Regime types and regime change: A new dataset on democracy, coups, and political institutions

Abstract

Social scientists have created a variety of datasets in recent years that quantify political regimes, but these often provide little data on phases of regime transitions. Our aim is to contribute to filling this gap, by providing an update and expansion of the Democracy-Dictatorship data by Cheibub et al. (Public Choice, 143, 67–101, 2010), originally introduced by Alvarez et al. (Studies in Comparative International Development, 31(2), 3–36, 1996), where we add the following three features: First, we expand coverage to a total of 192 sovereign countries and 16 currently self-governing territories between 1950 and 2018, including periods under colonial rule for more than ninety entities. Second, we provide more institutional details that are deemed of importance in the relevant literature. Third, we include a new, self-created indicator of successful and failed coups d’état, which is currently the most complete of its kind. We further illustrate the usefulness of the new dataset by documenting the importance of political institutions under colonial rule for democratic development after independence, making use of our much more detailed data on colonial institutions. Findings indicate that more participatory colonial institutions have a positive and lasting effect for democratic development after transition to independence.

1 Introduction

Empirical studies comparing political institutions, their determinants, and the possible outcomes they produce has increased almost exponentially in recent years. To this end, a variety of excellent datasets have been created by social science researchers that attempt to quantify different political regimes at the country level, most notably Polity IV, Freedom House (FH), the Database of Political Institutions (DPI), and most recently Varieties of Democracy (V-Dem).

In this paper, we introduce a dataset of political regimes and institutions that is specifically designed to facilitate empirical research on political transition phases and breakdowns, which we believe is currently somewhat underdeveloped in the field. It consists of three main elements. First, we update and expand the binary Democracy-Dictatorship (DD) data by Cheibub et al. (2010), originally introduced by Alvarez et al. (1996) and Przeworski et al. (2000), which has been widely used by researchers since its first publication and is arguably more sensitive to phases of regime transition. We expand time and regime coverage by updating the data to 2018 and including three additional countries and 16 currently self-governing territories. Additionally, we classify the political institutions of territories that either are or were formally under colonial rule, constituting a maximum number of 96 colonies in 1950. Second, we provide more institutional details for each country, such as parliamentary type, a spatial democracy measure, a binary indicator for the implementation of new constitutions, etc. Third, the dataset includes a new, self-created indicator of successful and failed coups d’état, which we uniquely code according to the regime categorization by Cheibub et al. (2010), operationalizing a definition similar to that of Powell and Thyne (2011).

Our dataset is therefore not intended to replace those of, e.g., Beck et al. (2001), Boix et al. (2013), Vreeland (2008), Cheibub et al. (2010), or Coppedge et al. (2016), but merely to complement already existing ones with important additional information on regimes and their political institutions. We thereby hope to enable researchers to gather more information on regime characteristics and particularly on the topic of regime change and political transitions.¹

2 The DD update

The DD indicator, first introduced by Alvarez et al. (1996) and Przeworski et al. (2000) and later updated and expanded in the much-cited Cheibub et al. (2010), is a dichotomous indicator of democracy based on a minimalist definition of the concept. Cheibub et al. (2010) coded almost 200 countries on whether or not elections were conducted, whether these were democratic in the sense of making legislative and executive offices de facto contestable, and if there had been a peaceful turnover of such offices following those elections. We follow this approach by defining democracy as a set of political institutions in which properly contested, repeated and repeatable elections are free and fair – as assessed by international observers from democratic countries – and create ex ante uncertainty for the incumbent government and de facto ex post irreversibility of election results.²

Including all years between 1946 and 2008 in which a country was sovereign, the full dataset consisted of exactly 9159 total observations. Ever since its first publication, the DD dataset has been remarkably successful, which is mainly due to the fact that it employs objective criteria and operational rules to capture political democracy. According to conservative estimates, it has already been cited by more than 700 published scholarly articles since its release in 2010.³

As noted, the DD democracy indicator, as well as further indicators in the database, is clearly based on a minimalist concept of democracy. We thus necessarily place ourselves in the still ongoing debate between minimalist and maximalist conceptions and measures of democracy, according to which measuring democracy is never an exercise driven by some obvious optimality considerations, but always involves a choice between different potential problems.

