The Court of Justice of the European Union (CJEU) is the principal judicial institution of the European Union (EU). It consists of two courts: the somewhat confusingly named, higher-instance Court of Justice (CJ) and the lower-instance General Court (GC). There used to be a third court, the Civil Service Tribunal (CST), which was abolished and whose jurisdiction was transferred to the GC.
This automatically updated research document gives a real-time overview of the EU courts’ output and engagement with national courts through the preliminary ruling procedure. It relies on data collected via the eurlex package for R1 Because neither Eur-Lex nor Curia contains the complete record of all court proceedings and documents, the data presented below should be viewed as a reasonable approximation of the true state of the world.2
As of 11 December 2024, EU courts delivered a total of
41684 decisions in about 49113 cases submitted
to them since 1953. According to public
records, 1147 cases are currently pending before the
Court of Justice and 1653 before the General Court. The
following table shows the most recent cases received by the CJ:
Even more than any other branch of the EU,
the CJEU’s caseload changed dramatically over the course of European
integration. In 1953 – the first year of its functioning – the CJ
received 4 cases. Last year (2023), it
received 816, while the GC received 1250.
# packages
library(eurlex)
library(ggplot2)
library(dplyr)
library(purrr)
library(tidyr)
library(stringr)
library(forcats)
library(rmarkdown)
library(modelsummary)
library(ggiraph)
library(DT)
library(lubridate)
library(maps)
# seed
set.seed(35239)
# current date
date_now <- Sys.Date()
date_now_f <- str_remove(format(date_now, "%d %B %Y"), "^0")
day_now <- as.integer(str_sub(date_now, 9, 10))
year_now <- as.integer(str_sub(date_now, 1, 4))
# citation
cit_page <- paste("Michal Ovádek, '",rmarkdown::metadata$title,"', available at https://michalovadek.github.io/eucourt/, accessed on ", date_now_f,
sep = "")
# data on the EU
# enlargement rounds
enlargements <- dplyr::tibble(
year = c(1952,1973,1981,1986,1995,2004,2007,2013,2020),
n_ms = c(6,9,10,12,15,25,27,28,27),
ms_composition = c(
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~United Kingdom",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~United Kingdom~~~Greece",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~United Kingdom~~~Greece~~~Portugal~~~Spain",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~United Kingdom~~~Greece~~~Portugal~~~Spain~~~Austria~~~Finland~~~Sweden",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~United Kingdom~~~Greece~~~Portugal~~~Spain~~~Austria~~~Finland~~~Sweden~~~Cyprus~~~Czechia~~~Estonia~~~Hungary~~~Latvia~~~Lithuania~~~Malta~~~Poland~~~Slovakia~~~Slovenia",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~United Kingdom~~~Greece~~~Portugal~~~Spain~~~Austria~~~Finland~~~Sweden~~~Cyprus~~~Czechia~~~Estonia~~~Hungary~~~Latvia~~~Lithuania~~~Malta~~~Poland~~~Slovakia~~~Slovenia~~~Bulgaria~~~Romania",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~United Kingdom~~~Greece~~~Portugal~~~Spain~~~Austria~~~Finland~~~Sweden~~~Cyprus~~~Czechia~~~Estonia~~~Hungary~~~Latvia~~~Lithuania~~~Malta~~~Poland~~~Slovakia~~~Slovenia~~~Bulgaria~~~Romania~~~Croatia",
"Belgium~~~France~~~Germany~~~Italy~~~Luxembourg~~~Netherlands~~~Denmark~~~Ireland~~~Greece~~~Portugal~~~Spain~~~Austria~~~Finland~~~Sweden~~~Cyprus~~~Czechia~~~Estonia~~~Hungary~~~Latvia~~~Lithuania~~~Malta~~~Poland~~~Slovakia~~~Slovenia~~~Bulgaria~~~Romania~~~Croatia"
)
)
# indicate EU membership
accessions <- tibble(
country = c("Austria",
"Belgium",
"Bulgaria",
"Croatia",
"Cyprus",
"Czechia",
"Denmark",
"Estonia",
"Finland",
"France",
"Germany",
"Greece",
"Hungary",
"Ireland",
"Italy",
"Latvia",
"Lithuania",
"Luxembourg",
"Malta",
"Netherlands",
"Poland",
"Portugal",
"Romania",
"Slovakia",
"Slovenia",
"Spain",
"Sweden",
"United Kingdom"),
year = c(1995,
1958,
2007,
2013,
2004,
2004,
1973,
2004,
1995,
1958,
1958,
1981,
2004,
1973,
1958,
2004,
2004,
1958,
2004,
1958,
2004,
1986,
2007,
2004,
2004,
1986,
1995,
