In producing the GxR, ARTICLE 19 selected the 25 indicators from the Varieties of Democracy (V-Dem) Dataset that best matched our broad and holistic view of freedom of expression. The V-Dem Dataset is described below, and the indicators are described in this document. 

We included these 25 indicators in a Bayesian measurement model for countries with available data between 2000 and 2022 to create the GxR metric, which underpins our report.

Our final data file contains the 161 countries (after combining Gaza and West Bank to report results for Palestine) for which at least one year of data is available between 2000 and 2022. 

Throughout our reporting, we refer to the results by ‘country’. However, we acknowledge that some of these are special regions that are part of another country in our data (e.g. Hong Kong). We acknowledge that there are various ways to group countries into regions and to analyse the data, and we continue to evolve our methodology each year. 

ARTICLE 19 Mexico and Central America has its own methodology for tracking freedom of expression in Mexico. To that end, Mexico is not included in GxR rankings or in any country-level analyses using the metric.

Scores, expression categories, and rankings

‍For each country, we calculate an Expression Score from the 25 variables, which is produced as a point estimate from the Bayesian measurement models. This initial score falls between 0 and 1, and we re-scale and round the value to report GxR as an integer (0–100). 

We calculate scores for both Global and Regional Expression by taking the average of the country-level scores described above. 

We also calculate the Human Score at both global and regional levels, taking into account the size of the population that is experiencing the expression environment. To calculate the Human Score, we adjust the Global and Regional Expression Scores by a population weight to ensure that each country is represented in proportion to its population size within those measures.

Countries are placed in their respective expression categories based on these final integers. 

We order the scores from highest to lowest at both global and regional levels, creating a list of global and regional rankings, to ascertain where a country sits in relation to other countries. 

Scores and rankings are available for each year between 2000 and 2022.

For our analyses, we pulled population data from the World Bank database. The total population counts represent all residents – regardless of citizenship – and the annual values are mid-year estimates (variable: SP.POP.TOTL). Neither Taiwan nor Palestine are represented in the World Bank data. The 2022 global population for the countries included in our GxR data is 7.9 billion. 

We calculated Palestine’s population using population weights based on data from Palestine’s 2007 census and the US Central Intelligence Agency’s 2019 and July 2021 estimates for Gaza and West Bank. We used the 2007 population for 2007–2010, taking the average of the 2007 population and the 2019 mid-year estimate for 2011–2015, the 2019 estimate for 2016–2019, and the July 2021 estimate for 2020–2022.

We identify countries that have seen significant changes (declines/advances) in their score based on movement outside the upper and lower bounds over the specified period (i.e. where the two intervals do not overlap, or where the prior-year observation falls outside the confidence interval for the current year). 

Note that the changes in scores that we examine to identify statistically significant declines and rises in expression are calculated from the original scale values (point estimates vs. reported rounded integers).

‍Key periods analyzed

‍We explore Global Expression Score changes over time across 3 time periods: the last year (2021–2022), the last 5 years (2017–2022), and the last 10 years (2012–2022). 

For each timeframe, we identify countries that have shown meaningful improvement or deterioration, defined by a statistically significant score change over the period. 

In our reporting this year, we have not included Eritrea in the list of countries that have experienced a 5- and 10-year advance because its score increase is less than 1 point. While that shift is statistically significant over the period (see above regarding significant declines and advances), we do not consider it to not be substantively significant. 

The complete V-Dem Dataset includes more than 600 indices and indicators that measure different aspects of democracy worldwide, making it the world’s most authoritative data resource for examining the health of democracies globally.

V-Dem draws on theoretical and methodological expertise from its worldwide team to produce data in the most objective and reliable way possible. Approximately half the indicators in the dataset are based on factual information obtainable from official documents, such as constitutions and government records. 

The remainder consists of more subjective assessments on topics like democratic and governing practices and compliance with de jure rules. On such issues, typically at least five experts provide ratings for the country, thematic area, and time period in which they have expertise.

To address variation in coder ratings, V-Dem works closely with leading social science research methodologists and has developed a Bayesian measurement model that, to the extent possible, addresses coder error and issues of comparability across countries and over time. Additional data (including coder score changes for previous years) are incorporated in every update, which improves the overall model.

V-Dem also provides upper and lower point estimates, which represent a range of probable values for a given observation. When the ranges of two observations do not overlap, we are relatively confident that the difference between them is significant.

V-Dem is continually experimenting with new techniques and soliciting feedback from experts throughout the field. In this sense, V-Dem remains at the cutting edge of developing new and improved methods to increase both the reliability and comparability of expert survey data.