First, minimalist definitions may arguably lack theoretically relevant attributes. Munck and Verkuilen (2002), for example, stress that the much-used Polity IV indicator is insensitive to restrictions on electoral participation. Another typical critique against minimalist measures is that they virtually always lack normatively desirable features, although there is also much discussion of which specific features ought to be associated with democracy. Indicators resting on maximalist definitions of democracy typically combine a number of sub-indicators measuring elements of electoral rights, political liberties, the rule of law, and social rights (e.g. Møller and Skaaning 2011). However, while such indicators can obviously get closer to measuring the quality of some form of normative ideal of democracy, maximalist measures risk providing quasi-answers to a priori interesting questions that are nevertheless merely resolved “by a definitional fiat” (Alvarez et al. 1996, 20). As such, while fully maximalist measures can be important in political debates, they are often of quite limited analytical value. When, for example, social rights are thought of as a normatively desirable feature of democracy and thus come to form part of the measurement of democracy, it is no longer possible to analyze the potentially important question whether democracies are more likely to implement and enforce such rights.

Our approach is conceptually minimalist, while other indicators such as the Freedom House Gastil Index or the V-Dem measure of polyarchy clearly represent more maximalist approaches to the measurement of democracy. As our aim is primarily to provide data of analytical value, we continue the minimalist approach of Cheibub et al. (2010), but add additional features that allow researchers to separately analyze questions of, e.g., participation restrictions.

Our version of this dataset started as a simple update, as all observations of the original DD variables end in 2008, meaning that they do not include the full extent of major recent transition phases, such as the Arab Spring. In this process, we have further added features, and also changed certain elements. First, while the original dataset covered the period from either 1946 or independence to 2008, we now include the years 2009–2018, such that all countries are coded between 1950 and 2018. Yet, we retain the main structure and continue to code regime types in six categories. Democracies are sorted into three types coded 0–2: Parliamentary democracies, mixed democracies (relatively weak/ indirectly elected presidents), and presidential democracies. Non-democracies are likewise separated into three types coded 3–5: Civilian autocracies, military dictatorships, and royal dictatorships. The latter category consists of absolutist hereditary monarchies, while the distinction between civilian autocracy and military dictatorship rests on whether the head of state or government has a military rank or not. The typology is described in two variables, where the first provides the numerical code (DD regime) and the second the respective name of the type (DD category). The latter also provides information on whether a country was a colony in a given year, which we specify in a dummy variable (Colony). For colonies, we further include a separate variable that provides information on which colonial power ruled the particular territory in the years between 1950 and independence (Colony of).

Second, the dataset includes three new sovereign countries that are not in the original data: Monaco, South Sudan, and Tuvalu. We count South Sudan as independent since 2011, and Tuvalu since 1979, when it gained its independence from the United Kingdom. Conversely, we do not include now defunct states, including the German Democratic Republic, the Republic of Vietnam (South Vietnam), or the USSR. We treat Czechoslovakia as the predecessor of the modern day Czech Republic, West Germany as the predecessor of the current Federal Republic of Germany, the USSR as the predecessor of Russia, Yugoslavia, as well as Serbia and Montenegro, as the predecessor states of modern Serbia, North Vietnam as predecessor of the current Vietnam, and North Yemen (the Yemen Arab Republic) as the predecessor of the current Yemen. On an institutional basis, all these decisions are easily justifiable as, for example, the German Democratic Republic at reunification was transformed to new Länder in the Federal Republic of Germany. Throughout, countries are included with their modern names, e.g. such that the Democratic Republic of Congo also includes the periods in which the country was called Congo-Kinshasa or Zaire and Belize also includes the period as British Honduras.

Third, we include 16 currently formally non-sovereign territories as part of our updated DD version. Even though these are all technically non-independent states, they all operate with their own political institutions as effectively separate, or quasi-separate, entities. The dataset therefore uniquely includes Anguilla, Bermuda, the British Virgin Islands, the Cayman Islands, Gibraltar, and Turks and Caicos, which are all technically dependencies of the United Kingdom; Aruba, Curaçao, and Sint Maarten, which are dependencies of the Netherlands; the Cook Islands that are administered by New Zealand; and Guam, the Northern Mariana Islands, Puerto Rico, and the US Virgin Islands that are all unincorporated territories of the United States. We naturally also include Hong Kong and Macao, which are nowadays administered by China, but used to be British and Portuguese colonies, respectively. The background for including these territories is that they all have their own parliaments and governments that control every major policy area, except for foreign policy and defense, and interference by the sovereign homeland in public affairs is (or ought to be) minimal. Bermuda, for example, remains the largest overseas territory of the United Kingdom, and has been effectively self-governing since 1620, boasting one of the oldest parliaments in the world. Tiny Anguilla has enjoyed a similar status ever since effectively seceding from Saint Kitts and Nevis in 1976, when the latter had begun its process towards independence. Conversely, we do not include any French possessions, due to the particular French institutional arrangement in which territories elect representatives directly to the legislative process in Paris, and interference in public policy issues is present on a regular basis. A similar institutional choice applies to Danish possessions in the North Atlantic, which we also exclude. This effectively makes our dataset one of the few empirical references that directly codes political institutions of colonies in a systematic way, which essentially goes beyond employing proxies like colonizer identity and settlement statistics (La Porta et al. 1998; Acemoglu et al. 2001).