1973),
eumember = 1L
)
# create MS panel
ms_years_panel <- dplyr::tibble(year = 1952:year_now) |>
dplyr::left_join(enlargements) |>
tidyr::fill(n_ms,ms_composition) |>
dplyr::mutate(dplyr::across(.cols = c(year,n_ms),
.fns = ~as.integer(.))) |>
tidyr::separate_rows(ms_composition, sep = "~~~") |>
dplyr::rename(country = ms_composition)
# get curia list
curia_list <- elx_curia_list("all", parse = TRUE)
# some corrections
curia_list <- curia_list |>
mutate(see_case = case_when(
str_detect(see_case, "[C|T|F] [:digit:]+/[:digit:]+") ~ str_replace(see_case, "(?<=[C|T|F]) (?=[:digit:])", "-"),
T ~ see_case
)) |>
mutate(case_id = case_when(
case_id == "6-64" ~ "6/64",
T ~ case_id
))
# code vars
curia_list <- curia_list |>
filter(nchar(case_id) > 3) |>
mutate(court = case_when(
str_detect(case_id, "OPIN|C-|^[:digit:]|RULING") ~ "CJ",
str_detect(case_id, "T-") ~ "GC",
str_detect(case_id, "F-") ~ "CST",
T ~ NA_character_
)) |>
mutate(see_case_court = case_when(
str_detect(see_case, "OPIN|C-|^[:digit:]|RULING") ~ "CJ",
str_detect(see_case, "T-") ~ "GC",
str_detect(see_case, "F-") ~ "CST",
T ~ NA_character_
)) |>
mutate(appeal_court = case_when(
str_detect(appeal, "OPIN|C-|^[:digit:]|RULING") ~ "CJ",
str_detect(appeal, "T-") ~ "GC",
str_detect(appeal, "F-") ~ "CST",
T ~ NA_character_
)) |>
mutate(case_status = case_when(
str_detect(case_id, "OPIN|RULING") ~ "Opinion",
str_detect(case_info, "^Judge?ment") ~ "Judgment",
str_detect(case_info, "^Order") ~ "Order",
str_detect(case_info, "^Removed") ~ "Removal",
str_detect(case_info, "^Pending") & is.na(ecli) ~ "Pending",
str_detect(case_info, "^Seizure order") ~ "Seizure order",
str_detect(case_info, "^Decision") & str_detect(case_id, "RX") ~ "Re-examination",
str_detect(case_info, "^Third-party proce") ~ "Third-party proceedings",
!is.na(see_case) & court != see_case_court ~ "Transferred",
!is.na(see_case) & court == see_case_court ~ "Joined",
T ~ NA_character_
)) |>
mutate(case_year = str_extract(case_id, "/[:digit:]{2}"),
case_year = str_remove(case_year, "/"),
case_year = case_when(
str_detect(case_year, "^0|^1|^2|^3") ~ as.integer(str_c("20", case_year)),
str_detect(case_year, "^5|^6|^7|^8|^9") ~ as.integer(str_c("19", case_year)),
T ~ NA_integer_
),
case_number = as.integer(str_extract(case_id, "[:digit:]+(?=/)")),
decision_year = as.integer(str_extract(ecli, "[:digit:]{4}"))) |>
mutate(decision_date = str_extract(case_info, "[:digit:]{1,2} (January|February|March|April|May|June|July|August|September|October|November|December) (19|20)[:digit:]{2}"))
#filter(is.na(ecli), is.na(see_case))
#filter(is.na(case_status)) |> View()
#count(case_status) |> arrange(-n)
# for initial para
n_cases <- curia_list |>
filter(case_status != "Transferred") |>
reframe(n = n_distinct(case_id))
n_decisions <- curia_list |>
#filter(!case_status %in% c("Transferred", "Joined", "Pending")) |>
reframe(n = n_distinct(ecli))
n_pending <- curia_list |>
filter(case_status == "Pending") |>
group_by(court) |>
reframe(n = n_distinct(case_id))
# n year court cases
n_cases_year_court <- curia_list |>
#filter(case_status != "Transferred") |>
group_by(court, case_year) |>
reframe(n = n_distinct(case_id)) %>%
right_join(., expand(., court, case_year)) |>
mutate(n = ifelse(is.na(n), 0L, n))# consistent palette for courts
court_palette <-
tibble(court = fct_inorder(factor(c("CJ", "GC", "CST"))),
clx_court = fct_inorder(factor(c("C","T","F"))),
court_long = fct_inorder(factor(c("Court of Justice", "General Court", "Civil Service Tribunal"))),
color = fct_inorder(factor(c("#fd014d", "#122771", "#3dcbb8"))))
# add labels
n_cases_year_court_labs <- n_cases_year_court |>
mutate(court_long = case_when(
court == "CJ" ~ "Court of Justice",
court == "GC" ~ "General Court",
court == "CST" ~ "Civil Service Tribunal",
T ~ NA_character_
)) |>
mutate(court_long = fct_relevel(court_long, "Court of Justice", "General Court"),
tooltip = str_c("N = ", n, " (", case_year, ")"),
data_id = str_c(court, case_year)) |>
left_join(court_palette)
# plot
iplot_cases_year_court <- n_cases_year_court_labs |>
group_by(court) |>
mutate(global_mean = mean(n)) |>
ggplot(aes(x = case_year, y = n, fill = color)) +
