Population Age Structure and the Vulnerability of States to Coups d’État

4.1 Temporal Extent of Data

At the outset of the study, the data chosen for analysis ranged over a 50-year period, from 1970 – a year when most overseas colonies of Western powers had achieved independence – to 2019. Exploratory analyses indicated that, to generate a model for the purpose of forecasting, data should cover only the post-Cold War period (discussed in more detail below, see Separating Cold-War and Post-Cold War Relationships).

4.2 Sample of States

As a matter of consistent practice, the list of recognized independent political entities is drawn from the United Nations. From this list, two types of entities were excluded from the analysis: (1.) non-independent political entities (e.g. Palestine, Western Sahara) whose state behavior may be constrained or induced by an occupying power; and (2.) independent states with a population under 500,000 (thus, currently excluding Belize, Iceland, Brunei, and numerous small island states).

A substantial number of states have entered the active data set between 1970 and 2019. These include several states became part of the data pool on the year that they surpassed a population of 500,000 (e.g. Bhutan, Cape Verde, Djibouti, Equatorial Guinea, Solomon Islands), and a large group of newly independent states (including Eritrea, former-Soviet republics, former-Yugoslav republics, Slovak Republic, South Sudan, and Timor Leste). Thus, the annual active data set has grown from 134 (in 1970) to 168 (in 2019).

The use of states as the unit of analysis has several analytical limitations. The country-level median age may obfuscate the presence of significantly populous minorities who display demographic dynamics differing substantially from the majority. Even when the country-level age structure has matured, minority–majority demographic differences can be associated with ethnic tensions that effect political stability (Leuprecht 2010).

4.3 Coup d’État Data

Annual data identifying the year of a successful coup d’état (a coup-year) were drawn from coup-related event data compiled by the Cline Center Coup d’État Project Dataset, Version 2.1.2 (Cline Center for Advanced Social Research 2023; Peyton et al. 2020). Multiple coups within a year were counted as a single coup-year.

4.4 Median Age

For these cases, annual values of median age were drawn from annual estimates published in the 2022 revision of the UN Population’s Division’s World Population Prospects database (UNDESA 2022).

4.5 Additional Data

Because the behaviors of two relatively small groups of countries have often deviated significantly from the age-structural patterns demonstrated by the majority of countries (Cincotta 2017; Weber 2013), it has been useful to statistically account for each of these groups as independent (dummy) variables in logistic regression analyses. The first statistically noteworthy group, hereafter referred to as resource-reliant states, includes all country-years of data in which the annual value of combined oil and mineral rents, reported by the World Bank’s World Development Indicators Database (World Bank 2022), exceeds 20.0 percent of GDP. The other group, hereafter referred to as the least populated states, includes all country-years of data that, according to the most recent UN Population Division estimates (UNDESA 2022), have less than 5.0 million residents.

In addition, data from the six Worldwide Governance Indicators (WGI) were the focus (outcome variables) of secondary analyses in this research and are discussed in more detail in a later methodological section (see Secondary Analyses).

4.6 Exploratory Analyses

To begin this study, various graphic and statistical exploratory analyses were undertaken to determine if an age-structural pattern could be discerned in coup d’état data, and if so, to detect aspects of this pattern that could influence modeling and forecasting. To limit the spikes and troughs that are common in political-event data, the frequency of coups d’état was retabulated using five-year aggregates of data (1970–74, 1975–79, … , 2015–19). For each five-year period, countries were grouped into discrete age-structural phases (youthful, intermediate, mature, and post-mature) according to a previously published median-age-based, four-phase schema (NIC 2012; see Figure 2), and the coup frequency for each age-structural phase was calculated and graphed for each five-year period, from 1970–74 to 2015–19.

4.7 Separating Cold-War and Post-Cold War Relationships

These exploratory analyses and initial attempts at modeling (not detailed in this essay) yielded three findings that influenced further analyses: (a.) relative to all other segments of the age-structural transition, the countries with the most youthful populations within the youthful phase appeared to be the most vulnerable to successful coups d’état (Figure 3); (b.) coup vulnerability across all phases of the age-structural transition declined precipitously after the end of the Cold War (see Figure 3); (c.) the 1990–94 period appeared as a statistically ambiguous period separating coup data occurring during the Cold War (1970–1989) from coup data during the post-Cold War years (1995–2019).