The fourth difference from the original dataset is that we have changed the approach to the temporal structure of the data. Cheibub et al. (2010) strictly code all features in calendar years, while we apply a different timing rule. In order to ease the empirical application of the data, we count all regime changes before July 1 of year x as pertaining to year x, and all regime transitions after that date as pertaining to year x + 1. Relative to the original approach, our data are therefore lagged by half a year. The reason we make this seemingly arbitrary choice is easily explained: For example, when Spain’s long-time dictator Francisco Franco died in November 1975, it started the country’s democratization phase, during which it first changed from a military to a civilian dictatorship, and only then became a democracy. Would we not introduce this half-year time lag, the change from military dictatorship to civilian autocracy would be ascribed to the whole year of 1975, even though it was only effective for about a month of that same year. However, we also add an indicator of whether a regime transition occurred in the second half of the year, which easily enables all users to restore the original timing rule, if needed (Regime change lag).

As specified above, the dataset includes a variable capturing whether countries enjoy full democratic rights, understood as the right for a full franchise to vote and run for office and generally free and fair elections that determine legislative and executive offices (Democracy). However, we here further introduce two innovations. First, we offer three dummy variables that provide more information on why a country is coded as democratic or autocratic: whether power has actually changed peacefully with the present institutions as a result of elections (Alternation), whether the elections offer a de facto choice between different parties (Multiparty), and whether international election observers from democracies considered the elections free and fair by regular standards (Free and fair election).

Second, we introduce the important innovation that our democracy variable extends to the full period from 1950, i.e. before all countries became independent. Thus, it captures the somewhat paradoxical situation that some territories received democratic representation rights (again, except foreign policy and defense) before becoming independent. A relatively inconsequential example is Botswana that held its first free elections in the spring of 1965 but only gained its formal independence from the United Kingdom on the 30th of September the following year.⁴ However, the dataset also includes several countries that experienced many years of colonial democracy before independence, such as Jamaica, where the population gained full political rights in 1942, some twenty years before its formal independence.⁵ A major feature of our dataset is therefore that the main political institutions of quasi-independent countries are coded, even though these are not fully sovereign territories in a legal sense. We derive the relevant information from the background data on suffrage in Przeworski (2007) combined with information on the fairness of elections from newspapers, magazines and the Encyclopedia Britannica (2018).⁶ In addition, we also code a variable (Electoral) that captures whether a country has no regular elections (a score of 0), elections in an effectively one-party state (1), elections with opposition parties but without an actual chance of government change (2), and full democracy (3). This variable also extends back to the colonial period of countries that were not independent for some or all of the period since 1950. It thus allows users to separate full democracy, a score of ‘3’, from a situation of category ‘2’, which corresponds to what is otherwise sometimes called electoral autocracy, competitive authoritarianism, and illiberal democracy (LeDuc et al. 2010; Levitsky and Way 2002; Zakaria 1997).

The result is a dataset covering 208 countries and self-governing territories, observed for all years between 1950 and mid-2018. Of all 212 fully or partially sovereign countries in the world, we only leave out four from our dataset: Andorra, the Vatican, Kosovo, and San Marino.⁷ Andorra is excluded due to its somewhat complicated status as a sovereign nation, where the function of its official head of state is shared between the French president ex officio and a bishop appointed by the Vatican. San Marino is, in most ways, de facto a part of Italy, and approximately 20 % of the population are Italian citizens. Finally, Kosovo remains under joint protection of the United Nations and the European Union and is only partially recognized internationally, while the Vatican City State is a unique construction among sovereign states that cannot easily be compared to anything else. The full data therefore covers 13,728 country-year observations, of which 10,281 pertain to formally sovereign nations.

A relevant question to ask is how the updated data compares to existing alternatives. Comparing the 8535 observations from the original DD dataset that coincide with ours, we find 94 instances in which we disagree with the coding. Most of these are essentially due to the different timing approach, such that only a year separates a regime change in the original data by Cheibub et al. (2010) and our present data update. We list all remaining discrepancies in the appendix (Table A2).

Differences to the recent alternative by Boix et al. (2013) are also minor (henceforth BMR): Of 9015 joint observations, we disagree with the BMR data on only 322 observations (3.6%). About half of these discrepancies are due to timing differences, where BMR also follow the calendar year and thus tend to record a regime change in the year before we do. The remaining half results from the different regime definitions that are employed by both. Discrepancies between our data and the V-Dem data presented by Coppedge et al. (2016) are similarly small, although substantially larger when using its polyarchy indicator.