geom_hline(aes(yintercept = global_mean, colour = color), lty = 2, alpha = 0.33) +
geom_col_interactive(color = "grey96",
aes(tooltip = tooltip, data_id = data_id)) +
scale_fill_identity() +
scale_colour_identity() +
scale_x_continuous(breaks = seq(from = 1950, to = year_now+1, by = 10)) +
facet_wrap(~court_long, scales = "free_y", dir = "v") +
theme_minimal(base_family = "Arial") +
theme(panel.grid = element_blank(),
panel.grid.major.y = element_line(color = "grey94"),
panel.grid.major.x = element_line(color = "grey94"),
axis.text = element_text(color = "grey10"),
plot.background = element_rect(fill = "white", color = "grey88"),
title = element_text(face = "bold"),
plot.subtitle = element_text(face = "italic"),
plot.caption = element_text(face = "italic", size = 8),
strip.text = element_text(hjust = 1, face = "bold")) +
labs(x = NULL, y = NULL,
title = "Number of cases received by EU courts",
subtitle = "Aggregated by year and court",
caption = "Dashed line shows the mean")
# interactive
girafe(ggobj = iplot_cases_year_court,
fonts = list(sans = "Arial"),
#width_svg = 12,
#height_svg = 8,
options = list(opts_sizing(rescale = TRUE),
opts_toolbar(saveaspng = FALSE),
opts_tooltip(css = "background-color:gray;color:white;font-style:italic;padding:9px;border-radius:5px;font-size:15px;",
use_fill = TRUE),
opts_hover_inv(css = "opacity:0.1;"),
opts_hover(css = "fill:gray;stroke:black;"))
)We can see that the increase in the work of the CJ and the GC was mostly gradual. Interestingly, the oursourcing of the civil service agenda to the CST did not prevent the GC’s caseload from rising during its period of operation (the jurisdiction was subsequently returned to the GC).
As most other organizations, EU courts operate under a budget constraint while facing some uncertainty about their future workload. When allocating its resources and preparing annual budgets, the CJEU needs to be able to forecast how many cases it is likely to have to deal with in the upcoming years.
With information on the number of cases received in the past in hand, we can create a simple model that will attempt to predict the number of cases this year and next. In the presence of significant autocorrelation, using the number of cases received in previous years as predictors (lags) should give us decent forecasting mileage. To account for the fact that the CJ’s caseload is linked to the GC’s through appeal procedures, a regression equation can also include a lagged term for GC cases:
\[ Y_{t}^{CJ} = \beta_0 + \beta_1 Y_{t-1}^{CJ} + \beta_2 Y_{t-2}^{CJ} + \beta_3 Y_{t-3}^{CJ} + \beta_4 Y_{t-3}^{GC} + \epsilon_t \] Estimating the parameters in this equation via Gaussian or Poisson3 regression provides a simple way of making predictions about \(Y_t^{CJ}\).
| OLS | Poisson | |
|---|---|---|
| (Intercept) | 270.400** | 5.919*** |
| [85.212, 455.587] | [5.806, 6.031] | |
| lag1_CJ | −0.079 | 0.000 |
| [−0.550, 0.393] | [0.000, 0.000] | |
| lag2_CJ | 0.275 | 0.000*** |
| [−0.108, 0.657] | [0.000, 0.001] | |
| lag3_CJ | −0.064 | 0.000 |
| [−0.483, 0.355] | [0.000, 0.000] | |
| lag3_GC | 0.478** | 0.001*** |
| [0.192, 0.765] | [0.001, 0.001] | |
| Num.Obs. | 23 | 23 |
| R2 | 0.848 | |
| R2 Adj. | 0.815 | |
| AIC | 257.6 | 282.5 |
| BIC | 264.4 | 288.2 |
| Log.Lik. | −122.802 | −136.262 |
| RMSE | 50.42 | 49.68 |
| + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001 |
# design matrix, trim current year
dat_reg <- n_cases_year_court_labs |>
filter(case_year > 1994,
court != "CST") |>
mutate(lag1 = lag(n),
lag2 = lag(n, 2),
lag3 = lag(n, 3),
lag4 = lag(n, 4),
lag5 = lag(n, 5)) |>
filter(case_year > 2000) |>
select(court, case_year, n, starts_with("lag")) |>
pivot_wider(id_cols = case_year, names_from = court, values_from = c(n, starts_with("lag")))
# OLS estimates
out_ols <- lm(
data = dat_reg |> filter(case_year < year_now),
formula = n_CJ ~ lag1_CJ + lag2_CJ + lag3_CJ + lag3_GC
)
# Poisson estimates
out_poi <- glm(
data = dat_reg |> filter(case_year < year_now),
family = poisson(),
formula = n_CJ ~ lag1_CJ + lag2_CJ + lag3_CJ + lag3_GC
)
