Working with manystates data

Have democracies become more free and accountable since the 1990s?

manystates provides data on states in the international system across time, assembling together datasets from reputable sources including The Correlates of War project, The Issue Correlates of War project, and Gleditsch and Ward’s dataset on interstate system membership. The package can also be used to study related features such as political regimes (regimes database), contiguity (contiguity database), leadership (leaders database), and economic indicators (economics database).

To illustrate the uses of the data, we explore here the question of whether states have indeed become more free and accountable. Since 2017, more than half of the states in the world have become democracies. However, at the same time, influential works have theorized about the recent “democratic decay” and “stretch”; especially with the recent(ish) elections of leaders with authoritarian tendencies in several democracies.

Below, we use manystates to study democratic accountability using data from the regimes database as well as the import_vdem() function to extract Variaties of Democracy Data (VDEM) data. The regimes database contains data on political regimes in the Polity5 and Freedom House datasets.

Data

For many years, Freedom House data has been widely relied upon, including in the media, as an indicator of states that are democratic. VDEM data has been developed more recently by scholars for a detailed and comprehensive understanding of democracy. In this vignette, we will focus on Freedom House and VDEM data which include electoral, judicial, and civil dimensions. Yet, Freedom House and VDEM projects also have slightly different approaches to the measurement of democracy, which we will explore below. Let’s import the data first. Due to the size of VDEM data, we rely on the import_vdem() function to download the data, contrary to the Freedom House dataset which is stored internally.

library(manystates)
fh <- manystates::regimes$FreedomHouseFull
vdem <- import_vdem()

Let’s examine first the data in VDEM by selecting a few variables that are helpful to investigate freedom and accountability in democracies since the 1990s.

First, we need to know if a state is democratic and to what extent. The “v2x_regime” variable in VDEM differentiates between two types of democratic regimes (electoral and liberal) :

Electoral democracy: free and fair multiparty elections…
Liberal democracy: free and fair multiparty elections, access to justice, transparent law enforcement, respect for personal liberties, rule of law…

Second, we need indicators of democracy. Let’s focus on transparency and accountability (“v2x_accountability_osp”), freedom of expression (“v2x_freexp_altinf”), and civil society participation (v2x_cspart).

These variables were transformed into scales (e.g. 0-1) from ordinal data from the surveys (ordered categorical) as follows:

Accountability index: low to high (0-1)
Freedom of expression: low to high (0-1)
Civil society participation: low to high (0-1)

library(dplyr)
library(tidyr)
vdem <- vdem %>% dplyr::select(StateName, stateID, Year,
                              v2x_regime, v2x_accountability_osp,
                              v2x_freexp_altinf, v2x_cspart) %>%
  # select only the variables of interest
  dplyr::filter(Year > 1989, v2x_regime == 2 | v2x_regime == 3) %>%
  # filter for last 30 years (bigger than 1989) and for democracies only
  dplyr::rename(regime = v2x_regime,
                freedom_expression = v2x_freexp_altinf,
                accountability = v2x_accountability_osp,
                civil_society = v2x_cspart) %>% # rename some of these variables
  dplyr::mutate(regime_type = case_when(regime == 2 ~ "Electoral Democracy",
                                        regime == 3 ~ "Liberal Democracy")) %>% 
  # create a new regime type variable and rename the categories within
  arrange(desc(Year)) %>% # arrange by year
  drop_na() # drop rows where there is at least one NA obs...

The Freedom House data contains state year data starting from the year 2012. Therefore, it is much smaller then VDEM. Freedom House also organises its data somewhat differently, disaggregating the concept of democracy into two components: political rights (“PR rating” - “A” + “B” + “C”) and civil liberties (“CL rating” - “D” + “E” + “F” + “G”). The FreedomHouseFull dataset also contains the raw scores of each state for each question, which are then grouped into categories labelled with alphabets from “A” to “G”. In terms of indicators of democracy, accountability is captured in category “C”, freedom of expression in category “D”, and civil society participation in category “B”. Let’s subset the data accordingly.

fh <- dplyr::select(fh, c(StateName, stateID, Year, PR, CL, B, C, D)) %>%
  # select only the variables of interest
  dplyr::filter(PR > 17) %>% # filter for democracies only
  manydata::transmutate(freedom_expression = D/16,
                        accountability = C/12,
                        civil_society = B/16) %>%
  # rename and standardise some of these variables
  dplyr::mutate(regime_type = ifelse(CL > 52, "Liberal Democracy",
                                     "Electoral Democracy")) %>%
  # create a new regime type variable and rename the categories within,
  # benchmark score derived from Scores table in Freedom House Codebook
  arrange(desc(Year)) %>% # arrange by year
  drop_na() # drop rows where there is at least one NA obs...

