The Political Settlements Dataset: Power Configurations and Political Blocs in the Global South, 1946-2018

Abstract

Political scientists have long sought to understand power dynamics across the Global South, but efforts to systematically capture political power have been limited, and existing datasets have faced conceptual and measurement challenges. The Political Settlements (PolSett) dataset was developed to address these issues. This original expert survey-based dataset includes over one hundred political economy variables coded for forty two countries from 1946 or independence to 2018 (totalling 2,718 country years). It allows for a detailed mapping of countries’ power configurations, capturing information on the relative size, power and social composition of competing political blocs in society, as well as their internal cohesion, accountability relations and distribution of benefits. The dataset is expected to enable more direct and rigorous analyses of power differentials in the Global South and their impact on political and economic developments. This letter describes the conceptual background of the dataset, how it was constructed, discusses its validity, and highlights some of its key features.

RESUMEN

Los politólogos han buscado, durante mucho tiempo, la forma de comprender las dinámicas de poder en todo el Sur Global. Sin embargo, sus esfuerzos para capturar, de manera sistemática, el poder político han sido limitados y los conjuntos de datos existentes se han enfrentado a desafíos, tanto conceptuales como de medición. El conjunto de datos de Acuerdos Políticos fue desarrollado con el fin de abordar estas cuestiones. Este conjunto de datos original se basa en encuestas de expertos e incluye más de 100 variables de economía política, codificadas para 42 países, desde 1946, o desde su independencia, hasta 2018 (un total de 2.718 años de países). Esto permite realizar un mapeo detallado de las configuraciones de poder de los países, capturando, de esta manera, información sobre el tamaño relativo, el poder y la composición social de los bloques políticos que compiten en cada sociedad, así como su cohesión interna, las relaciones en materia de rendición de cuentas y la distribución de beneficios. Esperamos que este conjunto de datos permita la realización de análisis más directos y rigurosos con respecto a las diferencias de poder en el Sur Global y su impacto en los acontecimientos políticos y económicos. Este artículo describe cómo se construyó el conjunto de datos, analiza su validez y destaca algunas de sus características clave.

RÉSUMÉ

Les politologues tentent depuis longtemps de comprendre les dynamiques de pouvoir dans les pays du Sud, mais les efforts de représentation systématique du pouvoir politique restent restreints, tandis que les ensembles de données existants sont confrontés à des défis conceptuels et relatifs aux mesures. L’ensemble de données Political Settlements (Règlement des différends) a été élaboré pour répondre à ces problématiques. Cet ensemble de données inédit fondé sur un sondage d’experts comprend 100 variables économico-politiques codées pour 42 pays entre 1946 ou leur indépendance et 2018 (pour un total de 2,718 années d’existence de pays). Aussi est-il possible de représenter précisément les configurations de pouvoir de chaque pays, en incluant des données sur la taille relative, le pouvoir et la composition sociale de blocs politiques concurrents au sein d’une société, ainsi que leur cohésion interne, leurs relations redditionnelles et la répartition des avantages. L’ensemble de données devrait permettre d’entreprendre des analyses plus directes et rigoureuses des différentiels de pouvoir dans les pays du Sud et de leur incidence sur le développement politique et économique. Cette lettre décrit la construction de l’ensemble de données, traite de sa validité et souligne certaines caractéristiques clés.

Keywords:

• leader support
• political cohesion
• political incorporation
• political threats
• power concentration
• societal power distribution

Introduction

Political scientists have long been invested in understanding the nature and distribution of political power in countries across the Global South. These countries are often characterized by complex political systems, where formal institutions and laws may reflect poorly the reality on the ground. As a result, scholars have sought to uncover how power is actually exercised in these contexts, and how it affects issues such as political stability, democratization, social justice, and economic development.