Differences are much more pronounced when compared to the Polity IV dataset (Marshall and Jaggers 2010). Of the 9131 joint observations, our data and Polity IV disagree on 860 observations (9.4%) for what Polity defines as “full democracy” at a minimum score of 6, and on 945 observations (10.3%) when setting this level at the lower score of 3. In about a third of these cases, the discrepancy occurs when Polity codes a country as democratic, while we code it as a civilian autocracy, i.e. a country in which the government most likely cannot lose an election. For example, this is the case for South Africa, which is counted as fully democratic in Polity IV and BMR, but not according to the DD dataset (see Cheibub et al. 2010 for a more extensive discussion). A comparison with the Freedom House (2016) indicator of political rights and civil liberties reveals similar discrepancies. In 375 of a total 7880 joint observations, Freedom House (FH) codes a country as democratic while the updated DD indicator does not, and in 447 cases it counts an observation as politically unfree, while the DD indicator codes it as democratic. This amounts to a total of 822 cases (10.4%), where both datasets differ in their evaluations of political regimes.

Overall, the updated DD indicator thus compares well to other democracy measures, even though the comparison also reveals that correspondence with the BMR dataset is much clearer than with other alternatives. Fig. 1 highlights differences for the illustrative case of Ghana, with full lines representing democracy, and dotted lines representing periods under autocratic rule. All indicators agree that Ghana is fully democratic from 2006 onwards, but vary considerably regarding the timing of Ghana’s democratization. In addition, only DD and BMR code the period 1969–1971 as democratic, and FH does not consider the brief democratic spell of 1979–1981. The main reason is that the BMR and DD datasets, as well as our update, all rest on a minimalist conception of democracy (i.e. electoral democracy). In turn, Polity IV is insensitive to participation restrictions and FH rests on a maximalist definition, which is why they both tend to be more restrictive in their definition of institutional democracy (Munck and Verkuilen 2002). In any case, it is clearly visible that our DD update maximizes observations on regime transition phases, because it codes all phases when Ghana alternated between electoral democracy and different autocratic regimes from 1950 to 2018, including its colonial period before 1957.

Fig. 1

Democratic and autocratic spells in Ghana, five indicators. Note: full lines denote periods categorized as democratic; dotted lines periods of autocracy

3 Additional institutional details

Besides the update of the democracy status, we further add a few institutional details to our broadened DD dataset for sovereign and non-sovereign entities that we believe will specifically facilitate the investigation of regime transitions. First, we include a spatial index of democracy, which is coded as the unweighted average of the democracy variable in geographical neighbors (Spatial democracy). This allows the direct comparison of political institutions to countries’ neighbors and thus permits the estimation of spatial spill-overs of regime change (cf. Ziblatt 2006; Aidt and Franck 2015). We provide a similar spatial index of the electoral variable (Spatial electoral). Second, we include a dummy for communist or socialist regimes (Communist). Third, we include some brief information on all monarchs and presidents, i.e. heads of state with some level of constitutional status. We report the name (Monarch name, President name), birth year (Monarch birthyear, President birthyear), year of accession (Monarch accession) or election / effective power transfer (President accession), and gender (Female monarch, Female president). Fourth, we also include a dummy capturing whether the status as head of state is temporary (Interim phase), because there have been more than two in the course of one year, as often occurs during an institutional crisis phase or a democratic regime change.⁸ Fifth, we further add a dummy, based on the Comparative Constitutions Project (Ginsburg et al. 2009), which denotes whether a new constitution was implemented in a given year (New constitution). This captures the existence of major de jure regime changes.

Sixth, we add a number of details on political institutions. These include the number of chambers in parliament, allowing users to separate uni- and bicameral institutions (No. of chambers in parliament). We add the number of members of the lower house in all cases, and the number of members of the upper house in bicameral systems (No. of members in lower house, No. of members in upper house). The data also include information on which particular system is used for the election of members of the lower house (Election system). Furthermore, we supplement specific information on electoral systems with a dummy for whether a majority of members are elected using proportional representation systems, or not (Proportional representation). We also include a dummy for whether the right to vote and run for office is extended to all citizens above the legal voting age, regardless of gender, race, or income (Full suffrage) and information on which types of restrictions on suffrage apply (Suffrage restriction). This information is intended to alleviate some of the well-known problems of minimalist indicators (cf. Munck and Verkuilen 2002).

As the final main feature, we include specific information on the political institutions of colonies. We add the name of the colonial legislature, more information on suffrage restrictions, such as if there were separate rolls for white and indigenous voters or other particular voting restrictions, and the distribution of elected and non-elected members of the legislature. This set of information also includes a dummy capturing whether the colony was formally self-governing or not (Self-governing). The dataset therefore enables the particular study of the evolution of political institutions before independence, and – as we illustrate in section five – their potential importance for modern institutions and democracy.

