1 Introduction

Sub-Saharan Africa (also referred to as SSA or just Africa) leads in several rankings: it is the world’s poorest region, it has the fastest-growing population, and it has the worst health outcomes. Data by the World Bank show that average GDP per capita (PPP-adjusted, in constant 2011 international \(\$\)) in Africa was about 30\(\%\) of the world average in 1990 (when the earliest data are available). This share has constantly decreased to only 22\(\%\) in 2018.Footnote 1 This development is accompanied by fast population growth. In 1960, SSA made up about 7\(\%\) of the world population. In 2018, that share has almost doubled.Footnote 2 Another issue is the health situation that has worsened over the last half century. In 1960, the average child mortality rate of the SSA countries was 40\(\%\) higher than the global average. In 2018, it was 100\(\%\) higher. In this study, we explore whether democracy provides a solution to this situation.

We estimate the within country effect of democracy on economic growth for countries in SSA. In our analysis though, we consider almost all countries in the world and the empirical model explores information for all of them in one and the same regression. As we discuss below, this method is beneficial because it enables comparisons between African and non-African countries estimated under the same econometric conditions. This approach provides deep insights for the development of Africa. Our estimation begins with fixed effects OLS estimates of the correlation between democracy and GDP per capita. The baseline estimates show evidence for a positive association. However, the positive link is significant only in the non-African countries. In the SSA countries, estimates turn out insignificant suggesting that democracy plays no important role for income growth. This pattern is supported by different measures for democracy.

The absence of significant correlation between democracy and income has to be taken with caution since the relationship is driven by endogeneity. Fixed effects models provide only some relief as they control only for the effects of those variables that are time-invariant such as geography, history, and institutions.Footnote 3 However, the effects of time-variant variables remain and they can bias estimates. In addition, there is reverse causality since income is also likely to have an effect on democracy (see, e.g., Cervellati et al. (2014) and Brückner and Ciccone (2011)). We address endogeneity with two methods: first, we employ the GMM technique developed by Arellano and Bond (1991), and second, we apply an instrumental variable (IV) method based on important work by Acemoglu et al. (2019). The methods are based on different identification assumptions (we discuss them below). Yet, both techniques confirm no significant positive effect of democracy on GDP per capita in the African countries. Although none of the methods is perfect in solving the endogeneity problem, they are helpful because in sum, they draw a clearer picture of the true effect.

Figure 1 gives an illustration of the issue. It shows the development of income growth around a democratization for countries that had a single transition to democracy without a reversal to autocracy. This restriction on the countries enables to depict the development of income before and after a democratization. As can be seen, both groups of countries experience some growth from a democratization. However, the gains are not significant for the SSA countries. The empirical analysis confirms this conclusion.

Fig. 1
figure 1

Average \(\%\)-change in GDP per capita around a democratization only for the countries that had a single transition from autocracy to democracy without a reversal. The identification of a transition is based on the dichotomous measure for democracy provided by Acemoglu et al. (2019). More information is given in Sect. 2

We also estimate the effect of democracy on total GDP. This is an important step because it provides insight how democracy affects GDP per capita. These estimates show that no countries’ GDP benefits from democracy. When democracy does not support total GDP but it increases GDP per capita in the non-African countries, the denominator of GDP per capita has to play an important role. We explore this theory by estimating the effect of democracy on the size of the total population. The estimates reveal an interesting pattern as they show that democracy has a decreasing effect on the population size in the non-African countries whereas in the African countries, it has an increasing effect on the total size. The relationship between population size and economic growth is the subject of a large body of work that has not concluded yet. On the one hand, a larger population means more people, i.e., brains that are more likely to develop advanced technology thus leading to more income growth. On the other hand however, the famous Malthusian theory postulates a negative effect since the contribution of each additional labor unit decreases due to limited resources (see Peterson (2017) for a discussion of the literature). Against this background, we explore how democracy affects the age structure of the population.Footnote 4 The estimates show that democracy decreases the share of young population (age 0–14 years) in the non-African countries whereas it increases that share in the African countries.

