.
overlap <- function(df, a, b) {
x <- filter(df, country_text_id == a)
y <- filter(df, country_text_id == b)
inner_join(x, y, by = "year", suffix = c("_a", "_b")) |>
transmute(year,
pair = paste(a, "vs", b),
leader = ifelse(v2x_libdem_a > v2x_libdem_b, a, b),
overlap = v2x_libdem_codelow_a <= v2x_libdem_codehigh_b &
v2x_libdem_codelow_b <= v2x_libdem_codehigh_a)
}
overlaps <- bind_rows(overlap(cee, "CZE", "POL"),
overlap(cee, "CZE", "HUN"),
overlap(cee, "POL", "HUN"))
overlaps |>
mutate(cell = ifelse(overlap, paste0(leader, " (overlap)"), leader)) |>
select(year, pair, cell) |>
pivot_wider(names_from = pair, values_from = cell) |>
knitr::kable(caption = "Higher point estimate in each pair; '(overlap)' = intervals intersect")| year | CZE vs POL | CZE vs HUN | POL vs HUN |
|---|---|---|---|
| 2005 | CZE (overlap) | CZE (overlap) | POL (overlap) |
| 2006 | CZE (overlap) | CZE (overlap) | POL (overlap) |
| 2007 | CZE (overlap) | CZE (overlap) | POL (overlap) |
| 2008 | POL (overlap) | CZE (overlap) | POL (overlap) |
| 2009 | POL (overlap) | CZE (overlap) | POL (overlap) |
| 2010 | POL (overlap) | CZE | POL |
| 2011 | POL (overlap) | CZE | POL |
| 2012 | POL (overlap) | CZE | POL |
| 2013 | POL (overlap) | CZE | POL |
| 2014 | POL (overlap) | CZE | POL |
| 2015 | POL (overlap) | CZE | POL |
| 2016 | CZE | CZE | POL |
| 2017 | CZE | CZE | POL |
| 2018 | CZE | CZE | POL |
| 2019 | CZE | CZE | POL |
| 2020 | CZE | CZE | POL |
| 2021 | CZE | CZE | POL (overlap) |
| 2022 | CZE | CZE | POL (overlap) |
| 2023 | CZE | CZE | POL |
| 2024 | CZE | CZE | POL |
| 2025 | CZE | CZE | POL |
Points for discussion:
- Between 2005 and 2015 the Czech Republic leads Poland in the point estimates, then Poland leads from 2008, yet their intervals overlap in every year: neither ranking is supported by the data.
- Before 2010 all three countries, including Hungary, are indistinguishable. The divergence that is now taken for granted only becomes measurable after Fidesz’s 2010 victory.
- In 2021 and 2022 Poland and Hungary overlap again, despite Poland’s higher point estimate. Descriptions of Poland as “not yet Hungary” at the time were more confident than the measurement allowed.
4. Comparison with Polity and Freedom House
V-Dem includes other democracy measures: Polity’s combined score
(e_p_polity, −10 to +10, available to 2018) and Freedom
House’s political rights and civil liberties ratings
(e_fh_pr, e_fh_cl, 1 = most free to 7 = least
free, from 1972). Polity’s special codes for interruption, interregnum
and transition (−66, −77, −88) are set to missing, and the two Freedom
House ratings are averaged and reversed so that higher means freer.