# OLS prediction
yhat_ols <- as.integer(round(predict(out_ols, dat_reg[nrow(dat_reg),], type = "response"), 0))
# Poisson prediction
yhat_poi <- as.integer(round(predict(out_poi, dat_reg[nrow(dat_reg),], type = "response"), 0))
# OLS prediction at t+1
newdat_t1 <- tibble(
case_year = year_now + 1,
lags = dat_reg[nrow(dat_reg),] |> select(starts_with("lag")) |> as.vector() |> unlist() |> lag(2),
cols = dat_reg[nrow(dat_reg),] |> select(starts_with("lag")) |> colnames()
) |>
pivot_wider(names_from = cols, values_from = lags, values_fill = 0L)
newdat_t1$lag1_CJ <- (yhat_ols + yhat_poi)/2 + sum(newdat_t1 |> select(contains("CJ")) |> unlist() |> diff(), na.rm = T)
yhat_ols_t1 <- as.integer(round(predict(out_ols, newdat_t1, type = "response"), 0))
# plot
iplot_cases_forecast <- n_cases_year_court_labs |>
mutate(tooltip = str_remove(tooltip, " \\(.*")) |>
filter(case_year > year_now - 7,
court == "CJ") |>
ggplot(aes(x = case_year, y = n)) +
geom_col(data = tibble(case_year = year_now, n = yhat_ols),
aes(color = court_palette$color[court_palette$court == "CJ"],
fill = court_palette$color[court_palette$court == "CJ"]),
alpha = 0.15,
lty = 2,
show.legend = FALSE) +
geom_col_interactive(data = tibble(case_year = year_now+1,
n = yhat_ols_t1,
tooltip = str_c("Predicted N = ", n),
data_id = case_year),
aes(color = court_palette$color[court_palette$court == "CJ"],
fill = court_palette$color[court_palette$court == "CJ"],
tooltip = tooltip,
data_id = data_id),
alpha = 0.15,
lty = 2,
show.legend = FALSE) +
geom_col_interactive(aes(fill = color, tooltip = tooltip, data_id = case_year)) +
scale_fill_identity() +
scale_x_continuous(breaks = (year_now-7):(year_now+1)) +
theme_minimal(base_family = "Arial") +
theme(panel.grid = element_blank(),
panel.grid.major.y = element_line(color = "grey94"),
#panel.grid.major.x = element_line(color = "grey94"),
axis.text = element_text(color = "grey10"),
axis.text.x = element_text(margin = margin(t = -6)),
plot.background = element_rect(fill = "white", color = "grey88"),
title = element_text(face = "bold"),
plot.subtitle = element_text(face = "italic"),
plot.caption = element_text(face = "italic", size = 8),
strip.text = element_text(hjust = 1, face = "bold")) +
labs(x = NULL, y = NULL,
title = "Number of cases received by the Court of Justice",
subtitle = "Forecast of this year and next based on previous years",
caption = "Predictions in dashed outlines")
# interactive
girafe(ggobj = iplot_cases_forecast,
fonts = list(sans = "Arial"),
#width_svg = 12,
#height_svg = 8,
options = list(opts_sizing(rescale = TRUE),
opts_toolbar(saveaspng = FALSE),
opts_tooltip(css = "background-color:gray;color:white;font-style:italic;padding:9px;border-radius:5px;font-size:15px;",
use_fill = TRUE),
opts_hover_inv(css = "opacity:0.2;"),
opts_hover(css = "fill:#fd014d;opacity:0.99;stroke:black;"))
)Both Gaussian and Poission regression yield comparable predictions
for 2024 (804 versus 804 cases).
The actual number of cases received so far this year represents
104 % of the predicted number.
In order to obtain a prediction for 2025 using our
model, we can plug in our prediction for 2024 as \(Y_{t-1}^{CJ}\). The result is shown in the
interactive plot. However, should our first prediction turn out way off
the mark, we will be compounding the error in our subsequent prediction.
Ours is an overly simplistic forecasting model – in the real-world, we
would hope to rely on other information than merely past values of \(Y_t\) to make predictions, as well as relax
some model assumptions.
The number of cases received by EU courts is not identical to the number of judicial decisions they produce. Some cases are withdrawn by the applicants or otherwise discontinued, others are joined and decided together.
The EU judicial system distinguishes between different types of public decisions. The two most common types are judgments and orders. Orders are typically employed to settle disputes more “economically” – they normally contain a shorter justification and address broadly speaking less important issues.
We can get an approximate overview of the number of decisions produced by EU courts using the Eur-Lex API.