This finding provides an explanation why democracy does not benefit income growth in Africa. Albeit, it also raises the question about the reason why the demographic figures react so differently to democracy? We find evidence for an explanation related to child health since democracy shows not to improve the child mortality rate in the African countries. In the non-African countries though, democracy leads to significant improvements. Figure 2 provides some illustrative support for this explanation. It shows the development of the child mortality rate for those countries that had a single democratization without a reversal to autocracy. The runs show that in the non-SSA countries, mortality decreases significantly after a democratization. However, in the SSA countries, a transition to democracy is shown to have no distinct effect on the health condition of children. The rate decreases in an almost linear way casting doubt as to whether the regime change plays any important role.

Fig. 2
figure 2

Average \(\%\)-change in the child mortality rate around a democratization only for the countries that had a single transition from autocracy to democracy without a reversal. The identification of a transition is based on the dichotomous measure for democracy provided by Acemoglu et al. (2019). More information is given in Sect. 2

Although we cannot rule out alternative explanations, the evidence suggests that since democracy has no beneficial effect on child health in Africa, it leads to a larger share of young people in the population.Footnote 5 The youth in turn represent a less productive part of the labor force. Thus, their contribution to economic performance is relatively small, if any. Consequently, democracy shows to have no significant effect on total GDP. At the same time, democracy shows to increase the size of the total population. In combination with no significant effects on total GDP, the pattern explains why democracy has no beneficial effect on GDP per capita in the African countries.

Our key contribution is the mechanism through which democracy affects GDP per capita. Existing research highlights channels such as investment, education, and trade (see, e.g., Papaioannou and Siourounis (2008) and Acemoglu et al. (2019)). We add the size of the total population to this list. As our estimates show, in the non-African countries, democracy leads to more GDP per capita not because it increases total GDP but because it decreases the total headcount. Both empirical (e.g., Acemoglu and Johnson (2007)) as well as theoretical contributions (e.g., Galor and Weil (2000)) have shown that fast population growth can prevent sustained rise in GDP per capita. Africa is particularly affected by this issue because of its fast population growth rate. Since democracy supports that growth, in its current version, democracy is unlikely to help the SSA countries out of the poverty trap.Footnote 6

2 Data and empirical framework

2.1 Data

We utilize data from different sources. The first category of data concerns information for the political situation in the countries. To this end, we collect data from the Polity IV Project (Marshall et al. (2017)) that includes the Polity 2 index of democracy. The index is a revised 21-scale measure that gauges democracy on the basis of competitiveness of executive recruitment, openness of executive recruitment, constraints on the chief executive, regulation of participation, and competitiveness of participation. The Polity 2 index is meaningful because it measures democracy on multiple intermediate stages ranging from absolute autocracy (−10) to pure democracy (+10). Due to this feature, we refer to it as the continuous measure for democracy. We follow the literature by normalizing the index to range between 0 (absolute autocracy) and 1 (pure democracy). Next to it, we employ the democracy measure by Acemoglu et al. (2019). The authors review numerous existing sources constructing a consolidated index that takes either value 0 (absolute autocratic) or value 1 (pure democratic). Due to its two-stage feature, we refer to the index as the dichotomous measure for democracy. Each measure has advantages and disadvantages. The continuous one reacts to all changes that concern the political process. This feature is disadvantageous when changes do not reflect meaningful changes in democratic institutions. The dichotomous measure in turn is more robust since it is not prone to such noises in the data. However, since it includes less information, it is also less variable. We employ both measures to show the robustness of our findings.

The second source for our data is the World Development Indicators (WDI) database maintained by the World Bank. The database includes information for GDP per capita (at constant 2000 US\(\$\)), size of the total population, share of population in the age of 0–14 years, and child mortality rate (per 1,000 births). We use the information on the former two measures to calculate total GDP (= GDP per capita \(*\) total population). This approach guarantees that we have the same sample for GDP per capita and total GDP (since information for total population is not missing).