Figure 3:

The five-year frequency of coups in the youthful phase and more mature phases (bars, vertical scale on the left), and the count of countries (points, vertical scale on the right) in the early segment of the youthful phase (median age 20.5 years or less). Data are from the Cline Center Coup d’État Dataset, Version 2.1.2 (Peyton et al. 2020) and UN Population Division demographic estimates (UNDESA 2022).

Three adjustments were made to accommodate these findings: (a.) in graphics, the youthful phase (median age 25.5 years or less) was disaggregated into early-youthful (median age 20.5 years or less) and late-youthful (median ages from 20.6 to 25.5 years) (see Figure 2); (b.) individual age-structural models were generated to capture the relationship over the course of data during the Cold War years (1970–1989), and the post-Cold War years (1995–2019); (c.) forecasting was conducted using a fitted form of the post-Cold War model, which omitted data from the 1990–94 interim period.

4.8 Primary Analysis

Consistent with the findings of previous exploratory analyses (discussed above), models were generated for both for data during the Cold War (1970–89) and post-Cold War period data (1995–2019). In these analyses, each annual country datum included: (a.) a dichotomous coding of the (dependent) outcome variable, Y, which records the presence ( y  = 1) or absence ( y  = 0) of a coup d’état; (b.) a continuous, quantitative observation of the status of the domain variable, m, measured in years of median age; and (c.) dichotomous codings of two independent (dummy) variables, x ₁ (resource-reliant states), and x ₂ (the least populated states). Thus, the iterative logistic regression algorithm is tasked with fitting parameters for the logit’s constant, k; the pitch, a, of its continuous domain variable; and the coefficients, b ₁ and b _2, of its two standard dichotomous independent variables (Eqs. (1.1) and (1.2)).

Notably, age-structural modeling’s generalized logit provides terms for experimentation with additional dichotomous independent variables, x _j . These can be used to identify other groups or conditions that might differ probabilistically from the resultant age-structural function, yielding a more accurate and informative forecast. For example, during the exploratory analyses of coups d’état, a dummy variable was added to differentiate Cold War from post-Cold War cases. The results further demonstrated the need to generate an individual model for each period and to omit the 1990–94 period.

4.9 Logistic Curve Fitting

When fit to a dichotomously coded set of country-level observations, where median age is the only continuous independent variable in the logit (Eq. (1.1)), the logistic regression algorithm (see Menard 2002) yields (by iteration) an expected outcome, which is read as the probability (from 0.0 to 1.0) of observing a successful coup d’état (i.e. the condition, y  = 1) at any median age across the age-structural domain, M. When graphed across M, the resultant probabilistic age-structural function, P( y  = 1) (Eq. (1.2)), assumes the form of a simple logistic curve – a positively or negatively sloped sigmoid monotonic function with its inflection point (the maximum of its first derivative) fixed, a priori, at a probability 0.50.

Ideally, an age-structural function’s inflection point falls within the boundaries of M, providing a theoretical indication of the general region of the domain at which analysts should expect the greatest net influx of countries into the outcome condition, y  = 1. However, the logistic regression algorithm can also generate, as a best fit to data, a partial segment of a logistic curve (i.e. a positively or negatively sloped tail). In the following analysis of the probabilistic age-structural timing of a successful coup d’état, the functional form that appears within the boundaries of M, is an example of a partial segment – part of the right-hand tail of a negatively sloping logistic curve, which is most elevated in the most youthful portion of M. The left-hand side of this curve, which includes its inflection point, falls below the domain’s lower boundary.

4.10 Linearization, Model Adjustment, and Forecasting

Linearization of a logistic post-Cold War age-structural model produces a model that can be easily adjusted to improve the fit to data, and ultimately be used by analysts to make forecasts for groups of countries of particular interest (see Cincotta and Weber 2021). The linear model (Eq. (2.1)) predicts the annual expected total of successful coups, E( y  = 1), by summing the probabilistic contributions from all countries that fall within each of the five discrete age-structural phases (eYTH, lYTH, INT, MAT, PMT; see Figure 2). Each phase’s contribution is the product of its coefficient of probability (p _i) and the annual count of countries in that phase (n _i).