# sector 6 court data from eurlex
allcourt <- elx_make_query(
"any",
sector = 6,
include_date = TRUE,
include_court_procedure = TRUE,
include_court_origin = TRUE,
include_original_language = TRUE,
include_court_formation = TRUE,
include_judge_rapporteur = TRUE,
include_ecli = TRUE
) |>
elx_run_query()
# bring data to celex level
decisions <- allcourt |>
mutate(clx_type = str_sub(celex, 6,7),
clx_num = as.integer(str_sub(celex, 8, 11)),
clx_year = as.integer(str_sub(celex, 2, 5)),
clx_court = str_sub(celex, 6, 6),
dec_type = case_when(
clx_type %in% c("CJ","TJ","FJ") ~ "Judgment",
clx_type %in% c("CO","CB","TO","TB","FO","FB") ~ "Order",
clx_type %in% c("CV","CX") ~ "Opinion",
T ~ NA_character_
)) |>
filter(!is.na(dec_type),
!str_detect(celex, "_INF|_SUM")) |>
separate_wider_delim(cols = courtprocedure,
delim = " - ",
too_few = "align_start",
too_many = "merge",
cols_remove = TRUE,
names = c("procedure", "procedure_outcome")) |>
mutate(procedure = ifelse(clx_type %in% c("FB","FJ","FO") & is.na(procedure), "Staff case", procedure)) |>
filter(!procedure %in% c("Rectification")) |>
group_by(celex) |> # CBs don't have ECLI
reframe(
decision = str_c(unique(dec_type), collapse = "~~~"),
ecli = str_c(unique(ecli), collapse = "~~~"),
date = str_c(unique(date), collapse = "~~~"),
procedure = str_c(unique(procedure), collapse = "~~~"),
procedure_outcome = str_c(unique(procedure_outcome), collapse = "~~~"),
rapporteur = str_c(unique(jr), collapse = "~~~"),
formation = str_c(unique(cf), collapse = "~~~"),
language = str_c(unique(origlang), collapse = "~~~"),
origin = str_c(unique(courtorigin), collapse = "~~~"),
) |>
ungroup() |>
mutate(across(everything(), ~str_squish(.)))
# where both CO and CB present, keep only CO
decisions <- decisions |>
mutate(clx_type = str_sub(celex, 6,7),
clx_num = as.integer(str_sub(celex, 8, 11)),
clx_year = as.integer(str_sub(celex, 2, 5)),
clx_court = str_sub(celex, 6, 6),
clx_dec = str_sub(celex, 7, 7)) |>
group_by(clx_court, clx_year, clx_num) |>
mutate(
dupl = any(clx_dec %in% "B") & any(clx_dec %in% "O")
) |>
ungroup() |>
filter(!(dupl == TRUE & clx_dec == "B")) |>
select(-dupl)
# title data to extract missing procedures
missing_procs_titles <- decisions |>
filter(is.na(procedure)) |>
mutate(title = map_chr(str_c("http://publications.europa.eu/resource/celex/",
celex),
possibly(eurlex::elx_fetch_data, otherwise = NA_character_), "title"))
# parse the titles
missing_procs_titles <- missing_procs_titles |>
mutate(title = str_squish(title)) |>
mutate(
procedure = case_when(
str_detect(title, "Rectification") ~ "Rectification",
str_detect(title, "Opinion of the Court|Request for an Opinion") ~ "Request for an Opinion",
str_detect(title, "DEP ") ~ "Other",
str_detect(celex, "FB") | str_detect(title, "[Ss]taff|[Cc]ivil [Ss]ervice") ~ "Staff case",
str_detect(title, "reliminary referen|for a prelim|reliminary ruli") ~ "Preliminary reference",
str_detect(title, "Commission( of the European Communities)? v (?!(Parliament|Council|European))") ~ "Failure to fulfil obligations",
str_detect(celex, "FB") | str_detect(title, "[Aa]nnulment|Trademark|Trade Mark") ~ "Action for annulment",
str_detect(celex, "FB") | str_detect(title, "damages") ~ "Action for damages",
str_detect(title, " (v|contre) (European Commission|Commission|Council|European Parliament|Parliament|EUIPO|OHIM|OHMI|ECB|Office for Har|European Union|Office de l'|EASO|EMA)") ~ "Action for annulment", # this is too broad but probably OK as approximation
T ~ NA_character_
),
origin = case_when(
str_detect(title, "Czech Republic|Nejvyšš|Czechia") ~ "Czechia",
str_detect(title, "Slovak Republic|Najvyšš|Prešov|Slovakia") ~ "Slovakia",