Investigate the data

Let’s first compare the 10 states with the highest accountability scores (on average) in each dataset since 2012.

library(kableExtra)

vdem_account_10 <- vdem %>% # call data without assigning it
  dplyr::filter(Year > 2011) %>% # filter for years in both datasets
  group_by(StateName) %>% # group by state name
  summarise(Mean_VDEM = mean(accountability)) %>%
  arrange(desc(Mean_VDEM)) %>% # arrange the data by mean, in decreasing order
  slice(1:10)

fh_account_10 <- fh %>% # call data without assigning it
  group_by(StateName) %>% # group by state name
  summarise(Mean_FH = mean(accountability)) %>%
  # summarize grouped accountability variable by mean
  arrange(desc(Mean_FH)) %>%
  slice(1:10) # arrange the data by mean, in decreasing order

full_join(vdem_account_10, fh_account_10) %>% 
  kbl() %>% 
  kable_styling("striped")

StateName	Mean_VDEM	Mean_FH
Denmark	0.9794545	1.0000000
Sweden	0.9743636	1.0000000
Germany	0.9712727	1.0000000
Norway	0.9711818	1.0000000
Costa Rica	0.9686364	NA
Switzerland	0.9683636	1.0000000
Estonia	0.9672727	NA
Australia	0.9663636	1.0000000
Finland	0.9658182	1.0000000
New Zealand	0.9627273	1.0000000
Netherlands	NA	1.0000000
Andorra	NA	0.9907407

It seems that the datasets differ somewhat on the states with highest accountability scores (on average).

Pathways: from electoral to liberal democracies (and vice-versa)

Let’s compare which states appear as having multiple democratic regimes since 2012 in both datasets.

vdem_transition <- vdem %>% # saving this object to use later
  dplyr::filter(Year > 2011) %>% # filter for years in both datasets
  group_by(StateName, regime_type) %>% # group by state and regime
  count() %>% # count grouped observations
  group_by(StateName) %>% # re-group by state name only 
  mutate(Duplicated = n()) %>% # add duplicated columns for state names
  filter(Duplicated > 1) %>%  # filter for names that appear more than once
  select(-c(Duplicated)) %>% # remove duplicated variables
  rename('Years VDEM' = n, 'Regime Type' = regime_type) # rename for clarity

fh_transition <- fh %>% # saving this object to use later
  group_by(StateName, regime_type) %>% # group by state and regime
  count() %>% # count grouped observations
  group_by(StateName) %>% # re-group by state name only 
  mutate(Duplicated = n()) %>% # add duplicated columns for state names
  filter(Duplicated > 1) %>%  # filter for names that appear more than once
  select(-c(Duplicated)) %>% # remove duplicated variables
  rename('Years FH' = n, 'Regime Type' = regime_type) # rename for clarity

dplyr::full_join(vdem_transition, fh_transition) %>%
  arrange(StateName) %>%
  kbl() %>% 
  kable_styling("striped")

StateName	Regime Type	Years VDEM	Years FH
Austria	Electoral Democracy	2	NA
Austria	Liberal Democracy	9	NA
Barbados	Electoral Democracy	3	NA
Barbados	Liberal Democracy	8	NA
Belize	Electoral Democracy	NA	8
Belize	Liberal Democracy	NA	1
Benin	Electoral Democracy	5	NA
Benin	Liberal Democracy	2	NA
Bhutan	Electoral Democracy	10	NA
Bhutan	Liberal Democracy	1	NA
Botswana	Electoral Democracy	2	NA
Botswana	Liberal Democracy	9	NA
Canada	Electoral Democracy	2	NA
Canada	Liberal Democracy	9	NA
Chile	Electoral Democracy	3	1
Chile	Liberal Democracy	8	8
Czechia	Electoral Democracy	2	NA
Czechia	Liberal Democracy	9	NA
France	Electoral Democracy	NA	5
France	Liberal Democracy	NA	4
Ghana	Electoral Democracy	4	NA
Ghana	Liberal Democracy	7	NA
Greece	Electoral Democracy	3	NA
Greece	Liberal Democracy	8	NA
Jamaica	Electoral Democracy	10	NA
Jamaica	Liberal Democracy	1	NA
Japan	Electoral Democracy	NA	1
Japan	Liberal Democracy	NA	8
Latvia	Electoral Democracy	2	NA
Latvia	Liberal Democracy	9	NA
Lithuania	Electoral Democracy	7	1
Lithuania	Liberal Democracy	4	8
Mauritius	Electoral Democracy	9	NA
Mauritius	Liberal Democracy	2	NA
Nauru	Electoral Democracy	NA	7
Nauru	Liberal Democracy	NA	2
Poland	Electoral Democracy	7	5
Poland	Liberal Democracy	4	4
Portugal	Electoral Democracy	2	NA
Portugal	Liberal Democracy	9	NA
Serbia	Electoral Democracy	1	NA
Serbia	Liberal Democracy	1	NA
Seychelles	Electoral Democracy	2	NA
Seychelles	Liberal Democracy	8	NA
Slovakia	Electoral Democracy	NA	2
Slovakia	Liberal Democracy	NA	7
Slovenia	Electoral Democracy	3	NA
Slovenia	Liberal Democracy	8	NA
South Africa	Electoral Democracy	10	NA
South Africa	Liberal Democracy	1	NA
Taiwan	Electoral Democracy	NA	4
Taiwan	Liberal Democracy	NA	5
Trinidad and Tobago	Electoral Democracy	2	NA
Trinidad and Tobago	Liberal Democracy	9	NA
United States	Electoral Democracy	NA	1
United States	Liberal Democracy	NA	8