To begin fitting the linear version of the post-Cold War model to observed data, its five coefficients of probability, p _i , were each initialized using discrete probabilities that were generated by the logistic post-Cold War model for each centroid (measured in median age) of the five corresponding age-structural phases. These median-age centroids were multiplied by 5 (to forecast over a five-year period) and then adjusted to a value, g _i, in order to maximize the average χ ² goodness-of-fit probability (α = 0.05) across all five-year previous post-Cold War periods (1995–99, 2000–04, … , 2015–19).

To forecast the expected number of coups for each future five-year period, the fitted linear model (Eq. (2.2), see Results) was extended into four future five-year periods (2020–24, 2025–29, 2030–34, and 2035–39) by using median ages from the UN Population Division’s most recent medium scenario (UNDESA 2022). To portray uncertainty surrounding this expected future scenario, upper and lower 0.95 confidence intervals (CIs) were computed (n = 165 countries) using the logistic model’s median-age-specific confidence intervals (upper CIs were rounded up to the nearest whole-number country count; lower CIs were similarly rounded down).

4.11 Secondary Analyses

Age-structural modeling of each of the World Bank’s (2022) six Worldwide Governance Indicators (WGI) (government effectiveness, control of corruption, rule of law, political stability and absence of violence/terrorism, voice and accountability, and regulatory quality) followed the logistic regression protocol outlined for the primary analysis (see Primary Analysis).

According to accompanying methodological notes, WGI scores are generated from the responses obtained in surveys of enterprises, citizens, and experts that are collected by survey institutes, think tanks, non-governmental organizations, international organizations, and private sector firms (Kaufmann, Kraay, and Mastruzzi 2010). Scores for each WGI indicator range approximately between −2.5 (indicating weak), and +2.5 (indicating strong). For each indicator, the annual mean for the countries used in this analysis has typically ranged slightly below zero (between −0.20 and 0.0).

For each governance indicator, a dichotomous outcome variable was generated to represent the set of negative (mostly weak) assessments. WGI data for this analysis span the temporal period from 1996 to 2019, with years 1997, 1999, and 2001 being absent from each data set. Similar to the primary analysis, data for each of the age-structural models of the six governance indicators were limited to independent states with more than 500,000 residents, which yielded a country count from 157 (in 1996) to 167 (in 2019). Annual estimates of median age were used as the continuous domain variable in each of the six logistic regressions. Similar to the primary analysis, the logit of these regressions included the two standard dichotomous independent variables (the set of resource-reliant states and the set of least populated states).

6.1 Non-Youthful Countries

Not only do countries in the transition’s intermediate and mature phases undergo only about one-fourth the risk of a coup as those in the youthful phase, later-phase coups tend to be qualitatively different. For example, coups in the intermediate and mature phases have been nearly twice as likely as those in the youthful phase to be instigated by a large-scale popular uprising (Cline Center for Advanced Social Research 2023; Peyton et al. 2020) – a so-called color revolution, such as Georgia’s Rose Revolution in 2003, Tunisia’s Jasmine Revolution in 2011, or Algeria’s Revolution of Smiles in 2019. Similarly, later-phase coups have been twice as likely as those in youthful countries to be successfully conducted by a faction within the ruling government (known as a palace coup).

6.2 Explanatory Narratives

If, as some analysts assert (Aboagye 2023; Jagne 2023), coups generally displace governments that fail to achieve legitimacy, why do coups occur so frequently in countries with very youthful populations? Moreover, in a system featuring numerous triggers, and a list of possible proximate and structural risk factors (Belkin and Schofer 2003; Hiroi and Omori 2013) – some of which vary between regions, among neighboring countries, between regime types, and from one chronological time period to the next – how can an age-structural model circumvent this causal complexity, yet identify the most coup-vulnerable states?

The short answer to both questions is “age-structural timing”. Evidence discussed in the two following sub-sections are intended to demonstrate that both the supply of governance, and demands placed on governance, tend to peak in the most youthful segment of the age-structural transition.

