Electoral Authoritarianism and Human Development

Democracy and Human Development

An enormous literature has developed on the association between democracy and human development, as well as economic growth, inequality, and countless other outcomes. ¹ The central logic is that democracy provides incentives for politicians to improve citizen welfare. Democracies are also more effective at implementing policies, as they have higher quality of government and lower corruption (Adserà, Boix, & Payne, 2003; Humphreys & Bates, 2005).

There is strong evidence that democracy leads to greater public investment in health and education (Brown & Hunter, 2004; Haggard & Kaufman, 2008; Lake & Baum, 2001; Stasavage, 2005; but see Mulligan, Gil, & Sala-i-Martin, 2004). Suffrage extensions, particularly for women and the poor, have also been shown to increase social spending (Husted & Kenny, 1997; Lindert, 2004). In turn, this spending translates into a positive association between democracy and infant mortality (Besley & Kudamatsu, 2006; Gerring et al., 2012; Lake & Baum, 2001; Zweifel & Navia, 2000), other measures of public health (Gerring, 2011; Lake & Baum, 2001; Sloan & Tedin, 1987; Wigley & Akkoyunlu-Wigley, 2011), educational attainment (Brown, 1999; Halperin, Siegle, & Weinstein, 2010; Sloan & Tedin, 1987), and gender equity in human development (M. Baum & Lake, 2003; Beer, 2009; Brown, 2004). However, the link between democracy and outcomes is more often disputed (Carbone, 2009; Nelson, 2007; Ross, 2006). For instance, Ross (2006) argues that democracy increases health spending targeted at the middle class, but is unrelated to infant mortality.

Autocratic Institutions

The spread of autocratic institutions

Despite the extensive literature on democracy, there exists little work distinguishing human development outcomes among autocratic types. This is a major oversight given the global spread of autocratic institutions traditionally associated with democracy, such as legislatures, parties, independent courts, and elections (Gandhi, 2008b; Schedler, 2006). Figure 1 shows the prevalence of closed autocracy, EA, and democracy in each year from 1946 to 2007 (from Boix, Miller, & Rosato, 2013; Cheibub, Gandhi, & Vreeland, 2010). Again, EA regimes are defined as autocracies with legal multiparty competition for the legislature (e.g., Russia, Malaysia, and Singapore). Closed autocracies either lack electoral institutions (Saudi Arabia, UAE) or feature single- or no-party elections (North Korea, Swaziland), which are largely ceremonial in nature. EA regimes distinguish themselves from democracies by manipulating electoral competition to below democratic standards. As seen, EA is not entirely new, but has recently become the large majority among autocracies.

OPEN IN VIEWER

Figure 1.

Distribution of three regime types by year.

This diffusion has stimulated a growing literature on the causes and consequences of autocratic institutions (Geddes, 1999; Levitsky & Way, 2010; Schedler, 2006; Svolik, 2012). A dominant theme in this work is the strategic value of institutions for regime survival, either through co-opting elites (Blaydes, 2011; Geddes, 1999; Magaloni, 2006, 2008; Svolik, 2012) or extending control over citizens (Lust-Okar, 2006; Magaloni, 2006). Autocratic elections, for instance, allow regimes to signal dominance (Magaloni, 2006; Simpser, 2013), disperse patronage (Lust-Okar, 2006; Magaloni, 2006), and gather information on opponents (Gandhi & Lust-Okar, 2009; Magaloni, 2006; Miller, 2015).

If autocratic elections were only adopted based on such calculations, it would be difficult to test their effects as they would be entirely endogenous to regime strategies. However, there are several other motivations for regimes to hold elections. The most widely recognized is international pressure, which rewards countries for adopting electoral institutions through aid, trade, and military benefits (Carothers, 1999; Levitsky & Way, 2010). Although usually packaged as democracy promotion, the end result is often transition to EA, as autocrats determine that they can profitably adopt the formal institutions of democracy without the substance (Carothers, 1999; Goldsmith, 2008). Clear cases include Leopold Senghor’s late-1970s adoption of EA in Senegal “to attract Western aid and investors” (Coulon, 1988, p. 157) and Burma’s recent liberalization. Other factors encouraging EA adoption include domestic protest, the country’s institutional history, post-civil war settlements (often under international mediation), and regional diffusion, which I expand upon below.

Electoral Pressure

By definition, autocratic elections are contested on an uneven playing field (Magaloni, 2006; Schedler, 2006), but this does not eliminate uncertainty or electoral turnover. In fact, autocratic elections can be highly competitive, providing an opening for citizens to sanction regimes for poor policy outputs. As Levitsky and Way (2010) argue, “Government officials fear a possible opposition victory (and must work hard to thwart it) . . . In competitive authoritarian regimes, incumbents are forced to sweat” (p. 12).