str_detect(title, "Polish|Poland") ~ "Poland",
str_detect(title, "Belgiqu|Hof van Cass|Brussel|Belgium") ~ "Belgium",
str_detect(title, "Finland") ~ "Finland",
str_detect(title, "Salzburg|Wien|Austria") ~ "Austria",
str_detect(title, "Veliko Tarn|Varna|Bulgaria|Sofiys") ~ "Bulgaria",
str_detect(title, "Juzgado|Audiencia Provincial|Tribunal Superior de Justicia|Spain") ~ "Spain",
str_detect(title, "România|Romania") ~ "Romania",
str_detect(title, "Rijeci|Croati|lučice") ~ "Croatia",
str_detect(title, "Slovenia") ~ "Slovenia",
str_detect(title, "Cyprus") ~ "Cyprus",
str_detect(title, "Estonia") ~ "Estonia",
str_detect(title, "Sweden") ~ "Sweden",
str_detect(title, "Luxembourg") ~ "Luxembourg",
str_detect(title, "Denmark|Dane?mark") ~ "Denmark",
str_detect(title, "Hellenic|Greece|Simvoulio|Simboulio") ~ "Greece",
str_detect(title, "Portugese|Tribunal da Relação|(?<!TAP )Portugal") ~ "Portugal",
str_detect(title, "Lietuv|Vilniaus|Lithuania") ~ "Lithuania",
str_detect(title, "Latvijas|Augstākās|Latvia") ~ "Latvia",
str_detect(title, "Italia|Italy|Tribunale Amministrativ|Tribunale di|Consiglio di|Corte Su") ~ "Italy",
str_detect(title, "Bíróság|Magyar|Budapest|Szeged|Hungary|Hongri") ~ "Hungary",
str_detect(title, "Conseil d|(?<!Air )France") ~ "France",
str_detect(title, "Irland|(?<!Northern )Ireland") ~ "Ireland",
str_detect(title, "Netherlands|Den Haag|Raad van Ber|College van|Rechtbank Am|Raad [Vv]an St|Hoge Raad") ~ "Netherlands",
str_detect(title, "England \\& Wales|England and Wales|Her Majest|United Kingdom") ~ "United Kingdom",
str_detect(title, "Deutschland|Regensburg|Germany|Landgericht K|Amtsgericht") ~ "Germany",
T ~ NA_character_
)
)
# simplify procedures
decisions <- decisions |>
filter(!celex %in% missing_procs_titles$celex) |>
bind_rows(missing_procs_titles |> select(-title)) |>
filter(!procedure %in% c("Rectification")) |>
mutate(procedure_class = case_when(
str_detect(procedure, "eference.*preliminary|Preliminary reference") ~ "Preliminary rulings",
str_detect(procedure, "Action for annulm") ~ "Annulment procedures",
str_detect(procedure, "[Ff]ailure to fulf") ~ "Infringement proceedings",
str_detect(procedure, "[Ss]taff") ~ "Staff cases",
T ~ "Other"
)) |>
mutate(court_long = case_when(
clx_court == "C" ~ "Court of Justice",
clx_court == "T" ~ "General Court",
clx_court == "F" ~ "Civil Service Tribunal",
T ~ NA_character_
)) |>
mutate(court_long = fct_relevel(court_long, "Court of Justice", "General Court")) |>
mutate(formation_simple = case_when(
str_detect(formation, "Grand Chamber") ~ "Grand Chamber",
str_detect(formation, "[Ff]ull [Cc]ourt") ~ "Full Court",
str_detect(formation, "Single|Vice-President|President|Judge hearing") ~ "Single Judge",
T ~ "Panel"
)) |>
mutate(formation_simple = fct_relevel(formation_simple, "Full Court", "Grand Chamber", "Panel", "Single Judge")) |>
mutate(decision_year = str_sub(date, 1, 4)) |>
mutate(rapporteur = case_when( # correct some rapporteurs
str_detect(rapporteur, "O.Higgins") ~ "OHiggins",
str_detect(rapporteur, "O'Keeffe") ~ "OKeeffe",
str_detect(rapporteur, "Cz.cz") ~ "Czucz",
str_detect(rapporteur, ".an .er Woude") ~ "van der Woude",
T ~ rapporteur
))
# some manual corrections
decisions <- decisions |>
filter(!celex %in% c("62019CO0519")) |> # corrigendum
mutate(
origin = case_when(
celex == "62019CJ0550" ~ "Spain",
celex == "62020CJ0339" ~ "France",
celex == "62014CB0196" ~ "Germany",
celex == "62022CB0578" ~ "Germany",
celex == "62023CB0052" ~ "Germany",
T ~ origin
),
language = case_when(
celex == "62017CO0707" ~ "Bulgarian",
celex == "62020CJ0306" ~ "Latvian",
celex == "62017CJ0122" ~ "English",
celex == "62018CJ0276" ~ "Hungarian",