4 The coup data

Besides the update of the regime classification in Cheibub et al. (2010), and the specific information on colonial political institutions that we document in the next section, another main feature of our dataset is the incorporation of several self-compiled variables on successful and unsuccessful coups d’état. We include data on all verifiable coup attempts since 1950 and separate successful and failed attempts, all according to the regime categorization scheme in Cheibub et al. (2010).

A first problem is how to arrive at a working definition of coups and coup attempts. As stressed by Powell and Thyne (2011), no standard definition exists although there is consensus that coups must have as their objective to overthrow the chief executive of a country. A first part of our coup definition is thus that coup attempts are events in which some actor or actors illegally attempt to take power over the executive branch of government. However, the definition in Powell and Thyne (2011, 250) also requires that coups are “undertaken by any elite who is part of the state apparatus.” We deviate from this definition by not requiring that coup makers are members of an ex ante identifiable elite, as such a requirement would have peculiar consequences. One such consequence would be that the successful coup in Burkina Faso in August 1983 would arguably not fall within the definition. The coup leaders – Thomas Sankara and Blaise Compaore – both held the rank of captain and thus could not be said to belong to any political or military elite.

We therefore do not require coup makers to hold any particular rank or position within the state apparatus, but merely to be in some way linked to the state apparatus. Even though we do not believe it is necessary for coup leaders to form part of a clearly identifiable elite, we do consider the linkage to the state as rather important, especially when considering different groups competing for power with the means of violence, or the threat thereof, and the necessity to distinguish such situations from exclusively foreign attempts at influence. This can naturally include actors, who have previously been removed from their positions inside the state. In addition, we delimit coups from other conflicts such as revolutions, civil wars or successful protest movements, which would also fit these requirements (Goodwin 2001), in another way. We define coups as events that can last up to a week, and may or may not be violent, but always entail a latent threat of violence, although most coups are essentially bloodless (Powell and Thyne 2011). All coups and coup attempts typically last very few days, while other types of events that last longer (i.e. protest movements or civil wars), some of which may have rather similar characteristics, mostly turn into substantially longer-lasting conflicts.

To summarize, coups and coup attempts therefore have to fulfil the following definition to count as such in our dataset: First, the objective must be to overthrow the executive branch. Second, actors have to be previously linked to the state apparatus in some way. Third, a coup or coup attempt cannot last longer than a week at most.

Using this definition, we compiled the database by searching all given information employing newspaper material from the Lexis-Nexis database and, in cases dating before 1970, historical information from the Encyclopedia Britannica, Luttwak’s (1968) handbook of coups, Singh’s (2014) coups study, and newspaper archives. To enter the data, we require that a coup attempt is verifiable by more than one source, and the information on failed coups in the sources cannot only derive from the incumbent government. Finally, we do not include any events in which there were merely rumors of an attempt or coup plots for which no independent information exists. As such, our database does not include rumored coups, which we define as cases where only government sources claimed a coup had been foiled, or cases in which only one international newspaper reported that a coup attempt had taken place. To avoid coding these false positives, all coup attempts in our data had to be reported independently by at least two international newspapers.⁹

Subsequently, we compared our list with the information available in Polity IV, MacGowan (2006), and Powell and Thyne (2011), who each provide a list of coups and coup attempts. While we are able to verify all successful coups in the alternative sources, we also found a substantial mismatch between these lists, and a number of cases where both lists include the same coup, but disagreed on the exact timing.

The resulting database has a number of notable features. First, we code whether a successful coup in a particular year occurred (Successful coups), and if the coup was primarily led by the military, a group of civilians, or in rare cases members of the royal family (Type). Second, the approach allows us to code if a failed coup occurred in a country (Failed coups), many of which are either not covered by existing databases, or seem to be coded on the basis of rumors. Third, the coup variables include four numerical variables capturing the number of coups in a country in a given year (All coups), the number of coup attempts that were successful (Successful coups) or failed (Failed coups), and an indicator of whether any more coup attempt occurred (First coup, Second coup, Third coup). In subsequent variables, we include information on the month of the coup (month), the coup leaders’ names (Coup leaders), military and civilian ranks (Military rank index, Civilian rank index), and whether the coup was military, civilian, or royal (Type). We do so for the first three coup attempts in each year, which is the maximum number of separable events that we find in a single country and year since 1950.¹⁰

In total, the coup database includes 498 country-years in which coups occurred. In 34 cases, more than one coup occurred during a single year, which brings the total count of coup attempts to 537. In a total of 243 cases, or slightly less than half, these were successful at overthrowing the government in power. A total of 393 coup attempts were led by current or former members of the military, 132 were led by civilians, while only 12 were insider coups within royal families. The success rate of military and royal coups is almost exactly 50 % while civilian coups are substantially less likely to succeed at 35 %.