Contrary to what many assume, autocratic elections feature dramatic and often unexpected swings in support. In each legislative election under EA, a one-standard-deviation shift in the ruling party’s seat share is about 19%, slightly higher than within democracies. ⁵ Examples of unanticipated electoral shocks under EA include Mexico in 1988, Singapore in 1991, Morocco in 2007, and Russia in 2011. As a result, electoral turnover is a real threat. Using Hyde and Marinov’s (2012) data, I find 51 elections since 1946 in which an EA regime’s incumbent party lost, with another 55 in which the incumbent leader left office due to the election. Together, these constitute nearly 1 in 5 national elections under EA. Either event roughly doubles the likelihood of democratization within 2 years (using Boix et al., 2013), as occurred in Benin in 1991, Guyana in 1992, and Mexico in 2000. Furthermore, a majority of the turnover cases experienced “regime failure” as defined by Geddes, Wright, and Frantz (2014). To avoid this fate, regimes spend copiously on election campaigns, clientelism, bonuses for public workers, and public goods to placate citizens, none of which would occur if elections were truly pre-ordained.

Even when ruling parties do not immediately risk losing power, several features of contested autocratic elections encourage policy responsiveness. First, even when they control elections in the present, autocrats are constantly reminded that complacency risks defeat in the future. It is tempting to make the faulty inference that because many autocratic elections are won by large margins, they cannot motivate regimes to provide public services. However, this is akin to arguing that because a champion boxer consistently wins his fights, he must not feel the need to train very hard. This foresight retains parties’ sensitivities to popular demands, while lengthening their time horizons, leaving them more inclined to invest in education and long-term development (Gehlbach & Keefer, 2011; Olson, 2000; Wright, 2008). Ruling parties can also benefit from successful development even after democratization, as most remain electorally competitive post-transition (Wright & Escribà-Folch, 2012). For instance, the ruling legacies of Mexico’s PRI and Taiwan’s Kuomintang (KMT) were central to their regaining power under democracy.

Second, even when national parties are highly secure, individual politicians may not be. For instance, between 1984 and 2005, reelection rates for Egyptian legislators varied from 19% to 42% (Blaydes, 2011). Local leaders also use strong electoral results to signal their competence and rise up the party hierarchy (Blaydes, 2011; Brandt & Turner, 2007). This electoral pressure motivates politicians to improve citizen welfare, at least in their local constituencies. Third, besides the fear of outright losing, ruling parties seek to maximize their winning margins to project strength and deter elite defections (Magaloni, 2006; Simpser, 2013). Fourth, even semi-competitive elections can function as dangerous focal points for opposition collective action, especially if there is a widespread perception of fraud (Fearon, 2011; Magaloni, 2006; Svolik, 2012). About one in three EA elections feature significant protests, riots, or violence surrounding the election (Hyde & Marinov, 2012). This vulnerability encourages regimes to at least temporarily reduce citizen discontent.

A final question is why many autocrats choose social welfare policies and public goods to defuse this popular pressure, rather than exclusively relying on fraud, repression, or vote buying. Although many EA regimes rely on some mix of these strategies, public goods often prove to be the safest option. Fraud and repression risk triggering protest, a loss of legitimacy, and international punishment, thereby forgoing the central benefits for which elections are adopted. Vote buying and patronage can control elections, but can also be exceedingly costly, especially for large, wealthy populations (Gandhi & Lust-Okar, 2009; Hicken, 2011). ⁶ When clientelism does occur, it often consists of an informal bargain in which local public goods are traded for electoral support. ⁷ As a result, when constructing electoral coalitions, EA regimes are naturally drawn to the poor as they are more easily co-opted by state assistance (Blaydes, 2011; Gandhi & Lust-Okar, 2009; Hicken, 2011). ⁸ In turn, this leads many EA regimes to strategically emphasize social assistance and basic development outcomes.

Numerous cases show that even minimally competitive elections can motivate EA regimes to improve social welfare policies. ⁹ Nelson (2007) points to a positive effect of competitive elections on education reform in autocratic Malawi, Uganda, Tanzania, Kenya, and Mexico. In several detailed case studies, Haggard and Kaufman (2008) identify autocratic elections as key to increased public goods provision and social assistance in Taiwan, Brazil, Mexico, Peru, and Venezuela. We can also find evidence of within-regime policy shifts: In a study of 86 EA regimes, electoral declines for ruling parties predict post-election increases in education and social spending and decreases in military spending (Miller, 2015).

A clear example of autocratic electoral pressure promoting human development is Taiwan’s path to universal health care in 1995. As late as 1980, only 16% of Taiwanese were insured and they spent a miniscule US$78 per capita on health care (Chiang, 1997). After the ruling KMT allowed opposition competition in 1986, the Democratic Progressive Party (DPP) strongly emphasized social welfare reform and garnered unexpected electoral support in 1986 and 1989. The KMT quickly pivoted to an expansion of social programs due to this newfound political pressure (Chiang, 1997; Haggard & Kaufman, 2008; Son, 2001). In particular, the KMT moved to nullify one of the DPP’s chief policy demands by passing health insurance for farmers in 1988 and proposing a National Health Insurance (NHI) system by 2000. ¹⁰ Although the KMT remained electorally dominant, the DPP was steadily gaining support, creeping up to 31% of votes in 1992. Anticipating the rematch in 1995 and facing mounting pressure to deliver on social demands, the KMT fast-tracked the NHI, implementing it 9 months before the election (Son, 2001). By February 1996, 92% of the population was insured, with a big expansion in health care utilization (Chiang, 1997).