VDEM and freedom house appear to differ quite a bit in the cases that transitioned from liberal democracies to electoral democracies (and vice-versa) since 2012. Let’s investigate further by taking a closer look at cases that appear in both datasets.

transition_vdem <- vdem %>% # saving this object to use later
  dplyr::filter(Year > 2011) %>% # filter for years in both datasets
  group_by(StateName, regime_type) %>% # group by state and regime
  summarise(Years_VDEM = paste(Year, collapse = ", ")) %>%
  group_by(StateName) %>% # re-group by state name only 
  mutate(Duplicated = n()) %>% # add duplicated columns for state names
  filter(Duplicated > 1) %>%  # filter for names that appear more than once
  select(-c(Duplicated)) # remove duplicated variables

transition_fh <- fh %>% # saving this object to use later
  group_by(StateName, regime_type) %>% # group by state and regime
  summarise(Years_FH = paste(Year, collapse = ", ")) %>%
  group_by(StateName) %>% # re-group by state name only 
  mutate(Duplicated = n()) %>% # add duplicated columns for state names
  filter(Duplicated > 1) %>%  # filter for names that appear more than once
  select(-c(Duplicated)) # remove duplicated variables

inner_join(transition_vdem, transition_fh) %>% 
  kbl() %>%
  kable_styling("striped")

StateName	regime_type	Years_VDEM	Years_FH
Chile	Electoral Democracy	2021, 2020, 2019	2019
Chile	Liberal Democracy	2022, 2018, 2017, 2016, 2015, 2014, 2013, 2012	2020, 2018, 2017, 2016, 2015, 2014, 2013, 2012
Lithuania	Electoral Democracy	2022, 2021, 2020, 2019, 2018, 2017, 2016	2020
Lithuania	Liberal Democracy	2015, 2014, 2013, 2012	2019, 2018, 2017, 2016, 2015, 2014, 2013, 2012
Poland	Electoral Democracy	2022, 2021, 2020, 2019, 2018, 2017, 2016	2020, 2019, 2018, 2017, 2016
Poland	Liberal Democracy	2015, 2014, 2013, 2012	2015, 2014, 2013, 2012

Even in the few cases of transtions from liberal democracies into electoral democracies (and vice-versa) since 2012 appear in both datasets, they seem to somewhat disagree in the specific years where these states are classified as liberal and/or electoral democracies.

Since VDEM has more state year observations and we are interested in democracy pathways since the 1990s, let’s work with VDEM data only from now onwards.

Let’s just look at the average differences for our democracy indicators for the states with multiple regimes.

library(ggplot2)

vdem %>%
  group_by(StateName, regime_type) %>% # group by state and regime
  summarise(Accountability = mean(accountability),
            Freedom_of_Expression = mean(freedom_expression),
            Civil_Society = mean(civil_society)) %>%
  # summarize various grouped variable by mean
  group_by(StateName) %>% # re-group by state name only 
  mutate(Duplicated = n()) %>% # add duplicated columns for state names
  filter(Duplicated > 1) %>% # filter for names that appear more than once
  pivot_wider(names_from = regime_type,
              values_from = Accountability:Civil_Society,
              names_sep = "-") %>% # make data wide
  mutate(Accountability_dif = `Accountability-Liberal Democracy` -
           `Accountability-Electoral Democracy`,
         Freedom_of_Expression_dif = `Freedom_of_Expression-Liberal Democracy` -
           `Freedom_of_Expression-Electoral Democracy`,
         Civil_Society_dif = `Civil_Society-Liberal Democracy` -
           `Civil_Society-Electoral Democracy`) %>% 
  # create variables with mean difference
  select(StateName, Accountability_dif, Freedom_of_Expression_dif,
         Civil_Society_dif) %>% # select only pertinent observations
  pivot_longer(cols = Accountability_dif:Civil_Society_dif) %>% 
  # pivot table longer again
  rename(State = StateName,
         Indicator_Difference = name,
         Mean_Difference = value) %>% # rename for clarity
  ggplot(aes(x = Mean_Difference, y = State,
             fill = Indicator_Difference)) + # plot
  geom_point(aes(color = Indicator_Difference)) +
  facet_wrap("Indicator_Difference") +
  theme_minimal() +
  theme(legend.position = "none") +
  labs(title = "From Electoral to Liberal Democracies",
       y = "State",
       x = "Mean Difference")