Despite significant recent advances, the discipline has struggled to systematically and comprehensively capture characteristics of power configurations across space and time. Several existing efforts have made important contributions, but are limited in several ways. Bueno de Mesquita et al. (2003) made an influential and parsimonious theoretical contribution to understanding power configurations through selectorate theory, but faced conceptual and measurement challenges in delineating the size and nature of the winning coalition, particularly in autocratic contexts (Gallagher and Hanson 2015). V-Dem (Coppedge et al. 2022) has more direct expert-based measures of the size of the winning coalition, provides important information on the socioeconomic characteristics of the winning coalition and the opposition, and has excellent cross-spatial and temporal coverage. However, it does not differentiate between major factions within the winning coalition and how they might differ in terms of relevant characteristics such as cohesion, hierarchy, relative power, or the means of integrating them into the regime. In contrast, important recent indices of authoritarian personalism (Gandhi and Sumner 2020; Geddes, Frantz, and Wright 2018; Weeks 2014) actively focus on the concentration of power in ruling coalitions, but have a narrow conceptualization of the term in that they only consider the power of the leader relative to other ruling elites, rather than taking a factional perspective. Vogt et al. (2015) provide the influential Ethnic Power Relationships (EPR) dataset, which allows us to examine the configuration of access to executive state power across autocracies and democracies, but not necessarily degrees of de facto political power.¹ By exclusively focusing on ethno-religious differences, moreover, power shifts based on other relevant political cleavages that occur between ethno-based regime changes are unfortunately overlooked.

Overall, there is a lack of fine-grained panel data that more broadly examines the diverse characteristics of power configurations across regimes in the Global South. Such a dataset would not only compare the ruling coalition with the opposition, but also open up the internal politics of the ruling coalition, allowing for a more nuanced understanding of power dynamics and patterns of societal incorporation in these countries. The Political Settlements (PolSett) dataset aims to fill this gap. Built over three years at the Effective States and Inclusive Development (ESID) Research Centre, this expert survey-based dataset includes over one hundred political economy variables coded for forty two countries in the Global South from 1946 or independence to 2018 (a total of 2,718 country years). Conceptually, it draws on “Political Settlements Analysis” (Kelsall et al. 2022; Khan 2010), which theorizes and studies how different distributions of de facto power among distinct groups in society shape political and economic developments. In the following sections, we define the core concepts underlying the dataset, describe how it was constructed, discuss its validity, and highlight some of its key features.

Conceptual Background

The initial motivation for creating the PolSett dataset was to better measure concepts and test hypotheses emerging from Political Settlements Analysis (PSA). PSA was developed in the mid-1990s by the political economist Mushtaq Khan (1995) in response to the difficulties of “New Institutional” approaches in explaining why certain institutional reforms (such as economic and political liberalization or industrial policy) succeeded in some countries and failed in others. Its central claim to date has been that the way powerful groups in society are configured or organized can help us explain such patterns; or, more ambitiously, help explain why different societies follow different development paths.

Khan’s (2010) point of departure, which has been highly influential, especially in UK development policy circles,² is that in developing countries especially, the building blocks of politics consist of patron–client networks or factions, that is, pyramidal aggregations of individuals and groups beneath a leader. These factions frequently mobilize politically to secure policies and institutions that will deliver benefits to them. Whatever their ideological, religious, or ethnic complexion, they are mainly about seeking rents or other economic benefits. If political and economic institutions are not delivering rents in line with these factions’ expectations, factions will mobilize to try and obstruct or change the institutions. Considerable political instability and perhaps violence are likely to result. However, when institutions deliver a distribution of benefits that is compatible with the distribution of factional power in society, we have what is called a political settlement, that is, a reproducible macro-social order.

Building on Khan, albeit with minor differences, Kelsall et al. (2022, 27) define political settlements as the ongoing agreement among a society’s most powerful groups over a set of political and economic institutions expected to generate for them a minimally acceptable level of benefits, which thereby ends or prevents generalized civil war. This is the definition applied in our dataset. Relatedly, (the degree or level of) power, in this context is defined by the ability of an actor or group, acting alone or with others, to change the political settlement or prevent it from being changed, where this ability might be shaped, inter alia, by constitutional authority, violence capabilities, or other disruptive potential.

Importantly, Khan (2010) argued that “clientelist political settlements” in the developing world differed significantly between each other with regard to how power was configured across groups and that this shaped their governments’ motivation and capability for developmental policy and implementation. Khan theorized (a) that the stronger the ruling coalition vis-à-vis the opposition (termed “horizontal power”), the longer its time horizon and thus the greater the motivation to engage in long-term (i.e., developmental) planning. And (b) that the more cohesive the ruling coalition and the weaker internal factions (termed “vertical power”), the less likely there will be pressure on leaders to permit unproductive rent-seeking.