Civil Liberties and Political Openness

An alternative perspective is that direct popular engagement with politicians is more effective at engendering responsiveness than electoral pressure (Cleary, 2007; Verba, Schlozman, & Brady, 1995). Protests, strikes, personal appeals, and civil society movements all motivate governments by providing information on policy demands and threatening to develop into electoral or violent challenges if left unaddressed. In turn, the ability of citizens to effectively pressure the government depends on a number of political factors, often termed the “political opportunity structure” (Kitschelt, 1986). I focus on two such factors: individual civil liberties and the openness of the political space to rival groups. Although democracies provide the freest environments by far, EA regimes are at least freer than closed autocracies and thus motivate their leaders to provide more extensive public goods.

First, protected rights to speech and association are critical to popular pressure. ¹¹ Here, the evidence is clear that EA regimes lie about midway between closed autocracies and democracies, using either the Freedom House (2010) rating of civil liberties or the Cingranelli–Richards (CIRI) data set’s rating of freedom of speech (Cingranelli & Richards, 2010; see Online Appendix Figure A1.) The freedom advantage of EA over closed autocracy is confirmed in the empirical section, which further shows that legislatures and parties by themselves are negatively related to civil liberties.

Second, political openness gives room to rival organizations that can pressure the ruling party or help to provide social services themselves (Lake & Baum, 2001; Gerring et al., 2012). By definition, EA regimes allow opposition parties, and they tend to also tolerate outside groups like unions, non-governmental organizations (NGOs), and civil society movements, albeit begrudgingly. Political openness can particularly influence development outcomes, as popular challenges often focus on the inadequate delivery of public services. In 1970s Brazil, for instance, an ideational movement within the state bureaucracy known as the sanitaristas coalesced around the goal of decentralizing and modernizing public health care (Falleti, 2009). In Russia, civil society groups such as Health Care for Children have organized around public health grievances, prompting a US$900 million government initiative to improve health care delivery (Englund, 2011).

Governmental Capacity

It is not enough for leaders to want to improve citizens’ lives—they must have the capacity to do so. A government’s ability to translate policy goals into outcomes encompasses a range of political qualities, including bureaucratic capacity, the rule of law, corruption, and territorial control by the state. Whereas democratic systems necessitate a variety of institutions that can effectively implement policies (Adserà et al., 2003; Humphreys & Bates, 2005), closed autocracies often exercise power through personal ties, military hierarchies, or ruling families. In particular, closed regimes often invest little in professionalized bureaucracies, which are costly and potentially threatening. As a result, their ability to deliver public services may be limited, even when so motivated.

In contrast, EA regimes establish a similar set of formal institutions as democracies. Even when manipulated, they contribute to bureaucratic development, specialization, and information-sharing, promoting governmental effectiveness in the long run (Charron & Lapuente, 2011). Looking at Malaysia, Indonesia, and the Philippines, Slater (2008) concludes that autocratic elections can spur institutional capacity by forcing regimes to develop competent parties and extend state control to marginal populations. EA may also help to select more skilled and technocratically competent leaders, who must rise through a party hierarchy (or found it) and win contested elections (Svolik, 2012). Other autocrats rule by virtue of birth or military leadership, which has little connection to governing ability.

Finally, EA regimes have an advantage in that autocracies are typically less constrained in implementing their policy goals. Democratic leaders face strong incentives for social reform but can be stymied by powerful interest groups (such as unions and bureaucracies) and legislative gridlock (Brown & Hunter, 2004; Nelson, 2007; Olson, 1984). In contrast, ruling parties in EA regimes usually dominate the legislature and face limited opposition from within the public sector (Levitsky & Way, 2010; Magaloni, 2006). Thus, EA regimes advantageously combine electoral incentives with political centralization.

The Endogeneity Problem

The central problem with testing the effect of EA is the endogeneity of regime type. Elections are adopted by autocrats who simultaneously determine governing strategies and policies. A variety of factors could mutually influence these decisions. For instance, strong opposition movements may lead autocrats to accede to elections and policy concessions. Simple correlations cannot therefore determine the causal effects of autocratic institutions. Unfortunately, most work in this area has ignored the endogeneity problem, with Gandhi (2008b) a notable exception. Although the problem also applies to democracy, it is less severe given that democracy is typically not chosen by the same rulers who determine policies. Endogeneity concerns are partly addressed here by testing regime history, which is driven by institutions up to 60 years in the past, lessening the confounding from near-term political factors.

IVs provide a fuller approach to the endogeneity issue. IVs are used to demonstrate that an endogenous variable X causes variation in an outcome Y. To recover a causal effect, we use a set of instruments Z that must satisfy two conditions: (a) They must be sufficiently predictive of X and (b) each instrument must satisfy an exclusion restriction whereby it only influences Y through its effect on X (net of any controls). The second condition, often ignored in empirical work, is critical: Only if we can substantively support the claim that Z does not directly influence Y can we infer the causal effect of X.

In at least three studies, Gandhi (2008a, 2008b; Kim & Gandhi, 2010) uses IVs to account for the endogeneity of autocratic legislatures. However, as noted in Pepinsky (2014), the choice of instruments is problematic, as none clearly satisfy an exclusion restriction. ¹² I now turn to an original IV strategy that more plausibly meets the exclusion restrictions.