Finally, we utilize the instrument for democracy as given by Acemoglu et al. (2019). The instrument implements the idea that political processes across countries affect each other ( Huntington (1991) speaks of “waves of democracy” in that matter). Thus, for a given country, the average democracy score of other countries serves as exogenous variation in democracy. Acemoglu et al. (2019) refine this idea by considering the average democracy score only of those countries that are in the same geographical region (as defined by the World Bank) and that have similar political history. The authors show that the so-constructed instrument is a powerful predictor of democracy. We refer to the instrument as the ANRR-IV.

Overall, our data set includes annual information for up to 175 countries of which 45 are in SSA. The period of analysis is 1960–2010. However, the data set is unbalanced meaning that not all data are given for all countries at any time. Table 3 in “Appendix” provides summary statistics for all the variables employed. It shows that in comparison with the non-SSA countries, the SSA countries are on average poorer (GDP per capita), younger (share of population between 0 and 14 years), and have much worse health conditions (children mortality rate).

2.2 Empirical framework

Our regressions are based on the following dynamic model:

$$\begin{aligned} y_{ct} = \beta D_{ct} + \sum _{j=1}^{p} \gamma _j y_{ct-j} + \alpha _c + \delta _t + \varepsilon _{ct}. \end{aligned}$$
(1)

This equation regresses (the log of) income \(y_{ct}\) on democracy \(D_{ct}\) for country c at time t.Footnote 7 According to the literature (Acemoglu et al. (2019) and Brückner and Ciccone (2011)), changes in the political process are proceeded by economic fluctuations. This dynamic is controlled by including p lags of income on the right hand of the model. The other factors in the equation are a full set of country fixed effects \(\alpha _c\) that absorb the effects of any time-invariant country characteristics, and a full set of year fixed effects \(\delta _t\) that capture the effects of any time-variant shocks that are common across the countries in the sample. The error term \(\varepsilon \) includes all other time-varying unobservable shocks to income. Throughout the analysis, reported standard errors are robust against arbitrary heteroskedasticity and serial correlation at the county level.

Equation (1) represents the standard model to estimate the average effect of democracy on income for a given sample of countries. We can apply it to two different samples of countries (African and non-African). Although intuitive, this approach is not beneficial. By excluding one group from regressions, not all variation in the sample is exploited. In this case, estimates are sensitive to the number of observations in the group (recall that there are only 45 African countries in the sample compared to 130 non-African countries). Moreover, the samples are not estimated under the same (econometric) assumptions since the year fixed effects differ across them. As mentioned above, these effects reflect those time-variant shocks that are common across the countries in the sample. Thus, for different samples, the year fixed effects differ as well. In no case though, do they capture global shocks.

For these reasons, we modify the standard model to a partially interacted model:

$$\begin{aligned} y_{ct} = \beta _{g} D_{ct} + \beta _m \left( D_{ct} * \textit{SSA}_{c} \right) + \sum _{j=1}^{p} \gamma _j y_{ct-j} + \alpha _c + \delta _t + \varepsilon _{ct}. \end{aligned}$$
(2)

This equation is a general version of Eq. (1). It differs by an interaction term between democracy and a country-specific dummy \({SSA}_{c}\) taking value 1 if country c is in the SSA region and 0 otherwise. The understanding of the model is that it estimates the effect of democracy on income in two terms. The first one estimates the effect of democracy that is common across all countries. We highlight this part of the effect by a subscript g clarifying that it represents the general effect of democracy (\(\beta _{g}\)). The second term identifies the marginal effect of democracy (subscript m) that exists only in countries where \({SSA}_{c}=1\). When \({SSA}_{c}=0\) \(\forall \) c, the partially interacted model overlaps with the standard model as in Eq. (1). Otherwise, the effect of democracy results from the accumulation of the first two terms. This can be seen better when we rearrange Eq. (2) as follows:

$$\begin{aligned} y_{ct} = \left( \beta _g + \beta _m * \textit{SSA}_{c} \right) D_{ct} + \sum _{j=1}^{p} \gamma _j y_{ct-j} + \alpha _c + \delta _t + \varepsilon _{ct}. \end{aligned}$$
Table 1 Estimates of the association between democracy and economic growth

This version clarifies that for the non-African countries (where \({SSA}_c=0\)), \(\beta _g\) shows the effect of democracy on income. For the African countries (where \({SSA}_c=1\)), the addition of general and marginal effects shows the cumulative effect of democracy on income. In this way, we exploit all variation in the sample. Moreover, the time fixed effects do not differ across the groups. This model enables fair comparisons across African and non-African countries since effects are estimated under the same econometric conditions (see Cervellati et al. (2014) for the same approach).