In combination with the updated information on democracy and regime types, the coup indicators allow identification of regime transitions that would not necessarily show up in the DD dataset. For example, the dataset includes successful military coups against incumbent military dictators, and civilian coups against incumbent civilian autocracies. In such cases, significant regime change events obviously occurred, but the regime categorization in the DD dataset does not change. The dataset also provides a potential background for subsequent institutional and regime changes – examples include several cases in which democratization occurred almost immediately after a coup or coup attempt – and the ability to identify the time since the last regime transition, as well as crucial information on the institutional set-up prior to home rule or independence, which we exploit in the following.

5 Colonial institutions and democratic development: An application

To further highlight the usefulness of the unique features in our dataset with a concrete example, we provide a direct test of one of the more disputed ideas in development economics: the question whether colonialism affects post-independence political institutions in a path-dependent manner, and if this effect can generally be evaluated as positive for democratic development. Recent contributions show how debated this question still is, with scholars like Ferguson (2012) claiming a largely positive legacy of (British) colonialism, while Acemoglu and Robinson (2012) highlight the substantially negative record of (extractive) colonial structures for present day political and economic development. Most scholars agree that the impact of colonialism for affected countries is strong and persistent to the present day, but somewhat disagree regarding the sign of the impact. Recently, even this has been called into question by Maseland (2018), who finds colonial legacies to be sharply declining in importance, at least for African political institutions and their overall economic development.

Related questions are not particularly new in the literature on political and economic determinants of democracy, where they have been discussed almost ever since the start of decolonization itself (e.g. Emerson 1960). Early empirical work was conducted by Hadenius (1992) and Lipset et al. (1993), although these contributions were nevertheless strongly limited by the availability of adequate data. Once more comprehensive datasets became available by the late 1990’s, the issue was subsequently addressed in large-n cross country studies, mostly using cross sectional data, and in some cases panel data analysis. Most of this literature effectively investigated the impact of colonial structures on comparative economic development, mainly treating the potential effect of colonialism on democratic institutions as one possible channel for outcomes on income per capita (Grier 1999; Sokoloff and Engerman 2000; Acemoglu et al. 2001; La Porta et al. 2008; Feyrer and Sacerdote 2009). Within this literature, a subset of contributions has more specifically analyzed the importance of colonialism for the democratic development of former colonial dominions and the exact transmission mechanisms of democratic and participatory legacies (Bernhard et al. 2004; Lange 2006; Olsson 2009; Jones 2013; Guardado 2018; Lee and Paine 2018). Notwithstanding, both strands of literature find overwhelming evidence that colonial history matters a great deal for the development of formal institutions and economic prosperity after transition to independence. Questions on the persistence of these effects have only been raised very recently, with some authors emphasizing the declining impact of colonialism and the growing importance of pre-colonial institutions (Michalopoulos and Papaioannou 2013; Maseland 2018).

What most of these studies have in common is that they are effectively forced to develop some type of empirical proxies to capture the nature of colonial institutions, as direct comparative measures of the latter are often not available. Here, a number of different strategies can be distinguished: First, a large share of the literature uses dummy variables to test whether the identity of the colonizer (or the type of legal system established by the colonizer) has a lasting effect on economic and political development (La Porta et al. 1998, 2008; Grier 1999; Przeworski et al. 2000; Bernhard et al. 2004; Lange 2006). This literature usually relates more favorable outcomes with British colonialism and a common law tradition, as compared to French or Spanish colonialism and the associated civil law tradition.¹¹

Second, and connected to the first group of studies, a number of authors have also focused on the temporal dimension of colonialism, mainly capturing its overall duration (Grier 1999; Feyrer and Sacerdote 2009; Olsson 2009). Generally, findings point in the direction that longer periods of colonial rule are beneficial for contemporary levels of income per capita and democracy, as state structures had more time to be consolidated during the phase of foreign rule.

Third, another strain of related literature basically argues that geographic and climatic conditions had a decisive impact on whether the colonizing power set up extractive or inclusive institutions in the host society, which in turn also largely determines current political and economic status. The corresponding authors have tried to proxy for this fact by employing several different measures, for example, by capturing climatic conditions (Sokoloff and Engerman 2000), or the mortality rates of settlers and the indigenous population density (Acemoglu et al. 2001). Findings in this literature generally point to the fact that former colonies with more extractive institutions are also less likely to be economically developed and politically less participatory to the present day.