The IV Strategy: Diffusion and Inheritance of Autocratic Elections

This study uses two sets of instruments for EA. Both are derived from theoretical reasons that regimes hold contested elections, independent of their general cooptation strategies: international diffusion and institutional inheritance. First, countries are known to be more likely to adopt and sustain democracy if higher fractions of their region and neighborhood are democratic (Gleditsch & Ward, 2006; Starr, 1991). I find the same diffusion effect for EA, implying that the regional and neighborhood prevalence of EA and democracy can serve as exogenous instruments for EA (Miller, 2014). This parallels a similar strategy in Acemoglu, Naidu, Restrepo, and Robinson (2014), which uses regional diffusion to instrument for democracy.

Why are autocracies more likely to adopt elections when nearby countries include them? Most simply, surrounding regime types proxy for region-specific international pressures (Levitsky & Way, 2010). Furthermore, there are at least four reasons that EA and democratic neighbors directly influence EA adoption: learning and emulation by autocratic leaders (Simmons, Dobbin, & Garrett, 2006), an absence of support for similar neighboring regimes from powerful closed autocracies (Ambrosio, 2010; Levitsky & Way, 2010), increased popular pressure for elections when citizens witness them in neighboring countries (Bratton & van de Walle, 1997; Gleditsch & Ward, 2006), and a desire by autocrats to avoid looking like illiberal outliers (Whitehead, 1996). For instance, Julius Nyerere instituted multipartyism in Tanzania based on a belief that the wave of liberalizations in early-90s sub-Saharan Africa “would inevitably catalyze pressures for similar changes in Tanzania” (Hoffman & Robinson, 2009, p. 125). Based on this theoretical setup, I include four instruments: the fraction of democracies and EA regimes in each country’s region (excluding the country itself) and among its neighbors. ¹³

Second, I leverage the inheritance of EA from previous rulers. Multiparty competition is very sticky, even after a change of leadership. When an autocratic executive loses power, the new ruler retains multiparty elections 93% of the time. As Pepinsky (2014) notes, the number of parties inherited from the previous regime is the most convincing instrument used in Gandhi (2008b). Jensen et al. (2014) use an identical instrument. Similarly, I use two dummies for EA under the previous leader and the previous autocratic regime. ¹⁴ When testing regime type history, I instead use weighted histories of the six variables as instruments.

As detailed below, diagnostic tests clearly indicate that these variables satisfy the two requirements for instruments. The first stages of the IV models show that they are highly predictive of EA and its history, well exceeding standard benchmarks for strong instruments. ¹⁵ In addition, I use the limited-information maximum likelihood (LIML) IV estimator, which is more robust to weak instruments than two-stage least squares (2SLS; Stock & Yogo, 2005).

The exclusion restrictions are also likely to be satisfied. First, it is improbable that external regime types directly influence development outcomes. A possible alternative causal channel is through policy diffusion, whereby external countries’ regime types predict their policies and they diffuse to the country’s own policies. To account for this possibility, all models control for the regional average of the dependent variable. This effectively blocks the causal pathway through policy diffusion (see Pearl, 2009 and Online Appendix Figure A2), as well as many other potential confounders. I also control for aid to account for direct intervention. Thus, for the exclusion restrictions to fail, there must be an omitted factor that predicts a country’s development outcomes and its neighbors’ regime types, but not it’s neighbors’ development outcomes, which is unlikely. ¹⁶ The case for EA inheritance is even simpler. As discussed in Jensen et al. (2014), instrumenting with inherited institutions eliminates the confounding due to leaders who establish contested elections and may also pursue distinctive policy strategies. ¹⁷

Second, given the multiple instruments, overidentification tests can further support the exclusion restrictions. These tests calculate whether the residuals from the outcome equation are correlated with the instruments. If so, this is evidence that the instruments violate the exclusion restrictions. For all models, the overidentification tests support the validity of the six instruments. ¹⁸

Although caution is warranted for any IV strategy, this represents the most thoroughly supported IV approach to testing autocratic institutions, both theoretically and diagnostically. The choice of instruments is also guided by the judgments of Pepinsky (2014), Jensen et al. (2014), and Acemoglu et al. (2014). Furthermore, the findings are highly robust to different choices of instruments. The online appendix shows the results for infant mortality and literacy after alternatively dropping the EA inheritance, regional, and neighbor instruments, as well as dropping each of the six instruments individually. For regime history, 18 of 20 tests are significant (at the .10 level). ¹⁹ Finally, the results hold across different samples and IV techniques, and consistently match the results testing regime history without IVs.

Dependent Variables

For outcome variables, I focus on four areas of human development: health, education, gender equality, and basic freedoms. I address infant mortality in the greatest detail, as it is an ideal indicator of inequality, development, and state effectiveness (Gerring et al., 2012; Ross, 2006). I then turn to two closely related measures: the mortality of children under 5 and overall life expectancy. For education outcomes, I look at literacy and school enrollment, as well as the gender balance in both these variables. Unless noted otherwise, the measures are taken from World Bank (2011).

All of these variables are directly affected by public policies and indicate responsive governance and general social welfare. Furthermore, they are well-represented in existing work on democracy, facilitating comparisons with this established literature. I focus on outcomes rather than spending for two reasons. First, public spending measures are often unreliable in autocracies. What goes on the books as social spending may in reality be diverted to patronage. Second, regimes vary greatly in their capacity to translate spending into outcomes, which are of course what we ultimately care about.