To capture Khan’s distinction between factions within and outside the ruling coalition as concisely as possible in our survey and dataset, we conceptually divided society into three distinct blocs:

• the Leader’s Bloc (LB): that is, the segment of the population whose political loyalty the current de facto leader perceives s/he can be reasonably assured of, at least in the short-term;

• the Contingently Loyal Bloc (CLB): the segment of the population that is currently aligned with the de facto leader but whose political loyalty s/he cannot be assured of; and

• the Opposition Bloc (OB): the segment of the population that is not currently aligned with either of the above.

The ruling coalition would thus consist of those members of the LB and the CLB who control executive state power, but not the OB. For more detail, please see Section 2.1.2.2 in the online-appended codebook.

To make this distinction more tangible, we provide a brief example of the composition of blocs in Kenya’s ethno-regionally dominated political settlement from 2013 to 2017, based on our experts’ assessment. The LB, led by President Uhuru Kenyatta, an ethnic Kikuyu, consisted mainly of members of ethnic groups from Kenya’s central regions, namely Kikuyu, Embu, and Meru, organized in the National Alliance party. The CLB was dominated by the United Republican Party of Deputy President William Ruto (a fierce rival of Kenyatta’s in the 2000s and 2020s), made up mainly of Kalenjin, Gusii, and some Luhya from central-western Kenya. Lastly, the OB was made up of a more regionally diverse coalition representing most Luos, Kambas, Luhya, and Coastals, organized in the multi-party Coalition for Reforms and Democracy, led by the eternal runner-up and Luo leader, Raila Odinga.

It is important to note that PSA is both relevant to and transcends the classic democracy-autocracy dichotomy that has often been the focus of development policy. While the institutions for selecting political leaders are an important part of a political settlement, they do not necessarily determine its power configurations. Although autocracies are, on average, more likely than democracies to concentrate power in the ruling coalition, concentrated power may also exist in the latter. Classic examples are Nehru’s India or Mbeki’s South Africa. Likewise, dispersed power can apply to autocracies such as China under Chiang-Kai Check, Al-Hafiz’s Syria or Acheampong’s Ghana. Accordingly, one of the purported strengths of political settlement theory is its ability to explain divergent outcomes while holding the regime type constant. And indeed, Kelsall et al. (2022) have conducted regression analyses using our dataset, showing that the degree of power concentration is significantly associated with economic growth, even when controlling for regime type.

Critically, despite these potential strengths, and having provided the theoretical framework for a number of empirical studies and programs in development research and practice over the past decade, there has been little attempt to measure core PSA concepts such as the configuration of power across countries or over time. And, as noted in the introduction, existing data sets do not fill this gap. As a result, it is difficult to test the claims of the PSA more systematically and to address fundamental questions in political science about how political power is structured across societies and how different social groups are incorporated by rulers, among many others. To this end, we have created the PolSett dataset. In what follows, we will describe how we have gone about doing so.

Data collection

Understanding the composition and relationships of a society’s more and less powerful groups requires in-depth knowledge of a country’s political history. For this reason, we chose an expert survey as our approach to data collection. For each country, the survey relies on the judgment of at least three widely recognized experts in its modern political history, identified through personal networks and web searches.

Although ideally we would have covered all countries in the Global South, given resource limitations, we confined our survey population to a subset of countries (see Figure 1 below). Three key objectives guided our case selection. First, we wanted our sample to be as relevant as possible to ("clientelist") political settlement theory, which meant selecting countries that began as developing economies with substantial informal sectors (with agriculture classically being one of the most important informal sectors). Second, we wanted our sample to be as similar as possible on a range of structural economic and geographic variables that are classically thought to be relevant to development outcomes. This would allow small- or medium-N-based comparative research to reduce the number of confounding variables that need to be held constant when using the dataset and studying the political determinants of development. Finally, we wanted our sample to cover as much of the population of the Global South as possible within resource constraints. To this end, we selected all thirty seven coastal countries in the world that had low- or lower-middle-income statuses and a predominantly rural population in the 1960s, an agricultural sector that contributed at least ten percent of GDP until the 1980s, and a population that averaged more than five million throughout their post-independence history.³ We also added five other countries: Ethiopia, Rwanda, South Africa, Uganda, and Zambia. Although these countries narrowly missed our selection criteria, they represented critical existing case studies from the ESID research center, which specifically used PSA to study effective and inclusive policymaking in the Global South. Overall, our sample covers seventy six percent of the population of the Global South in 2018, while having a particularly strong coverage for African and Asian countries.