Finally, some recent studies develop more fine-grained indicator of political institutions under colonial rule: Jones (2013) measures the comparative pay of colonial governors, arguing that better paid colonial posts also attracted more able administrators. These territories were consequently better managed, leaving a path dependent trajectory after transition to independence that is visible in the currently higher economic and political development of affected territories. Similarly, Guardado (2018) captures prices at which the Spanish royal government sold colonial governorships in Peru during the late 17th and early eighteenth century, showing that provinces with a higher extraction potential sold at higher prices and that these same areas of Peru present comparatively worse socio-economic outcomes to the present day.

In an attempt to shed some new light on this discussion, we use the colonial features of our new dataset, which offers more direct indicators of colonial institutions than most previous studies. Employing a sample of former colonies that are all observed in our data before and after their transition to independence, this gives us a panel data set covering a total of 66 countries/colonies, and up to almost 3000 individual observations. Apart from a paper by Lee and Paine (2018), which conducts a rather similar exercise with the more restricted V-Dem data by Coppedge et al. (2016), we are unaware of any further study that directly associates pre-independence political institutions of colonies with the democratic outcomes of post-independence.

Our main dependent variable is the extended democracy indicator by Cheibub et al. (2010), which is observed on an annual basis, or in overlapping five year averages such that we require that the country has been democratic for at least five consecutive years. In further tests, we use three additional dependent variables to test if colonial political institutions shape more than merely the electoral structures. Specifically, we are interested in whether colonial institutions have also instilled stable post-independence mechanisms that enable the peaceful negotiation of conflicts. We use the coup indicator from the new database (whether any coups occurred in a given year), a dummy capturing whether any political assassinations occurred in a year, which we derive from Banks and Wilson (2013), and a measure of the absence of general government repression from Fariss (2014).

Our main independent variables directly capture political institutions under colonialism. In particular, the electoral variable from our data allows us to observe whether a colony had fully democratic institutions at least five years prior to gaining independence, or if it had an elected parliament or other similar type of consultative assembly (see Section 3). For the present analysis these variables are denoted as colonial democracy and colonial representation, respectively. The first is equal to one, if colonial political structures were fully democratic at least five years prior to its independence, and zero otherwise. The second is equal to one, if the colony had any type of elected domestic representation at least five years prior to its independence, and zero otherwise.

To control for the formal status of territories, we introduce a dummy that is equal to one, if the country is fully independent, and zero if it has the status of a colony. Further control variables are taken from the literature reviewed above, where we introduce the log of purchasing-power adjusted GDP per capita and the log of total population size, both of which derive from the Penn World Tables, mark 9 (Feenstra et al. 2015). We also introduce a dummy for former colonies of the British or Dutch commonwealth, a dummy for former French colonies, a dummy that captures former colonies situated in Africa, and another dummy that captures former colonies situated in Latin America and the Caribbean, all based on information in Encyclopedia Britannica (2018). Descriptive statistics of all variables are shown in Table A1 of the appendix.

Table 1 shows the results for our analysis of colonial representation and its association with democratic development under independence. Here, columns 1 to 3 employ annual observations of the dependent variable, while columns 4 to 6 use the overlapping five year averages. The corresponding equations are all estimated with a random effects logit regression model.¹² As to the controls, we observe that neither income per capita nor total population are significantly associated with democratic political structures, while becoming independent presents a significant and negative relationship with political democracy in our sample. Coefficients on colonizer origin are all insignificant, as is a location in Africa, while being located in Latin America and the Caribbean makes political democracy significantly more likely in some models.

Table 1 Colonial institutions and subsequent democratic development

In column 1, our primary controls of interest both enter the equation with a significant and positive sign. By itself, both these variables do not carry too much valuable information though. In order to capture the impact of colonial representation and colonial democracy for democratic governance after transition to independence, we further introduce an interaction of the colonial representation variable with the independence dummy in column 2; collinearity problems prevent us from adding an interaction with colonial democracy. The marginal effects of colonial representation and colonial democracy during the phase of independence are shown in the lower part of the table, where both are highly significant and enter the equation with a positive sign. All else equal then, having domestic representation or democratic structures during the colonial phase, makes the presence of political democracy after transition to independence significantly and substantially more likely. In column 3, we find that these effects also hold even 30 years after independence. We thus observe no evidence that their influence is decreasing over time.

A separate question, nevertheless, is if the political institutions inherited from colonial times prove to be stable. In columns 4 to 6, we therefore repeat the previous estimation models, but employ a measure capturing if democracy is stable across overlapping five-year periods as the dependent variable. Interestingly, coefficients on the income per capita and total population variables remain insignificant but increase substantially in size. Finally, the lower part of columns 5 and 6 again shows the marginal effects for our variables of interest. We find that both enter the equation with a positive sign, but only colonial democracy achieves significance at conventional levels.