3 Main results

Table 1 shows estimates of the association between democracy and economic performance. We begin by estimating Eq. (2) with common panel estimators including fixed effects OLS, Arellano–Bond GMM (Arellano and Bond (1991)), and IV based on the ANRR-IV instrument. With each method, we employ the continuous and dichotomous measure for democracy in alternate order. Moreover, the table shows estimates in two separate panels: in Panel A, the outcome variable is (log) GDP per capita and in Panel B, we measure income by (log) GDP. As discussed above, the model includes several lags of income to control for the dynamics that proceed changes in the political process. The specifications presented here include four lags of the respective income measure. Tables 46 in “Appendix” clarify that higher lags are irrelevant (as in Acemoglu et al. (2019)). To keep clear overview, we do not report the lags in Table 1. Instead, we show a summary statistic (persistence of the growth process) that informs about the joint significance of the lags of the corresponding measure for economic growth.Footnote 8

Column (1) in Panel A shows a coefficient of 0.231 (with standard error 0.438) for the general correlation between the continuous measure for democracy and GDP per capita. For the non-African countries, this estimate suggests that there is no significant relationship between democracy and income growth. For the African countries, the coefficient of the marginal effect is −1.716 (0.755). The negative sign clarifies that it offsets the general effect of democracy leading to a cumulative effect of −1.485. In sum, the cumulative effect suggests that democracy and income growth are negatively correlated in the SSA countries. The coefficient is precisely estimated (the standard error term is 0.671) clarifying that the negative link is significant at the 95\(\%\) level. The specification in column (2) shows a similar pattern. Therein, democracy is measured by a dichotomous index. The coefficient for the general correlation is 1.174 with a small error (0.278) making it significant at the 99\(\%\) level. Yet, the marginal effect is −1.087. Since this coefficient is almost identical in size and negative in sign, the cumulative effect for the SSA countries turns out to be marginal with 0.087. Moreover, the large error term (0.312) makes the coefficient statistically insignificant. Taken together, the correlation analyses provide a first indication that the relationship between democracy and GDP per capita follows a different path in the African countries compared to the non-African countries.

The OLS estimator suffers from the issue that democracy is not strictly exogenous to income. The GMM estimator developed by Arellano and Bond (1991) tackles the issue by constructing internal instruments. Under the condition of no serial correlation in the error term, higher lags of the endogenous variable can be used as its own instruments. The estimates in columns (3) and (4) are based on this method. To show the validity of the GMM estimation technique, we report p values for the AR(2) test of serial autocorrelation at the bottom of the columns. In both cases, the assumption of serial correlation is clearly rejected. In terms of the effect of democracy on GDP per capita, the GMM coefficients differ to some extent from the OLS estimates. In column (3), for example, the general effect of democracy on income is 0.672. This is about three times as large as the OLS estimate in column (1). The estimate in column (4) deviates from that in column (2) by showing that the marginal effect of democracy is not statistically significant. However, in terms of conclusion, both specifications confirm that democracy does not support income growth in the African countries. This is clarified by estimates for the cumulative effect that are either significantly negative (column (3)) or insignificant (column (4)).

Although powerful, the GMM estimator cannot solve the entire endogeneity problem. A particular issue is time-varying autocorrelated omitted variables that are also correlated with democracy and that have delayed effects on income, e.g., education. In the presence of such variables, GMM estimates mix the effect of democracy with the effects that such variables have on the outcome variable. The IV estimator can overcome this issue when external instruments are applied. These are exogenous sources that affect democracy (relevance criterion) without having a direct effect on the outcome variable (exclusion restriction). The former condition can be tested by the 1st-stage F-statistics where a value \(\ge \) 10 indicates that the IV bias is smaller than that of the corresponding OLS estimator. With respect to the exclusion restriction though, there is no econometric test to confirm the validity of the instrument. It is rather based on economic reasoning. Acemoglu et al. (2019) present numerous specifications showing that their instrument is exogenous to several theories that have the potential to violate the exclusion restriction.