Figure 1. Number of political periods per surveyed country.

The survey instrument itself was developed over the course of a year. Before being distributed to all 129 experts, it underwent three rounds of piloting and subsequent feedback discussions and workshops with a dozen ESID researchers and ESID-external country experts respectively. The final survey then consisted of two distinct phases.

Periodization

To create temporal units for coding, we worked with our country experts to define political periods for each country, rather than coding temporal dynamics annually. We found this approach to have three potential advantages. First, it addresses concerns that experts might code excessively long periods out of convenience or fatigue, potentially leading to inaccurate variation. Second, the use of fixed periods not only provides coders with a reference point for variables, especially when coding across questions or blocks within the same question, but also ensures comparability across consistent time frames. Finally, this approach helps identify unsettled periods in a country’s history, which are exempt from coding.⁴

Accordingly, in Phase I of the survey we asked experts to confirm or challenge a list of political periods into which we had tentatively divided countries’ political history (since 1946 or independence, whichever was later, until 2018). Our aim was to capture significant changes in the political settlement of a country and its related components. As described above, at the macro level we divide society into three main blocs based on their relationship to a country’s de facto leader. Following this conceptualization, we decided to use leadership change as a good initial proxy for changes in the configuration of powerful groups. Using the Archigos database of (de facto) political leaders (Goemans, Gleditsch, and Chiozza 2009), each change in a country’s leadership was thus identified as a basic breakpoint. We then used additional databases and web resources to identify other potential breakpoints that could signal a major change or evolution in the de facto rules of the game and/or the configuration of power, namely:

1. the composition and power of the governing coalition;

2. formal and informal political institutions;

3. the degree of violent contestation and/or the propensity of the government losing a war;

4. the economic and social ideology of the head of government; and/or

5. the degree to which the state can conduct domestic policy autonomously of foreign entities.

We then asked coders to question and suggest changes to our initial periodization. This process involved several iterations of back-and-forth discussions with and between coders. The average country in our dataset is covered for 61.77 years, had 8.59 different leaders, and 14.31 political periods (resulting in an average period length of 4.31 years). Importantly, our method goes beyond simply using leader transitions as breakpoints, as the example of Cameroon illustrates. The nation has had only two de facto leaders in its post-independence history – Ahmadou Ahidjo and Paul Biya – yet our methodology divides it into ten distinct political periods. Overall, this fine-grained periodization system results in strong temporal variation within our variables, allowing for greater precision in the analysis.

Main Survey

Having established the temporal units of analysis, in Phase II of the survey we asked experts to characterize in more detail all periods that lasted at least two years and did not involve a full-scale civil war, using a set of twenty one mostly closed-ended questions, resulting in a total of ninety five raw indicators. Implemented using the online survey tool Qualtrics, the survey took experts an average of a full day to complete and included five sections.

Table 1 provides an overview of all indicator-based variables, divided into sections (and those variables available for each bloc marked by an asterisk). The indicators in the first section aim to capture different aspects of the configuration of power in a political settlement, such as the size of different blocs, their relative power, as well as their internal cohesion and hierarchical power distribution. Two important qualitative variables in this section consist of our coders’ lists of the most powerful and powerless groups in each bloc for each country-period, which could be useful for qualitative researchers as well as for additional coding. In contrast, the variables in Section II are designed to capture the relationship of each bloc to the settlement. Specifically, we have created variables that measure the importance of various strategies (e.g., violent repression, clientelistic material cooptation, or universalistic ideological legitimation) to incorporate a bloc’s followers or leaders into the settlement, as well as variables describing how (un)equally material benefits are distributed across and within blocs. Section III is an extension of the first section in the sense that here we asked experts more directly about different degrees of power concentration in a country’s leader. Section IV then moves away from the more classical power or bloc-related variables and looks at the role of threats and foreign support. That is, two variables capture the extent to which foreign military and/or financial and technical support has been critical in maintaining the political settlement. And several variables capture the extent to which various groups (such as the urban subordinate classes, the military, or a neighboring country) were perceived by a country’s leadership as posing a significant threat to its political or physical survival. Finally, since a prominent strand of PSA treats the economic capabilities and political power of manufacturing firms as important additional variables in explaining economic policymaking capacity and development outcomes, we added these indicators in Section V. Notably, we also constructed 107 other indices based on our raw indicators, such as Khan’s two dimensions of the configuration of power of ruling coalitions, which are presented in more detail in the online appendix codebook.