celex == "62013CJ0512" ~ "Dutch",
celex == "62022CJ0142" ~ "English",
celex == "61987CJ0245" ~ "German",
celex == "62000CJ0167" ~ "German",
celex == "62012CJ0293" ~ "English~~~German",
T ~ language
)
)
# appeals and outcomes# data frame for plotting
n_dec_type_proc <- decisions |>
filter(decision != "Opinion",
procedure_class != "Other",
clx_court == "C") |>
count(decision_year, decision, procedure_class, court_long) |>
drop_na()
# make plot
iplot_dec_procs <- n_dec_type_proc |>
left_join(court_palette) |>
mutate(tooltip = str_c(decision, "s = ", n, " (", decision_year, ")"),
data_id = str_c(clx_court, decision_year, procedure_class, decision)) |>
ggplot(aes(x = as.integer(decision_year), y = n, fill = decision)) +
geom_col_interactive(color = "grey96", show.legend = FALSE,
aes(tooltip = tooltip, data_id = data_id)) +
#scale_fill_identity() +
#scale_colour_identity() +
scale_fill_manual_interactive(values = c("#fd014d","#eda1b8")) +
scale_x_continuous(breaks = seq(from = 1950, to = year_now+1, by = 10)) +
facet_wrap(~fct_infreq(procedure_class), dir = "v", scales = "free_y") +
theme_minimal(base_family = "Arial") +
theme(panel.grid = element_blank(),
panel.grid.major.y = element_line(color = "grey94"),
panel.grid.major.x = element_line(color = "grey94"),
axis.text = element_text(color = "grey10"),
plot.background = element_rect(fill = "white", color = "grey88"),
title = element_text(face = "bold"),
plot.subtitle = element_text(face = "italic"),
plot.caption = element_text(face = "italic", size = 8),
strip.text = element_text(hjust = 1, face = "bold")) +
labs(x = NULL, y = NULL,
title = "Number of decisions handed down by the Court of Justice",
subtitle = "Aggregated by year and procedure")
# interactive
girafe(ggobj = iplot_dec_procs,
fonts = list(sans = "Arial"),
#width_svg = 12,
#height_svg = 8,
options = list(opts_sizing(rescale = TRUE),
opts_toolbar(saveaspng = FALSE),
opts_tooltip(css = "background-color:gray;color:white;font-style:italic;padding:9px;border-radius:5px;font-size:15px;",
use_fill = TRUE),
opts_hover_inv(css = "opacity:0.2;"),
opts_hover(css = "opacity:0.99;stroke:black;")))The Court of Justice sees the highest diversity of procedures, even though the General Court is set to adjudicate a greater variety of cases in the future in response to workload pressures and past court reforms.
The four main types of procedures are the preliminary ruling procedure (Article 267 TFEU), the annulment procedure (Article 263 TFEU), the infringement procedure (Article 258 TFEU) and staff cases. We can see that the CJ has been handing down a seemingly ever-increasing number of preliminary rulings on questions referred by national courts. As the CJ’s workload expanded, the Court increased the frequency with which it uses orders rather than judgments.
Decisions can be further classified by the deciding court formation. The most important cases at the CJ are decided by more than ten judges in a formation known as the “Grand Chamber”. On special occasions, the CJ even sits as the “Full Court” (all judges decide). Most cases are decided by a panel (chamber) of five judges.
Each decision is assigned a judge-rapporteur. This judge has the primary responsibility for drafting the decision. The main predictor of how many decisions a judge writes is the length of their tenure at the court. In principle, judges can be reappointed to EU courts any number of times.
Historically the most important subset of CJEU cases is generated through the preliminary ruling procedure (Article 267 TFEU). The most well-known CJ decisions – such as Costa v ENEL and Cassis de Dijon – originate in references from national courts to Luxembourg.