In columns (5) and (6), we present second-stage IV results where democracy is instrument by the first lag of the ANRR-IV. To safe space, the corresponding first-stage results are shown in Table 7 in “Appendix.” Several findings emerge for the second-stage results. First, with the continuous measure for democracy, column (5) shows that the general effect of democracy on income is significant. The estimates in columns (1) and (3) are insignificant (and much smaller in magnitude). An addition, the marginal effect is more than twice as large in magnitude as the previous specifications. This leads to an even larger negative cumulative effect of democracy on income in the SSA countries, 2.688. However, it is imprecisely estimated (error term 1.837) making the coefficient statistically insignificant.

Second, with the dichotomous measure for democracy, column (6) shows that while the estimate for the general effect of democracy is comparable to the estimates in columns (2) and (4), the coefficient for the marginal effect in the SSA countries is more than three times larger in magnitude than the corresponding estimates in columns (2) and (4). This leads to a negative cumulative effect that is also much larger in magnitude than the previous estimates. In any case, the cumulative effect for the SSA countries is still negative and statistically insignificant.

Third, when comparing the different estimators to each other, it turns out that the IV estimates of the cumulative effects are much closer to each other than the estimates of the other methods. This increases confidence in the precision of the IV method. The 1st-stage F-statistics of 24.7 and 9.4 provide further support.Footnote 9 Each of the estimation methods has disadvantages.Footnote 10 Yet, they all suggest that democracy does not support GDP per capita in the African countries.

Panel B repeats the exercises of Panel A using (log) GDP as measure for economic performance. Most of the estimates question a significantly positive effect of democracy on income for both groups of countries. For the non-SSA countries, the estimates in columns (1), (3), and (5) suggest that there is neither significant correlation nor evidence for a positive effect of democracy on income when the spectrum of the political process is measured by a continuous index. This pattern is confirmed by all three estimation techniques. When democracy is measured by a dichotomous index though, the OLS and GMM estimators in columns (2) and (4) show significantly positive coefficients. However, the IV estimate in column (6) turns out insignificant (p value 0.130, not shown in the table).Footnote 11 For the group of the SSA countries, all estimates for the marginal effect of democracy are insignificant. Consequently, most of the estimates for the cumulative effect are also insignificant. The only exception is the coefficient for the correlation between democracy and GDP in column (2).

When comparing the estimates in Panel A to each other, there are many potential explanations for the divergent development path in the African countries. The pool of potential explanations however, can be reduced when comparing the estimates of Panel A to those of Panel B. Despite the correlation analyses in columns (1) and (2), the estimates in Panel A indicate that there is a positive causal effect of democracy on income growth in the non-SSA countries. The estimates in Panel B question this effect even for the non-African countries. The two panels differ only by the fact that one of them takes into account the size of the population while the other does not. This distinction represents a good starting point for the investigation of an explanation.

Table 2 Estimates of the association between democracy (continuous) and growth-related channels

4 Explanations for no economic growth in Africa

In this section, we aim at providing an explanation for the pattern shown in Table 1. There, the evidence suggests that the denominator of GDP per capita plays an important role in explaining why democracy does not lead to more economic growth in the African countries. Thus, we begin by exploring how democracy affects the size of the total population. To this end, we modify Eq. (2) as follows: First, we replace the left-side variable by a demography figure as mentioned below. Second, we augment the right side of the equation by the first four lags of the respective figure (summarized in the table as persistence of the outcome variable). Third, we keep the first four lags of (log) GDP per capita as controls in order to exclude the effect of economic growth. The statistics for the growth process are not reported in the table.