Table 1. Overview of PolSett indicators.

Variables	Scale
Section I: The Settlement’s Configuration of Power
Population share (in %)*	0–100%
Relative power*	1 (Powerless)–5 (Extremely Powerful)
Likelihood CLB splits from government	1 (High)–3 (Low)
Likelihood OB joins government	1 (None)–4 (High)
Most powerful groups*	Text/List
Relatively powerless groups*	Text/List
Percentage of relatively powerless*	0–100%
Hierarchical power concentration*	1 (Power rests w. ordinary members)–6 (Power rests w. high-level leaders)
Cohesiveness*	1 (very incohesive)–4 (very cohesive)
Section II: Blocs’ Relationship to the Settlement
Importance of Strategy to Incorporate Bloc Leaders OR Followers into/under the Political Settlement:* (a) Violent repression; (b) Non-violent repression; (c) Clientelistic material co-optation; (d) Clientelistic non-material co-optation; (e) Programmatic material legitimation; (f) Universalistic ideological legitimation; (g) Procedurally democratic legitimation	1 (not important)–4 (very important)
Relative material benefits received*	1 (Very low)–5 (Very high)
Within-bloc egalitarian material distribution*	1 (Massively inegalitarian)–5 (Egalitar.)
Section III: Decision-making and Implementing Power of the Leadership
Concentrat. of policy-decision-making power in leader	1 (high. dispersed)–4 (high. concentrat.)
Concentrat. of policy-implementation power in leader	1 (high. dispersed)–4 (high. concentrat.)
Section IV: Foreign Support and Internal & External Threats
Foreign military support importance Foreign financial and technical support importance Political OR Physical Threat to Top Leadership from: (a) Rural subordinate classes; (b) Rural dominant classes; (c) Urban subordinate classes; (d) Urban dominant classes; (e) Ethnic, regional or religious groups; (f) The military; (g) Opposition group in exile; (h) Neighboring country; (i) Non-neighboring country	1 (not important)–4 (very important) 1 (not important)–4 (very important) 1 (No threat)–4 (High threat)
Section V: Economic Organizational Power
Manufacturing firms’ capabilities	1 (Low)–3 (High)
Manufacturing firms’ political power	1 (Low)–3 (High)

Variables marked with an asterisk are disaggregated by bloc.

Recognizing the sometimes ineffable nature of power and related variables, and that not all experts will have the same knowledge for all periods and questions, coders were asked to record their level of confidence in their answer for each question-period (the average coder being “fairly confident” in their responses). While not a fail-safe method for eliminating error and bias, we feel this provides some indication of where the evidence is stronger or weaker and a safeguard against making exaggerated claims for our data. For a more detailed discussion and overview of our different reliability measures – including an overview of confidence ratings and relative standard deviations across questions, countries and time – please see Section 2.1.3 of the online appendix.

In addition, we used these confidence ratings as partial weights when aggregating the country-coder-variable-period scores into our final individual country-period scores. Specifically, we used a weighted arithmetic mean to aggregate individual responses, where the weight for each individual expert’s score was lower the less confident they were in their assessment and the further their coding was from the mean score of all three experts.⁵ A more detailed description of the aggregation process, including a worked example and an overview of the different variable versions, can also be found in Section 2.1.3 of the online appendix.

Lastly, it should be noted that achieving cross-country intercoder equivalence in expert surveys poses challenges given the lack of a benchmark common to all coders. While employing vignettes and consistent coders across countries is ideal, resource constraints and the length of the survey limited their implementation in our project. Instead, coders were provided with careful question wording, detailed notes, and empirical examples, informed by several prior rounds of discussion and piloting with experts for very different country contexts. Additionally, rigorous validation of cross-country patterns was conducted by affiliated scholars. Despite these efforts, comparisons across cases are prone to errors, thus including country fixed-effects in models is recommended for more reliable estimates.