I also test four measures of political rights, two concerning civil liberties, one women’s political rights, and the last an indicator of physical repression by the state. Besides their normative importance, political freedoms are critical to human development and economic productivity. Furthermore, the results on civil liberties can be compared against Gandhi (2008b) and Conrad (2011). Summary statistics are shown in Online Appendix Table A1.

Definitions of Regime Types

Democracy is measured using Boix et al.’s (2013) dichotomous coding, which requires competitive elections and a minimal level of suffrage. Although highly correlated with other measures of democracy, this coding is advantageous in that it explicitly differentiates democracies from EA regimes based on the freedom and fairness of elections.

EA regimes are defined as autocracies in which multiple political parties exist and legally compete in legislative elections (from Cheibub et al., 2010). Thus, they tolerate organized opposition and provide citizens a regular, legal route to pressure the government. About 90% of these regimes experienced a full legislative election within the previous 5 years (using Hyde & Marinov, 2012). ²⁰ This definition has the advantage of concreteness, as it is based on a formal legal requirement. In contrast, concepts like competitiveness are worth testing, but are more subjective and will suffer from reverse causation if development outcomes influence electoral behavior.

Empirical Setup

I test four distinct panel models for each outcome variable Y. As a baseline, I first run ordinary least squares (OLS) with dummy variables for EA and democracy (lagged by 1 year):

Y i t + 1 = α 0 + α 1 E A i t + α 2 D i t + α 3 R ( Y i t + 1 ) + α 4 X i t + γ t + ε i t + 1 ,

where EA_it and D_it are indicators of EA and democracy for country i in year t. Closed autocracy is the omitted category. R(·) stands for the regional average (excluding the country i), X_it is a set of control variables, the γ _t are year fixed effects, and ε is an error term.

Although useful, this approach cannot demonstrate the causal effect of regime type. The second model thus instruments for EA using four measures of external regime types and two of institutional inheritance. Democracy is assumed to be exogenous, but violations of this should not affect inference on EA. The first-stage equation is thus the following:

E A i t = β 0 + β 1 R ( E A i t ) + β 2 R ( D i t ) + β 3 N ( E A i t ) + β 4 N ( D i t ) + β 5 R ( Y i t + 1 ) + β 6 E A ( P r e v i o u s L e a d e r ) + β 7 E A ( P r e v i o u s R e g i m e ) + β 8 X i t + γ t + μ i t ,

where N(·) stands for the neighbor average and µ is an error term. The IV models are calculated using the ivreg2 command in Stata (C. F. Baum, Schaffer, & Stillman, 2010). I use the LIML form of the IV, which has attractive robustness properties (Stock & Yogo, 2005). I also check the results using 2SLS and the generalized method of moments (GMM)IV estimator.

The third and fourth models test the cumulative, historical effects of regime types, mirroring the approach in Gerring et al. (2012). For each country, I calculate the weighted average of each regime type back to 1946. As in Gerring et al., I apply a 1% annual depreciation to the weight. In contrast, I normalize the weighted history to the same scale as an average, so that the three measures sum to 1. For instance, EA History is measured as,

E A H i s t o r y i t = ∑ τ ≥ 0 0 . 99 τ E A i , t − τ ∑ τ ≥ 0 0 . 99 τ ,

and similarly for Democracy History. This approach captures long-term experience, but unlike a stock variable, avoids conflating it with the time period and years of independence. The third model type replaces the regime dummies in equation (1) with the history measures. The fourth, which I consider the benchmark model, instruments for EA History using weighted histories of the instruments.

In the main results, the sample is limited to 158 countries from 1960 to 2007, although this varies by dependent variable. The main models use robust standard errors clustered by country. As a robustness check, I use Driscoll-Kraay standard errors, which account for multiple lags of serial correlation and contemporaneous correlation across units.

Control Variables

For each outcome Y, the models control for the regional average of Y (excluding the country itself) in the same year. This accounts for distinct regional characteristics, as well as shocks specific to time and region, such as the diffusion of medical innovations.

The models include a common set of further control variables, which are lagged by 1 year. First, I account for Foreign Aid (official development assistance as a percentage of gross national income, from World Bank, 2011). While obviously influencing outcomes like infant mortality, aid is often tied to political reform and so represents a potential confounder.

Second, I control for five economic variables: logged Gross Domestic Product (GDP) per capita (in real 2000 dollars, from Haber & Menaldo, 2011; World Bank, 2011), Economic Growth (the percentage change in GDP/capita), Resource Dependence (fuel and metal revenues as a percentage of GDP, from Haber & Menaldo, 2011), Economic Inequality (Gini, from Galbraith & Kum, 2003; United Nations University–World Institute for Development Economic Research, 2005; World Bank, 2011), and a Communist dummy. Government resources and the level and distribution of income strongly condition human development. Haggard and Kaufman (2008) show that Communist regimes have highly distinct policy platforms. ²¹

Third, I control for several socioeconomic characteristics that affect the ease of delivering public services: Population (logged, from Heston, Summers, & Aten, 2011), Urbanization (the percentage living in cities of 100,000+, from Correlates of War Project, 2010), and Ethnolinguistic Fractionalization (ELF, from Roeder, 2001). Fourth, because domestic conflict can disrupt government services and impede human development, I control for Political Violence (a 0-10 rating of domestic civil and ethnic violence, from Marshall, 2010).