Table 2 displays the estimates. Therein, democracy is measured by the continuous index. The estimates with the dichotomous measure led to the same qualitative results (see Table 8 in “Appendix”). Columns (1) to (3) show the association between democracy and (log) population size. The specifications differ only by the estimation method (as mentioned beneath the outcome variable). All three estimators show that democracy has a decreasing effect on the population size in the non-African countries.Footnote 12 This is clarified by estimates of −0.09, −0.07, and −0.23, all negative in sign. Only with the GMM estimator is the coefficient not significant (p value is 0.151, not shown in the table). For the African countries in turn, the marginal effect of democracy is positive and highly significant in all three cases (see Table 10 in “Appendix” for the first-stage results of the IV specifications). Since the marginal effects are all larger in magnitude than the associated estimates for the general effects, estimates of the cumulative effect of democracy on population size turn out positive suggesting that democracy increases the headcount in the African countries. This result is the opposite of what happens in the non-African countries and it provides an explanation for the heterogeneous effect of democracy on GDP per capita across the African and non-African countries.

Larger populations are not necessarily an obstacle to income growth. The effect depends on how the additional units contribute to economic performance. The estimates in columns (4) to (6) shed some light on this question as they show how democracy affects the age structure of the population. In detail, we estimate the effect of democracy on the \(\%\)-share of population that is 0–14 years. Again, it turns out that there is meaningful heterogeneity across the two groups of countries. In the non-African countries, democracy decreases the share of young population. This is shown by the coefficient of the general effect of democracy that is significantly negative with all three estimators. In the African countries in turn, the effect is reversed due to marginal effects that are all positive and larger in magnitude. This leads to cumulative effects that are significant and positive. In sum, these estimates suggest that democracy decreases the share of young people in the non-African countries but increases that share in the SSA countries. This pattern explains why democracy does not support total GDP growth in Africa.

Finally, the question remains why Africa’s demography figures show this picture. We explore this issue by estimating the effect of democracy on (log) child mortality rate. The estimates in columns (7) to (9) provide an interesting pattern as they show that democracy decreases the child mortality rate only in the non-African countries. In the SSA countries though, the marginal effect is positive suggesting that democracy’s effect on child mortality follows a different path in Africa. Although the magnitudes of the marginal effects are smaller than those of the general effects, they are large enough to offset the general effects such that the cumulative effect of democracy on the child mortality rate turns out insignificant in Africa. This pattern is confirmed by all three estimators.

In sum, the specifications in Table 2 provide an explanation why democracy does not lead to more economic growth in the SSA countries. The evidence draws a line from poor health to changes in demographic figures that affect economic growth. We are not aware of exogenous variation in the demographic figures. This complicates the analysis of the causal effects that these factors have on growth. However, in Table 11 in “Appendix,” we provide some insights. In detail, we employ the OLS and GMM estimator exploring the effects that total population size and the share of young people have on GDP per capita. The estimates too show a heterogeneous pattern across the SSA and non-SSA countries. In detail, they show that an increase in the population size has negative effects on growth only in the SSA countries. Outside of that region, the effect is insignificant. The share of young population in turn has negative effects on GDP per capita only in the non-SSA countries (note that this share decreases due to democracy). In the SSA countries, that share has no significant effect on GDP per capita. This confirms the suggestion that the youth do not contribute to growth in Africa. This pattern has important policy implications.

5 Conclusion

The question whether or not the political orientation of a country matters for economic performance is of central interest to the political economy literature. In this study, we explore this issue for the countries in Sub-Saharan Africa in comparison with other countries. Regression analyses show that unlike in the non-African countries, democracy has no significantly positive effect on GDP per capita in the African countries. We show that heterogeneous effects of democracy on the size and age structure of the African populations explain the different development paths. The demographic changes in turn are related to the finding that democracy fails to reduce the child mortality rate in the African countries. In the non-African countries though, democracy is shown to significantly improve the child mortality rate. Taken together, the analysis shows that democracy does not support income growth in the African countries because it does not support a demographic transition.Footnote 13

Future research may want to explore in more detail why democracy does not have the same pro-health effect on child health in Africa than elsewhere.Footnote 14 A famous branch of the literature argues that this is related to the continent’s geographical position that is more prone to tropical diseases than other regions of the world (see, e.g., Sachs and Warner (1997), Bloom et al. (1998), Gallup et al. (1999), Gallup and Sachs (2001), and Sachs and Malaney (2002)).