Validity

Several measures were taken to ensure and test the validity of the data. After submission, we checked each survey for missing values and common misinterpretations that we had identified in previous coder responses. Second, after all coder responses had been received, cleaned and aggregated, we asked seven regional experts from within and outside ESID to systematically check whether the temporal movements and cross-country position of their countries of expertise appeared to be consistent with their broader research findings. In doing so, we provided them with theoretical and empirical benchmarks to increase intercoder equivalence across countries. Coder-level replies for the countries assessed as outliers were then inspected and coders asked to respond.

Finally, following the final review by coders, we used additional statistical and case-based methods to test the construct, content, and convergent/discriminant validity of the data. Testing convergent validity was not straightforward given the paucity of closely related external variables. Nevertheless, Table 2 illustrates some of the pairwise correlation tests (Cohen 1988) we conducted for indicators against existing measures, primarily derived from Coppedge et al. (2022). Since we are comparing conceptually related but distinct variables, we expected to find moderate correlations.

Table 2. Correlational tests of concurrent validity for selected variables.

Section I & III	Regime support groups size	Ruling coalition size (Vogt et al. 2015)	Polity2 (Marshall and Jaggers 2020)	Personalism (Geddes, Frantz, and Wright 2018)
Population share (in %) of LB	0.41***	0.34***	0.09***	0.07***
Relative power of LB	−0.06***	−0.04*	−0.34***	0.17***
Concentration of policy-decision-making power in leader	−0.20***	−0.05**	−0.43***	0.27***
Section II	CSO repression	Ideology	Clientelism Index	Power distributed by socioecon. position
Procedurally democratic legitimation for leaders of OB	0.58***	−0.28***	−0.04*	0.06***
Violent repression for leaders of OB	−0.62***	0.30***	0.15***	−0.16***
Universalistic ideological legitimation for followers of LB	−0.20***	0.45***	−0.36***	0.31***
Clientelistic material co-optation for leaders of LB	0.22***	−0.15***	0.41***	−0.24***
Relative material benefits received by LB	−0.17***	0.02	0.36***	−0.47***
Within-bloc egalitarian material distribution of LB	0.14***	0.15***	−0.52***	0.44***
Section IV & V	Active regime opposition by ethnic or racial group	Regime opposition by working class	Regime support by business elites	State ownership of economy
Political threat by ethnic, regional or religious groups	0.36***	0.05**	−0.08***	−0.12***
Political threat by urban subordinate classes	0.03	0.30***	0.21***	0.21***
Manufacturing firms’ political power	−0.07***	0.12***	0.38***	0.52***

*p < .10, **p < .05, ***p < .01. Unless otherwise specified, all external variables (in italics) are derived from V-Dem (Coppedge et al. 2022).

Indeed, this is mostly what we find. For example, our measure of LB population share is moderately positively correlated, at 0.41 and 0.34, with the size of V-Dem’s regime support group and the aggregate population size of the ethnicities represented in a ruling coalition (Vogt et al. 2015) respectively. This is despite the fact that both variables are quite different from our own. While our measure looks only at supporters of the LB, the V-Dem measure asks coders to assess the size of all powerful supporters of a regime, and the EPR focuses on the ethnicities that make up the entire ruling coalition.

Looking at our more power-related variables, we find that, as expected, both the power of the LB (to singlehandedly change or prevent changes to the settlement) and the concentration of political decision-making power in the leader decrease as the regime becomes more democratic, characterized by correlations of −0.34 and 0.43 with Polity2. This correlation also remains when visually comparing the levels of power of the three blocs against Cheibub, Gandhi, and Vreeland (2010) six regime types, as done in Figure 2 on the left. As expected, LBs in dictatorships are on average more powerful than their democratic counterparts. At the same time, the kernel density plot of LB power across regime types in Figure 2 on the right also confirms that power configurations and regime types are not congruent. That is, consistent with PSA assumptions, there is considerable variation within each regime type in terms of existing levels of power, as well as significant overlap with other types (and is consistent with findings from other datasets, e.g. Gandhi and Sumner (2020)). For example, 31 percent of parliamentary democratic observations in our sample have higher LB relative power scores than the median civilian dictatorship. For instance, while the LB under Thai Prime Minister Prem Tinsulanond from 1980 to 1988 or Ecuador under President Correa from 2007 to 2010 was very powerful (5 and 4.7, respectively), the civilian dictatorships of Préval in Haiti from 1999 to 2000 and the military dictatorship of Al-Hafiz in Syria from 1963 to 1965 had comparatively weak LBs (at 2.7 and 2.8, respectively). Thus, our data suggest that researchers should exercise caution before using existing categorizations of regime types as proxies for the variables more directly captured by our dataset.