compare <- vdem_global |>
mutate(polity = ifelse(e_p_polity < -10, NA, e_p_polity),
fh = 8 - (e_fh_pr + e_fh_cl) / 2,
focal = country_text_id %in% names(countries))
compare_long <- compare |>
select(country_text_id, year, focal, v2x_polyarchy, polity, fh) |>
pivot_longer(c(polity, fh), names_to = "measure", values_to = "value") |>
filter(!is.na(value), !is.na(v2x_polyarchy)) |>
mutate(measure = recode(measure,
polity = "Polity (-10 to +10)",
fh = "Freedom House (1 to 7, reversed)"))
compare_long |>
group_by(measure) |>
summarise(n = n(), correlation = cor(value, v2x_polyarchy)) |>
knitr::kable(digits = 2, caption = "Correlation with V-Dem electoral democracy, all country-years")| measure | n | correlation |
|---|---|---|
| Freedom House (1 to 7, reversed) | 8332 | 0.93 |
| Polity (-10 to +10) | 16255 | 0.86 |
ggplot(compare_long, aes(value, v2x_polyarchy)) +
geom_jitter(data = filter(compare_long, !focal), width = 0.15, height = 0,
alpha = 0.06, size = 0.6, colour = "grey30") +
geom_jitter(data = filter(compare_long, focal), width = 0.15, height = 0,
alpha = 0.6, size = 0.9, colour = "#d95f02") +
geom_smooth(method = "loess", se = FALSE, colour = "black", linewidth = 0.7) +
facet_wrap(~ measure, scales = "free_x") +
scale_y_continuous(limits = c(0, 1)) +
labs(title = "V-Dem electoral democracy against Polity and Freedom House",
subtitle = "All country-years; orange = the five focal countries",
x = "Alternative measure", y = "V-Dem electoral democracy index",
caption = "Source: V-Dem Country-Year dataset.")
How much variation does Polity’s top score hide?
compare |>
filter(!is.na(polity), !is.na(v2x_polyarchy), year >= 1946) |>
mutate(polity_band = cut(polity, c(-11, -6, 5, 9, 10),
labels = c("Autocracy (-10 to -6)", "Anocracy (-5 to 5)",
"Democracy (6 to 9)", "Full democracy (10)"))) |>
group_by(polity_band) |>
summarise(country_years = n(),
edi_min = min(v2x_polyarchy),
edi_median = median(v2x_polyarchy),
edi_max = max(v2x_polyarchy)) |>
knitr::kable(digits = 2, caption = "V-Dem electoral democracy within Polity bands, 1946-2018")| polity_band | country_years | edi_min | edi_median | edi_max |
|---|---|---|---|---|
| Autocracy (-10 to -6) | 3394 | 0.01 | 0.14 | 0.44 |
| Anocracy (-5 to 5) | 2237 | 0.01 | 0.30 | 0.78 |
| Democracy (6 to 9) | 2108 | 0.10 | 0.62 | 0.91 |
| Full democracy (10) | 1794 | 0.29 | 0.85 | 0.92 |
And the three measures over time for the focal countries, each rescaled to 0-1:
compare |>
filter(focal, year >= 1900) |>
mutate(country = factor(countries[country_text_id], levels = countries),
`V-Dem electoral` = v2x_polyarchy,
`Polity` = (polity + 10) / 20,
`Freedom House` = (fh - 1) / 6) |>
select(country, year, `V-Dem electoral`, Polity, `Freedom House`) |>
pivot_longer(-c(country, year), names_to = "measure", values_to = "score") |>
ggplot(aes(year, score, colour = measure)) +
geom_line(linewidth = 0.6, na.rm = TRUE) +
facet_wrap(~ country, ncol = 2) +
scale_colour_manual(values = c("V-Dem electoral" = "black",
"Polity" = "#1f78b4",
"Freedom House" = "#e31a1c")) +
scale_x_continuous(breaks = seq(1900, 2025, 25)) +
labs(title = "Three democracy measures rescaled to 0-1, 1900 to present",
x = NULL, y = "Rescaled score", colour = NULL,
caption = "Polity ends in 2018; Freedom House starts in 1972.") +
theme(legend.position = c(0.75, 0.12))
Points for discussion:
- Overall the measures agree closely, so aggregate correlations are reassuring. The disagreement is concentrated in the middle of the scale, among hybrid regimes, which is precisely where most substantive questions lie.