Although the plot is not very informative, it appears that some states averages for these indicators were worse when they were classified as liberal democracies in comparison to when they were electoral democracies. This is puzzling.

Let’s try a different approach, plotting averages for all indicators together in time and by state.

vdem %>% # saving this object to use later
  group_by(StateName, regime_type) %>% # group by state and regime
  count() %>% # count grouped observations
  group_by(StateName) %>% # re-group by state name only 
  mutate(Duplicated = n()) %>% # add duplicated columns for state names
  filter(Duplicated > 1) %>%  # filter for names that appear more than once
  select(-c(Duplicated)) %>% # remove duplicated variables
  select(-n) %>% # remove years
  inner_join(vdem, by = c("StateName", "regime_type"), multiple = "all") %>%
  # keep only obs in present in both datasets
  mutate(Aggregated_Mean_Indicators = (accountability +
           freedom_expression + civil_society)/3) %>%
  # add rows to form the aggregated indicator
  select(StateName, regime_type, Year, Aggregated_Mean_Indicators) %>%
  # select only a few variables
  rename(State = StateName, Regime = regime_type) %>%
  ggplot(aes(x = as.numeric(Year), y = Aggregated_Mean_Indicators,
             colour = (Regime == "Liberal Democracy"))) + 
  geom_line(aes(group=1)) +
  facet_wrap("State") +
  theme_minimal() +
  theme(legend.position = "bottom") +
  labs(title = "From Electoral to Liberal Democracies (and vice-versa)",
       y = "Aggregated Mean Indicators") +
  scale_color_discrete(name="Regime",
                       labels=c("Electoral Democracy",
                                "Liberal Democracy"))

Once again, it appears that for some states, aggregated indicators for accountability, freedom of expression, and civil society participation got better as they moved from electoral democracy to liberal democracy (e.g. Taiwan), or worse as they moved from liberal democracy to electoral democracy (e.g. Hungary). In other cases, it is not clear these changed at all (e.g. Benin) or, even more puzzling, aggregated indicators got worse as they moved from electoral democracy to liberal democracy (e.g. Greece), or better as they moved from liberal democracy to electoral democracy (e.g. Slovakia).

Indicators by region

Perhaps, however, we are missing important regional information that mediates the relationship between these indicators and whether states are classified liberal or electoral democracies. How about we check the evolution of these indicators by region? Let’s get and join region data from the HUGGO_STATES database to see the evolution of the aggregated indicators per region since the 1990s.

HUGGO_STATES <- states$HUGGO_STATES %>%
  select(StateName, Region) %>% # select only the variables we want
  distinct() # keeps only distinct rolls (no duplicates)

vdem %>%
  dplyr::left_join(HUGGO_STATES, by = "StateName") %>%
  group_by(Region) %>%
  mutate(max_region_year = max(as.numeric(Year)),
         min_region_year = min(as.numeric(Year))) %>% # get min and max values for each regions
  filter(Year == min_region_year | Year == max_region_year) %>%
  # filter for first and last year of data
  group_by(Region, Year) %>% # group region and year
  summarise(Aggregated_Mean_Indicators = (mean(accountability) +
                                            mean(freedom_expression) +
                                            mean(civil_society))/3) %>% 
  # add aggregated means by groups
  drop_na() %>%
  ggplot(aes(x = Aggregated_Mean_Indicators,
      y = reorder(Region, Aggregated_Mean_Indicators),
      Fill = Region)) +
  geom_point(aes(color = Region)) +
  geom_line(aes(color = Region)) +
  geom_text(aes(label = Year), nudge_y = 0.3) +
  theme_minimal() +
  theme(legend.position="none") +
  labs(title = "Are democracies better off in 2021 than in 1990?",
       subtitle = "Difference aggregated means for indicators of
       accountability, freedom of expression, and civil society indicators",
       y = "Region",
       x = "Mean Difference")

Perhaps a better question is where have democracies become more free and accountable since the 1990s.

Henrique Sposito and Jael Tan

2023-05-26

Have democracies become more free and accountable since the 1990s?

Data

Investigate the data

Pathways: from electoral to liberal democracies (and vice-versa)

Indicators by region