Figure 2. Relative power of blocs across regime types (left) and kernel density distribution of LB Power across regime types (right).

We also examine how our power measures compare with Geddes, Frantz, and Wright (2018) personalism index, which aims to capture the extent to which dictators have concentrated power in their hands at the expense of other members of the ruling elite. While the correlations are significant and positive as expected, the coefficients are fairly small at 0.17 and 0.27 respectively. We believe this to be primarily due to conceptual differences. First, both our measures view power as relative to the opposition as well, in contrast to the personalism index, which explicitly excludes the opposition. Second, our LB measure does not focus on the leader’s personal relative power, but on the leader and his or her loyal supporters’ relative power. Figure 3 illustrates how this translates into some real-world examples with significant differences between our scores. While leaders like Deng Xiaoping in China’s 1980s, Houphouet-Boigny in Côte d’Ivoire’s 1960s, Abdul Razak Hussein in Malaysia’s 1970s, or Verwoerd and Vorster in South Africa’s 1970s might not have personalized their rule in the sense that they personally controlled the security apparatus or appointments to the party’s executive committee (key indicators of the personalism index), together with their bloc of loyal supporters, they have often been cited in the literature as classical examples for wielding particularly high levels of political power. This is reflected in our data.

Figure 3. Personalism index and PolSett indicators compared. Both PolSett measures have been transformed to a 0 to 1 scale for better comparability.

We find stronger levels of convergence for our variables in our subsequent sections of the dataset. For example, we find that governments that are less repressive of civil society organizations according to V-Dem coders are strongly negatively associated (–0.62) with our measure of violent repression of OB leaders. Similarly, higher political power of manufacturing firms, as determined by our experts, correlates positively with V-Dem’s measure of business elites’ support for a regime (0.38) and inversely with state ownership of the economy (0.52). This aligns with the notion that governments find it easier to control firms they own.

Overall, the analysis shows reassuring levels of convergent and discriminant validity. Concurrently, the difficulty in finding closely related measures further demonstrates, we believe, the novelty and added value of our dataset.

Patterns and Trends

While we believe that the primary use of our dataset is hypothesis testing, demonstrating this here in the required depth is beyond the scope of this dataset introduction and has been attempted in greater detail elsewhere (Kelsall et al. 2022). Rather, we wish to illustrate some of the variety of descriptive insights that can be gained from using our data, with a particular focus on our bloc incorporation and threat-related variables. To this end, in Figure 4, we first take a closer look at the significance of various selected methods of incorporating LB and OB followers into or under the political settlement over time. Several notable findings emerge. First, clientelistic material cooptation has been the most important form of incorporation of LB followers throughout post-independence history. Interestingly, this is consistent with Khan’s suggestion that political settlements in developing countries are fundamentally clientelistic in the way politics operates. Interestingly, universalist ideological legitimation seems to move in the opposite direction to clientelist incorporation, with the former declining since the 1970s and the latter increasing. This makes sense in two ways. First, ideological claims to leadership were strongest in much of the Global South in the 1950s to 1970s, particularly in the form of nationalist independence movements and pro- or anti-communist positioning. Second, it makes sense that when ideological claims fail to attract followers, clientelistic means become an important alternative for holding coalitions together. It is also interesting to note that the importance of clientelism has not diminished with the rise of democratization in the Global South. This seems to cast doubt on the assumption that leaders with larger “selectorates” are less reliant on the provision of material incentives to maintain the loyalty of their followers (Bueno de Mesquita et al. 2003).