However, it remains an open question if having representative colonial institutions affect any ‘quality’ of democracy. In Table 2, we therefore test if societies with a democratic colonial past are also more likely to be peaceful and non-repressive. We apply the same approach as in Table 1 with the exception of columns 5 and 6, in which we employ a random effects OLS estimator as the Fariss (2014) repression measure is a continuous variable.

Table 2 Colonial institutions and subsequent peace

In columns 1 and 2, we find no evidence that colonial institutions affect the risk of observing a coup attempt after independence: only economic development and being placed in Latin America and the Caribbean are significant. Conversely, we find that both larger and poorer countries are more likely to experience assassinations (columns 3 and 4), and that assassinations are less likely after independence. The latter finding is likely to reflect that many independence movements resorted to assassinations prior to gaining independence. More importantly, the findings show that countries with representative institutions during their periods as colonies are also less likely to experience assassinations after transition to independence, an effect that even seems to be increasing over time. Similarly, we observe that independent, richer and less populous countries have less repressive governments in columns 5 and 6. Again, government repression is also shaped by the colonial institutions in a significant and highly important way: Countries that inherited either representative or fully democratic institutions from their colonial past are substantially less likely to have repressive governments. We also find no significant evidence that this association is decreasing over time.

Overall, these results confirm the idea that the presence of representative organs or democratic structures during colonialism could make the peaceful maintenance of political democracy significantly more likely for newly independent states. Still, whether or not countries had democratic colonial institutions can, in principle, also reflect deep cultural factors that persist to this day, which also affect the probability of having a functioning and peaceful democracy. Given that historically more developed colonies are also those same ones for which more data is currently available, selection bias might be an additional driver of our findings. As Przeworski and Limongi (1997) further highlight, apart from unmeasured factors that facilitate democratic rule during colonial times and that also facilitate democracy following independence, it is also feasible to distinguish between the emergence of a democracy and its sustainability to assess regime dynamics after independence. While the former is greatly influenced by economic development in Western states, development seems to have little impact on the sustainability of democracy in former colonies, according to these authors.

In order to somewhat address the issues of endogeneity and regime dynamics, we further conduct a cross-sectional analysis with instrumental variables in Table 3. Here, we relate colonial democracy and colonial representation in 1955 to two different dependent variables that are supposed to somewhat capture democratic emergence and stability: One is the share of time a country has been democratic from independence to 2015 (i.e. share democratic), while the other is a dummy that is equal to one, if the country has ever been democratic for a full year since independence (i.e. ever democratic). Colonial democracy and colonial representation before independence are instrumented via population density data for that same year, which we obtain from information in the Statesman’s Yearbook (1950). Outside of the influence that population density in 1950 should have for the probability of observing colonial democratic and representative structures at the time, it is not clear how this variable could possibly be related to current political institutions. Given the relatively small degrees of freedom in this cross-sectional dataset, we only introduce the log of per capita GDP and total population in 2015 as further control variables. Summary statistics are also shown in Table A1.

Table 3 Colonial institutions in 1955 and subsequent democratic development

Findings in Table 3 suggest that colonial democracy and colonial representation have a substantial influence on both the sustainability of democracy (columns 1 to 4) and the probability of its emergence (columns 5 to 8). In both cases, tests indicate that the instrument is exogenous and coefficients show that the pure numerical effect might actually be substantially underestimated in our previous estimations. In general, we are therefore confident that the presence of representative democratic structures during colonialism makes the establishment and maintenance of political democracy significantly more likely after independence, rather than just picking up common unmeasured factors that drive both outcomes.

6 Summary

As the famous British economist Paul Collier has recently put it, large datasets and the corresponding effort to collect the information contained in them are the public goods of economic research (Collier 2013). Recent studies on regime transition point to a number of open questions that require more accurate data, data capturing different aspects than what is available, or simply more data (Ziblatt 2006; Aidt and Franck 2015). Our aim is to contribute to partially filling this gap by providing an update and expansion of the DD dataset, and a new and expanded dataset on coups that includes several novel features. This data is available for 192 sovereign countries and 16 currently self-governing territories between 1950 and 2018. Hereby, we hope to enable researchers to drive forward important research questions on regime transitions and the political and economic outcomes they produce that are somewhat underrepresented among the empirical studies of political regimes.

We attempt to show the usefulness of our data collection efforts with an application of our colonialism dataset, where we re-investigate the importance of colonial structures for the political institutions of countries after transition to formal independence. Contrary to earlier studies, our data offers a direct measure of colonial institutions; in particular, it captures whether the colonial government had a representative body, or whether its government was fully democratic. Findings show that having domestic representation or democratic structures during the late colonial phase makes the presence of political democracy after transition to independence significantly more likely. The additional finding that the colonial institutions also appear to affect how peaceful and non-repressive the political process is points to the usefulness of further research using the present and other datasets that capture elements of the institutional setup of the developing world.