- Polity has a ceiling. About a fifth of post-1946 country-years score +10, and among them V-Dem’s electoral index ranges widely. Polity cannot register erosion that stays within its top category: Hungary and Poland still score +10 in 2018, when V-Dem has Hungary as an electoral autocracy. The United States scores 9 or 10 throughout the 20th century while much of its Black population was excluded from voting.
- Freedom House’s 1-7 scale is coarse, so its series move in steps, and it registers Hungary’s and Poland’s backsliding in the 2010s less sharply than V-Dem does.
5. Detecting autocratization
The ERT algorithm (Maerz et al.) identifies episodes from annual changes in the electoral democracy index. With the default thresholds, an episode starts with an annual change of at least 0.01, becomes a manifest episode once the cumulative change reaches 0.10, and ends after five years of stagnation or a reversal (an annual change of 0.03, or a cumulative change of 0.10, in the opposite direction).
outcome_aut <- c("0" = NA, "1" = "Democratic breakdown", "2" = "Preempted breakdown",
"3" = "Diminished democracy", "4" = "Averted regression",
"5" = "Regressed autocracy", "6" = "Outcome censored")
outcome_dem <- c("0" = NA, "1" = "Democratic transition", "2" = "Preempted transition",
"3" = "Stabilised electoral autocracy", "4" = "Failed liberalisation",
"5" = "Democratic deepening", "6" = "Outcome censored")
regimes <- c("Closed autocracy", "Electoral autocracy",
"Electoral democracy", "Liberal democracy")
window <- tibble(country_text_id = c("HUN", "POL"),
from = c(2000, 2010))
ep_data <- eps_default |>
inner_join(window, by = "country_text_id") |>
filter(year >= from) |>
mutate(country = countries[country_text_id],
regime = factor(regimes[v2x_regime + 1], levels = regimes))
episode_spans <- bind_rows(
ep_data |> filter(aut_ep == 1) |>
group_by(country, id = aut_ep_id) |>
summarise(start = min(year), end = max(year), .groups = "drop") |>
mutate(type = "Autocratization"),
ep_data |> filter(dem_ep == 1) |>
group_by(country, id = dem_ep_id) |>
summarise(start = min(year), end = max(year), .groups = "drop") |>
mutate(type = "Democratization"))
ggplot(ep_data, aes(year, v2x_polyarchy)) +
geom_rect(data = episode_spans,
aes(xmin = start - 0.5, xmax = end + 0.5, ymin = -Inf, ymax = Inf, fill = type),
inherit.aes = FALSE, alpha = 0.18) +
geom_ribbon(aes(ymin = v2x_polyarchy_codelow, ymax = v2x_polyarchy_codehigh),
fill = "grey70", alpha = 0.5) +
geom_line() +
geom_point(aes(colour = regime), size = 2) +
facet_wrap(~ country, scales = "free_x") +
scale_fill_manual(values = c(Autocratization = "#e31a1c", Democratization = "#1f78b4")) +
scale_colour_manual(values = c("Closed autocracy" = "#67000d", "Electoral autocracy" = "#ef3b2c",
"Electoral democracy" = "#6baed6", "Liberal democracy" = "#08306b"),
drop = FALSE) +
labs(title = "Episodes of regime transformation: Hungary and Poland",
subtitle = "Shading = ERT episodes (default thresholds); points = Regimes of the World",
x = NULL, y = "Electoral democracy index", fill = "Episode", colour = "Regime type",
caption = "Source: V-Dem; ERT package.") +
theme(legend.position = "right")
All episodes since 1985 for the five countries:
bind_rows(
eps_default |> filter(aut_ep == 1) |>
distinct(country_text_id, id = aut_ep_id, start = aut_ep_start_year,
end = aut_ep_end_year, outcome = outcome_aut[as.character(aut_ep_outcome)],
censored = aut_ep_censored) |>
mutate(type = "Autocratization"),
eps_default |> filter(dem_ep == 1) |>
distinct(country_text_id, id = dem_ep_id, start = dem_ep_start_year,
end = dem_ep_end_year, outcome = outcome_dem[as.character(dem_ep_outcome)],
censored = dem_ep_censored) |>
mutate(type = "Democratization")) |>
filter(start >= 1985) |>
mutate(country = countries[country_text_id],
censored = ifelse(censored == 1, "ongoing", "")) |>
arrange(country, start) |>
select(country, type, start, end, outcome, censored) |>
knitr::kable(caption = "ERT episodes with default thresholds")| country | type | start | end | outcome | censored |
|---|---|---|---|---|---|
| Czechoslovakia / Czech Republic | Democratization | 1990 | 1993 | Democratic transition | |
| Czechoslovakia / Czech Republic | Autocratization | 2013 | 2021 | Averted regression | |
| Hungary | Democratization | 1988 | 1991 | Democratic transition | |
| Hungary | Autocratization | 2006 | 2025 | Democratic breakdown | ongoing |
| Poland | Democratization | 1989 | 1992 | Democratic transition | |
| Poland | Autocratization | 2016 | 2021 | Averted regression | |
| Poland | Democratization | 2023 | 2025 | Democratic deepening | ongoing |
| United States | Autocratization | 2024 | 2025 | Outcome censored | ongoing |
When does a decline count as an episode?