Finally, to account for variation across time, I include year fixed effects. I do not include country fixed effects, which problematically negate the influence of the 57 countries that do not vary by regime type. Whereas the purpose of country fixed effects is to account for country-level omitted variables, I rely here on instruments and regime history to obtain causal estimates. However, the online appendix shows that the infant mortality results are robust to including fixed or random effects (Online Appendix Table A5). In other checks, I control for region fixed effects and the Geddes autocracy categories (military, party-based, and personalist).

Main Results

Infant mortality

Table 1 displays results predicting Infant Mortality (per 1,000 live births). Models 1 and 2 test EA and Democracy, while Models 3 and 4 test their historical averages. Models 2 and 4 instrument for EA and EA History, respectively. ²² Again, Model 4 is the benchmark model. All four models show superior outcomes under both EA and democracy, with all eight coefficients significantly negative for infant mortality. The results are particularly strong in the IV models, implying a large causal effect of EA on health. According to Model 4, a long-term EA regime has about 27 fewer infant deaths per 1000 live births compared with closed autocracy. This is equivalent to the estimated effect of shifting from Vietnam to Sweden on average income. For democracy, the effect is nearly identical.

OPEN IN VIEWER

Table 1.

IV and OLS Regressions Predicting Infant Mortality.

	OLS	IV	OLS	IV
DV = Infant mortality	(1)	(2)	(3)	(4)
EA	−7.319*	−18.253*
	(−2.15)	(−2.07)
Democracy	−13.089***	−18.561**
	(−3.62)	(−3.19)
EA history			−13.982*	−26.612**
			(−2.57)	(−2.77)
Democracy history			−20.902**	−26.835***
			(−3.23)	(−3.55)
Infant mortality	0.418***	0.425***	0.412***	0.425***
(Regional average)	(4.93)	(4.81)	(4.91)	(4.82)
Foreign aid	0.387**	0.371*	0.290*	0.233 †
	(2.76)	(2.48)	(2.38)	(1.86)
GDP/capita (ln)	−15.132***	−15.175***	−13.776***	−13.346***
	(−7.29)	(−7.08)	(−6.68)	(−5.87)
Economic growth	−0.141*	−0.114 †	−0.125 †	−0.099
	(−2.00)	(−1.67)	(−1.77)	(−1.50)
Economic inequality	0.125	0.087	0.117	0.112
	(0.93)	(0.63)	(0.86)	(0.77)
Resource dependence	0.161	0.133	0.116	0.081
	(1.30)	(1.03)	(0.88)	(0.77)
Communist	−14.991***	−12.805***	−17.037***	−15.079**
	(−4.14)	(−3.62)	(−3.96)	(−3.16)
Urbanization	−0.181*	−0.154 †	−0.200*	−0.177*
	(−2.28)	(−1.92)	(−2.58)	(−2.25)
ELF	3.107	3.871	4.857	5.024
	(0.48)	(0.58)	(0.74)	(0.75)
Population (ln)	2.643**	2.480*	2.459**	2.260*
	(2.95)	(2.54)	(2.75)	(2.34)
Political violence	−0.140	−0.120	0.113	0.125
	(−0.19)	(−0.16)	(0.18)	(0.19)
Year dummies?	Y	Y	Y	Y
n	4,435	4,237	4,455	4,257
Countries	151	144	152	145
Adjusted R2	.798	.790	.805	.798
Weak identification (F)		30.99***		12.95***
Overidentification (χ2)		10.28		7.50

The table displays models predicting infant mortality from regime type. Model 1 uses OLS to compare EA regimes and democracies with closed autocracies. Model 2 instruments for EA. Model 3 uses OLS to test histories of EA and democracy. Model 4 instruments for EA History. In each model, both EA and democracy are significantly negative for infant mortality. Years are 1960 to 2007. t statistics (based on robust standard errors clustered by country) are shown in parentheses. IV = instrumental variables; OLS = ordinary least squares; DV = dependent variable; EA = electoral authoritarianism; ELF = ethnolinguistic fractionalization.

†

p < .10. *p < .05. **p < .01. ***p < .001.

The table also displays the two primary checks for the validity of the instruments. The weak identification test addresses whether the instruments explain a sufficient amount of variance in the endogenous variable (EA or EA History). The specific test shown is the Kleibergen–Paap F statistic (measured against the Stock–Yogo critical values), where the common rule of thumb is a threshold of 10 to be considered valid. In both models, the instruments exceed this threshold. The overidentification test (the Hansen J statistic) calculates whether the instruments are correlated with the error term in the outcome equation, violating the exclusion restrictions. A rejection of the null questions the validity of the instruments, but the test is insignificant in both models. The two checks are virtually identical for the other dependent variables because the only variation in the first stage is the regional DV average.

Results for the control variables are largely expected. Higher income is strongly negative for infant mortality, whereas a higher regional average is positive. Consistent with Haggard and Kaufman (2008), Communist regimes feature fewer infant deaths. More populated and less urbanized countries have higher rates, reflecting a difficulty in delivering public health services. Surprisingly, no relationship with inequality or resource dependence is evident.

Other health outcomes

Table 2 summarizes results for two other health outcomes and four education outcomes. The four model types are identical to those in Table 1, but results for the controls are omitted. Each pair of coefficients (for EA and Democracy) represents a separate regression. A total of 24 are shown in Table 2. Again, the third and fourth models test EA History and Democracy History.