Figure 4. Importance of methods of incorporating LB and OB followers across time.

In contrast and less surprisingly, the use of violent repression and of democratic legitimation to incorporate opposition followers into or under the settlement closely correlate with democratization. Until around the mid-1980s, violent repression was a particularly relevant mode of incorporation. This declined significantly with the third wave of democratization, during which democratic legitimation became an equally important method of incorporating opposition actors (although this trend seems to have reversed somewhat in the early 2010s, potentially supporting discussions in the literature on democratic backsliding).

Finally, Figure 5 shows the evolution and levels of how leaders perceive different social groups as threats to their political survival over time. Although the trends are more stable, some patterns and (non-)developments are striking. For instance, we see a significant decline in the threat posed by the military in the Global South, in line with the significant decline in coups since the 1980s. Another notable pattern is the relatively low threat level of rural groups compared to their urban counterparts, consistent with findings from the urban bias literature. Interestingly, however, and contrary to the expectations of this literature, this pattern has not changed much with democratization. Although peasants now have the vote, on average they do not appear to have become more threatening to leaders vis-à-vis the urban classes.

Figure 5. Political threats from societal groups across time.

Conclusion

Questions about whether and how different configurations of power have shaped the development of societies in the Global South have long preoccupied the social sciences. Until now, however, a lack of adequate data has made it difficult to test the validity of competing frameworks and hypotheses across time and space. Complementing other recent datasets, the Political Settlements dataset represents an important step in filling this gap. Correlation tests and illustrations of historical trends demonstrate not only the validity of the data, but also its added value, in particular the disaggregation of societal blocs.

Ultimately, PolSett promises to be a rich resource for a variety of comparative political analyses. For instance, future research could examine how different power configurations are associated with political and economic outcomes, such as the propensity for coup attempts, civil war onset, democratization, corruption reduction, or economic development. Together with our variables capturing the means by which societal actors are incorporated into a political settlement, our power configuration data should allow for more valid tests of selectorate as well as various social contract theories. Future research could also use the dataset’s variables on the severity of internal and external threats to test, for example, “bellicist” theories of state and economic development (e.g., Doner, Ritchie, and Slater 2005). And further back in the causal chain, researchers could examine what predicts the emergence and maintenance of powerful leaders and ruling coalitions. In closing, we hope that researchers will find the data useful in better understanding how different configurations of power, incorporation, and threat have shaped political economies in the Global South.

Acknowledgments

We would like to thank Whitney Banyai-Becker and Anthony Bladen for their excellent research assistance. We are also grateful to Julia Brunt and Julie Rafferty for their administrative support. We are indebted to Kunal Sen, Samuel Hickey, and David Hulme, who provided critical substantive and methodological input as well as the opportunity to conduct this project at ESID. We would also like to thank the participants in numerous workshops who helped us to refine the survey design, validate expert responses and interpret initial findings. We are also grateful to Carl Henrik Knutsen and Bastian Herre for their advice on publishing large-scale political datasets. Finally, our particular thanks go to the 129 country experts who agreed to share their knowledge and time with us to produce this dataset.

Disclosure statement

The authors report there are no competing interests to declare.

Data availability statement

The dataset and replication files of this study are openly available in the Harvard Dataverse at https://doi.org/10.7910/DVN/UYPEWU.

Additional information

Funding

The research for this paper was conducted as part of the Effective States and Inclusive Development Research Centre (ESID) at the University of Manchester, UK, a project funded by UK Aid from the UK government (Department for International Development, grant no. PO 5113).

Notes

1 It does not tell us, for example, whether an ethnic group formally excluded from access to political power, for example the Chinese Uighurs, nevertheless represent a credible threat to the Han Chinese monopoly on power.

2 As of March 2024, Khan’s seminal (2010) paper had 1051 citations on Google Scholar, which if anything underestimates the extent of its influence.

3 Given our limited resources, we had to choose between coastal and landlocked countries. We opted for the former, larger group.

4 Please see Section 2.1.2.1 in our online appendix codebook for more details on the periodization process.

5 In the very rare instances where data points are based on fewer than three replies, intercoder-distance based aggregation becomes ineffective. Users are encouraged to exclude such data points from their analyses, facilitated by a variable indicating the number of replies for each country-year-variable.