6.3 Supply of Governance: Indications from the WGI

Results of logistic regression analysis (see Appendix, Statistical Table 2) on all six scored Worldwide Governance Indicators (WGI) – government effectiveness, control of corruption, rule of law, political stability and absence of violence/terrorism, voice and accountability, and regulatory quality – share a strikingly similar relationship with median age. In each of their graphic functional forms (Figure 9), the probability of a below-zero score (more weak than strong) is highest in the youthful portion of the age structural transition. Notably, each of these governance trends reaches a 0.50 probability of having an above-zero assessment – and thus the transition to a positive score – near the middle of the demographic window.

Figure 9:

Age-structural functions and 0.95 confidence intervals (CI) for each of the World Bank’s Worldwide Governance Indicators (Kaufmann, Kraay, and Mastruzzi 2010; World Bank 2022): (a.) control of corruption, and government effectiveness; (b.) rule of law, and absence of political instability and violence/terror; and (c.) voice and accountability, and regulatory quality.

Whereas this repeated age-structural pattern may surprise political scientists, it is wholly consistent with the conclusions of computational studies of lifecycle consumption, income, and public and private transfers, which draw on data from the National Transfer Accounts Project (UNDESA 2013, pp. 16–19; Lee and Mason 2011). In the absence of resource rents and external sources of income (such as remittances, foreign loans and grants) of substantial per-capita value, taxation in youthful countries and their resultant government budgets are generally constrained by low fiscal support ratios (the ratio of the effective number of taxpayers to beneficiaries, weighted by their respective contributions). Thus constrained, youthful countries generally have limited fiscal resources to build and maintain their institutions – i.e., fund their operations and services, train service providers, and exercise quality control. Yet, their government’s legitimacy depends on these institutions.

6.4 Demand for Governance: Implications of a Youth Bulge

Since its earliest elaboration (Möller 1968/69), proponents of the youth-bulge narrative have focused on the ease of mobilizing idealistic, identity-seeking, risk-taking young adults in youthful societies to engage in political violence as members of non-state or state-sponsored organizations. Researchers have generally assumed that the risks of recruitment, particularly among young males vying to enter the workforce, would be highest where alternative avenues for economic and social mobility are perceived as limited by the large and rapidly growing numbers of competitors and by elite capture of jobs ((Huntington 1997; Goldstone 2002). While the preponderance of evidence indicates that young men have been more prone to acts of violence than either older males or women (Daly and Wilson 1988; Simon and Blaxter 1989), and more likely to participate in coalitional violence (Archer 1994), the youth-bulge narrative – like many other mechanistic social narratives – has been difficult to directly test and conclusively demonstrate.

To operationalize this narrative, its proponents have hypothesized that in countries having populations with large numbers of young adults, relative to a measure of the adult population – i.e., countries in the youthful phase of the age structural transition – governments are expected to face an elevated vulnerability to political violence. Several studies have found this expectation to be statistically validated by available data (Mesquida and Wiener 1999; Urdal 2006). Moreover, this demographic operationalization has succeeded in opening the youth-bulge narrative to further refinement, testable criticism, and to use in national and sub-national studies of criminal violence.

Whereas, in statistical competition with other indications of low levels of socioeconomic development, youth-bulge measures have shown only mixed results in national and sub-national studies of violence (see Corona Juárez, Urdal, and Vadlamannati 2022), this demographic measure has demonstrated an ability to identify high-risk groups of countries. For example, as part of a lengthy critique, a critic of the youth-bulge narrative questioned the validity of this country-level hypothesis (Sommers 2011), noting that “most of the recent wars are over–Angola, Burundi, Liberia, Mozambique, Rwanda, Sierra Leone, South Sudan … despite all of them taking place in countries that still have youth bulge demographics.” Yet, regardless of peaceful pauses and the demobilization of insurgents in several of these seven states, four returned to some form of civil conflict within five years (from 2012 to 2016). After 10 years, this group has fared even worse. All have remained youthful, and five of “Sommers’ seven” have experienced multiple years of civil conflict during this extended period (2017–2021). According to UCPD/PRIO data (2023), only Liberia and Sierra Leone remain conflict-free since 2011.