OPEN IN VIEWER

Table 2.

IV and OLS Regressions for Health and Education Outcomes.

	Regime Level		Regime History
	OLS	IV	OLS	IV
	(1)	(2)	(3)	(4)
Under-5 Mortality
EA	−14.03*	−35.82*	−26.52**	−53.51***
	(−2.43)	(−2.51)	(−2.78)	(−3.49)
Democracy	−20.56**	−31.25**	−34.36**	−47.10***
	(−3.34)	(−3.18)	(−3.06)	(−3.50)
Life Expectancy
EA	1.30 †	2.32	2.31*	6.12**
	(1.90)	(1.17)	(2.07)	(3.00)
Democracy	2.80***	3.06*	4.41**	6.18***
	(3.61)	(2.57)	(3.07)	(3.66)
Literacy
EA	2.51	12.30*	10.12*	24.78**
	(0.99)	(2.09)	(2.47)	(2.62)
Democracy	6.53*	12.02**	15.03**	23.64**
	(2.32)	(2.66)	(3.06)	(2.95)
Schooling
EA	2.06	9.52	7.76*	20.77**
	(0.97)	(1.60)	(2.10)	(2.76)
Democracy	3.95 †	7.94 †	8.60*	14.91**
	(1.68)	(1.91)	(2.20)	(2.78)
Literacy Equality
EA	−0.01	0.22*	0.09*	0.23*
	(−0.23)	(2.37)	(2.28)	(2.29)
Democracy	0.01	0.17*	0.11*	0.20*
	(0.51)	(2.33)	(2.51)	(2.36)
Schooling Equality
EA	0.04*	0.16**	0.10**	0.18*
	(2.03)	(3.27)	(2.80)	(2.38)
Democracy	0.05*	0.12**	0.10*	0.14*
	(2.34)	(3.09)	(2.57)	(2.34)

The table summarizes models predicting health and education outcomes from regime type. The six dependent variables are listed in the first column. Four models are then shown for each dependent variable. Model 1 uses OLS to compare EA regimes and democracies to closed autocracies. Model 2 instruments for EA. Model 3 uses OLS to test histories of EA and democracy. Model 4 instruments for EA History. The standard controls are included, but not shown. t statistics (based on robust standard errors clustered by country) are shown in parentheses. IV = Instrumental variables; OLS = ordinary least squares; EA = electoral authoritarianism.

†

p < .10. *p < .05. **p < .01. ***p < .001.

Under-5 Mortality (per 1,000 live births) is a slightly broader measure of basic health compared with infant mortality. Life Expectancy (in years) is broader still, but heavily influenced by child mortality. For both variables, EA and democracy are strongly associated with better outcomes (although EA misses significance at the .05 level for Life Expectancy in Models 1-2). According to Model 4, a long-term EA regime produces 53.5 fewer child deaths per 1,000 births and 6.1 additional years of life compared with closed autocracy. For democracy, the effect is 47 fewer deaths and 6.2 additional years of life.

Political rights

The four dependent variables listed in Table 3 measure basic freedoms and political rights. All are scaled to run from 0 to 1, with 1 highest in freedom. Civil Liberties is from Freedom House (2010) and Free Speech from Cingranelli and Richards (2010). For both, democracies are much freer than closed autocracies, whereas EA regimes lie in between. However, EA History is only weakly related in the IV tests. This may be because long-lived democracies experience a consolidation process of democratic deepening, whereas EA regimes do not—liberties always remain at the mercy of the ruling party’s interests.

OPEN IN VIEWER

Table 3.

IV and OLS Regressions Predicting Political Rights.

	Regime Level		Regime History
	OLS	IV	OLS	IV
	(1)	(2)	(3)	(4)
Civil Liberties
EA	0.11***	0.20***	0.13***	0.15*
	(5.33)	(4.00)	(3.64)	(2.33)
Democracy	0.38***	0.43***	0.38***	0.39***
	(15.57)	(12.24)	(10.41)	(8.19)
Free Speech
EA	0.12***	0.18*	0.10 †	0.11
	(3.41)	(2.29)	(1.92)	(1.32)
Democracy	0.36***	0.41***	0.36***	0.35***
	(10.13)	(7.13)	(6.58)	(5.31)
Women’s Rights
EA	0.06**	0.06	0.05	0.14*
	(2.66)	(0.99)	(1.65)	(2.13)
Democracy	0.07**	0.08*	0.11**	0.18***
	(3.10)	(1.98)	(3.06)	(3.55)
Physical Integrity
EA	0.00	0.03	−0.01	0.03
	(0.07)	(0.48)	(−0.35)	(0.45)
Democracy	0.11***	0.13**	0.11**	0.14**
	(4.54)	(3.01)	(2.89)	(3.15)

The table summarizes models predicting basic freedoms and women’s rights from regime type. The four dependent variables are listed in the first column. Four models are then shown for each dependent variable. Model 1 uses OLS to compare EA regimes and democracies to closed autocracies. Model 2 instruments for EA. Model 3 uses OLS to test histories of EA and democracy. Model 4 instruments for EA History. The standard controls are included, but not shown. t statistics (based on robust standard errors clustered by country) are shown in parentheses. IV = Instrumental variables; OLS = ordinary least squares; EA = electoral authoritarianism.