# from decision level to reference level
references <- decisions |>
filter(str_detect(procedure_class ,"Preliminary")) |>
separate_rows(origin, sep = "~~~") |>
filter(!is.na(origin))
# could add validation of origin and language here (done as of 5/10/23)
# references |>
# mutate(origin_lang_match = case_when(
# origin == "Czechia" & str_detect(language, "Czec") ~ T,
# origin == "Slovakia" & str_detect(language, "Slovak") ~ T,
# origin == "Italy" & str_detect(language, "Ital|German") ~ T,
# origin == "Poland" & str_detect(language, "Polish") ~ T,
# origin == "France" & str_detect(language, "French") ~ T,
# origin == "Belgium" & str_detect(language, "French|Dutch|German") ~ T,
# origin == "Ireland" & str_detect(language, "English|Gael|Irish") ~ T,
# origin == "Austria" & str_detect(language, "German|Sloven") ~ T,
# origin == "Bulgaria" & str_detect(language, "Bulgar") ~ T,
# origin == "Spain" & str_detect(language, "Spani") ~ T,
# origin == "Romania" & str_detect(language, "Romani") ~ T,
# origin == "Croatia" & str_detect(language, "Croati") ~ T,
# origin == "Greece" & str_detect(language, "Greek") ~ T,
# origin == "Cyprus" & str_detect(language, "Greek") ~ T,
# origin == "Estonia" & str_detect(language, "Estonian") ~ T,
# origin == "Finland" & str_detect(language, "Finnish|Swedish") ~ T,
# origin == "Denmark" & str_detect(language, "Danish") ~ T,
# origin == "Sweden" & str_detect(language, "Swedish") ~ T,
# origin == "Luxembourg" & str_detect(language, "French") ~ T,
# origin == "Malta" & str_detect(language, "Maltese|English") ~ T,
# origin == "Slovenia" & str_detect(language, "Sloven") ~ T,
# origin == "Portugal" & str_detect(language, "Portug") ~ T,
# origin == "Lithuania" & str_detect(language, "Lithuan") ~ T,
# origin == "Latvia" & str_detect(language, "Latvia") ~ T,
# origin == "Hungary" & str_detect(language, "Hungar") ~ T,
# origin == "Netherlands" & str_detect(language, "Dutch") ~ T,
# origin == "United Kingdom" & str_detect(language, "English") ~ T,
# origin == "Germany" & str_detect(language, "German") ~ T,
# T ~ FALSE
# )) |>
# #filter(origin == "Austria", language == "English") |> View()
# count(origin_lang_match, origin, language) |> View()
# country ref number
n_refs_country <- references |>
arrange(date) |>
group_by(origin) |>
summarize(
year_first = clx_year[1],
n_refs = n()
) |>
mutate(ttip = str_c(origin,": ", n_refs," references. <br>", "First case referred in ", year_first, "."))
# prepare map data
refs_map_dat <- map_data("world",xlim=c(-10.6600,51.5500), ylim=c(34.5000,71.0500)) |>
mutate(region = ifelse(region == "Czech Republic", "Czechia", region),
region = ifelse(region == "UK", "United Kingdom", region))
#map("world",xlim=c(-10.6600,31.5500), ylim=c(34.5000,71.0500))
# map plot
iplot_map_refs <- n_refs_country |>
ggplot() +
geom_map_interactive(
aes(map_id = origin,
fill = n_refs,
tooltip = ttip,
data_id = origin),
map = refs_map_dat,
show.legend = FALSE,
color = "grey28",
size = 0.2
) +
#annotate("text", label = "The UK was an EU member between 1973 and 2020.", size = 2.6,
# x = -1.8, y = 65) +
expand_limits(x = c(-13.6600,38.5500), y = c(34.5000,71.0500)) +
scale_fill_gradient_interactive(low = "#8993b8", high = "#fe6794") +
theme_minimal(base_family = "Arial") +
theme(panel.grid = element_blank(),
axis.text = element_blank(),
plot.background = element_rect(fill = "white", color = "grey88"),
title = element_text(face = "bold"),
plot.subtitle = element_text(face = "italic"),
plot.caption = element_text(face = "italic", size = 8),
strip.text = element_text(hjust = 1, face = "bold")) +
labs(x = NULL, y = NULL,
title = "Preliminary references by country of origin",
subtitle = "Total number of closed cases referred by all national courts")
# interactive
girafe(ggobj = iplot_map_refs,
fonts = list(sans = "Arial"),
#width_svg = 12,
#height_svg = 8,
options = list(opts_sizing(rescale = TRUE),
opts_toolbar(saveaspng = FALSE),
opts_tooltip(css = "background-color:gray;color:white;font-style:italic;padding:9px;border-radius:5px;font-size:15px;",
use_fill = TRUE),
opts_hover_inv(css = "opacity:0.2;"),
opts_hover(css = "opacity:0.99;stroke:gray;")))From a geographical perspective, preliminary references from national courts are very unevenly distributed. By far the highest number of references has come from Germany. Naturally, there is a relationship between the length of a country’s membership and the total number of preliminary references. Research has shown that there are also important subnational disparities.
In general, referral activity increased over time in most countries. The CJEU and its preliminary ruling procedure gradually gained prominence. Interestingly, the gradual increase in references is also visible for countries that joined the EU more recently when the importance of the CJEU was already well-established. Despite the overall trend, there are also notable differences in the temporal patterns between countries (compare Ireland and Denmark, for example).4 Note that the data comes from closed cases – many referrals in recent years are still pending before the CJEU.
Cite this document as Michal Ovádek, ‘Judicial Proceedings in the European Union’, available at https://michalovadek.github.io/eucourt/, accessed on 11 December 2024.
For more details see the website. More example use cases, as well as arguments for integrating open data APIs into research workflows, can be found in this open-access paper.↩︎
The IUROPA project is attempting to create more comprehensive datasets.↩︎
Because we are working with count data, Poisson regression should be in principle more appropriate.↩︎
However, because the data used here is not fully accurate, we should not make too much of any individual data point.↩︎