The episodes above depend on researcher-chosen thresholds. Below, the start threshold (annual change needed to open an episode) and the cumulative threshold (total change needed for the episode to count) are varied, holding the termination rules at their defaults.
sens_long <- eps_sens |>
mutate(eps = lapply(eps, \(e) select(e, country_text_id, year, aut_ep, dem_ep))) |>
unnest(eps) |>
filter(year >= 1985) |>
mutate(status = case_when(aut_ep == 1 ~ "Autocratization",
dem_ep == 1 ~ "Democratization"),
country = factor(countries[country_text_id], levels = rev(countries)),
start_lab = paste("start_incl =", start_incl),
cum_lab = paste("cum_incl =", format(cum_incl, nsmall = 2)))
ggplot(filter(sens_long, !is.na(status)), aes(year, country, fill = status)) +
geom_tile(height = 0.8) +
facet_grid(start_lab ~ cum_lab) +
scale_fill_manual(values = c(Autocratization = "#e31a1c", Democratization = "#1f78b4")) +
scale_x_continuous(limits = c(1984.5, 2025.5), breaks = seq(1990, 2025, 10)) +
scale_y_discrete(drop = FALSE) +
labs(title = "Episode detection under alternative thresholds",
subtitle = "Default ERT settings: start_incl = 0.01, cum_incl = 0.10",
x = NULL, y = NULL, fill = NULL) +
theme(legend.position = "bottom", panel.grid.major.y = element_blank())
Points for discussion:
- With default thresholds the algorithm dates Hungary’s autocratization to 2006, not 2010. The start is triggered by a small dip in 2006 (just over the 0.01 threshold); the index then recovers slightly but not enough to end the episode, and the real decline begins in 2010. Raise the start threshold to 0.02 and the episode starts in 2010 and ends around 2018, once Hungary has stabilised as an electoral autocracy.
- Hungary’s episode crosses the Regimes of the World threshold into electoral autocracy (2018), so it is classified as a democratic breakdown. Poland’s 2016-2021 episode stops within electoral democracy and is coded as an averted regression, followed by a democratization episode from 2023, provisionally coded as democratic deepening and still ongoing.
- With default settings the Czech Republic also registers an autocratization episode (2013-2021) and the United States one starting in 2024. Raise the cumulative threshold to 0.15 and both disappear, while Hungary and Poland remain. Lower it to 0.05 and a 2016-2018 US episode appears. Whether the Czech Republic or the United States “autocratized” is partly a function of where the bar is set. France registers no episode under any setting.
- At a cumulative threshold of 0.20, Poland’s post-2023 recovery is not yet large enough to count as an episode.
- The termination rules matter too: an episode ends only after five years of stagnation or a clear reversal, so the ongoing US episode will be classified only once enough years have passed.