†

p < .10. *p < .05. **p < .01. ***p < .001.

The Women’s Rights variable (Cingranelli & Richards, 2010) relates positively to EA and democracy and, unlike for civil liberties, results are similar in magnitude. Finally, EA is unrelated to Physical Integrity, an index of freedom from torture, killings, and physical repression by the state (Cingranelli & Richards, 2010), whereas democracy is strongly positive in all four tests. This presents an interesting contrast with EA’s positive relationship with civil liberties. A possible reason is that EA regimes feature a high degree of contention relative to closed autocracies. Thus, their legal recognition of civil liberties may be balanced by their needs to suppress opposition activity (Vreeland, 2008). In comparison, closed regimes facing quiescent populations do not need to resort to widespread physical repression.

Robustness Checks

Table 4 summarizes eight sets of robustness checks for the IV results on infant mortality and literacy, with the baseline results included for comparison. The table shows the IV estimates for EA (Models 1 and 3) and EA History (Models 2 and 4). Each coefficient represents a separate regression. As clearly seen, the regime history results are highly robust, remaining significant at the .05 level in all 16 models. For regime level, results remain significant in 11 of 16 models, narrowly missing in another 3.

OPEN IN VIEWER

Table 4.

IV Robustness Checks for Infant Mortality and Literacy.

	Infant Mortality		Literacy
	Level	History	Level	History
	(1)	(2)	(3)	(4)
Baseline	−18.25*	−26.61**	12.30*	24.78**
	(−2.07)	(−2.77)	(2.09)	(2.62)
Spatial dynamics	−11.89***	−20.96***	5.17**	13.58***
	(−4.50)	(−11.12)	(2.78)	(4.95)
Geddes types added	−18.20 †	−27.76**	13.83*	29.12*
	(−1.83)	(−2.60)	(2.08)	(2.47)
2SLS	−15.83*	−24.53**	9.02*	18.22**
	(−2.18)	(−2.86)	(2.15)	(2.96)
GMM-IV	−16.12*	−22.32**	5.00	11.52*
	(−2.22)	(−2.71)	(1.35)	(1.97)
Region FE	−16.79*	−21.14*	7.23	18.13*
	(−2.00)	(−2.03)	(1.00)	(1.98)
Autocracies only	−17.35 †	−33.36**	13.37 †	42.59**
	(−1.91)	(−3.07)	(1.77)	(3.04)
No state failure	−18.44*	−26.02**	11.82*	24.21**
	(−2.06)	(−2.72)	(1.99)	(2.58)
Inequality/aid omitted	−22.64**	−33.66***	17.86**	29.73***
	(−2.93)	(−4.05)	(3.16)	(4.55)

The table summarizes several robustness checks for the IV models predicting infant mortality and literacy. Only the results for EA and EA History are shown. The first column lists the model checks, which are described in the text. Models 1 and 3 instrument for EA and test the effect of regime level. Models 2 and 4 instrument for EA History and test the effect of regime experience. t statistics (based on robust standard errors clustered by country) are shown in parentheses. IV = instrumental variables; EA = electoral authoritarianism; 2SLS = two-stage least squares; GMM = generalized method of moments; FE = fixed effects.

†

p < .10. *p < .05. **p < .01. ***p < .001.

The first check includes two alterations taking into account possible spatial dynamics. First, I control for the neighbor average of the outcome in addition to the regional average. Second, I use Driscoll-Kraay standard errors, which are robust to three lags of serial correlation and arbitrary correlations across countries. EA’s significance increases in all four tests.

The second check controls for the autocracy categories coded in Geddes et al. (2014). For Models 1 and 3, I add dummies for military and party-based regimes, with personalist dictatorships and monarchies as the base category. For Models 2 and 4, I instead test their historical averages. The Geddes categories are unrelated to both outcomes (see Online Appendix Table A7).

In the next two checks, I replace the LIML IV estimator with 2SLS and a GMM estimator, with little effect on the findings. The fifth check adds region dummies. In the sixth check, I limit the sample to autocracies. The level models necessarily omit Democracy, but the history models still include Democracy History. In the seventh check, I remove cases of state failure (from Marshall & Jaggers, 2010) because these are over-represented among closed autocracies. In the final check, I remove Economic Inequality and Foreign Aid as controls, as they have the most missing data. To compare with Table 1, Model 2’s sample size increases to 5,645 (with 150 countries). For the literacy models, the sample is expanded to 1946-2007. The IV estimates for EA become slightly stronger.

Other Autocratic Institutions

How does EA compare with other institutions that have become widespread in autocracies? Studies of autocracy have focused the greatest attention on legislatures and parties, but there is a shortfall of work explicitly comparing these different institutions. ²⁴ I therefore compare EA with legislatures and parties (measured from Cheibub et al., 2010) in the absence of multiparty competition. Because EA incorporates all three elements, it is not yet clear which is really doing the work. As a robustness check, I also test the effect of a Multiparty Legislature, which differs from EA in additionally requiring that multiple parties are seated in the legislature. About 80% of EA regimes satisfy this requirement. ²⁵ I expect the two measures of multiparty contestation to have similar effects.

To make the comparisons, I adapt the third model type, namely, a non-IV test of regime type histories. Ideally, an IV setup would be used, but I lack a reliable set of instruments for legislatures and parties. ²⁶ For each outcome variable, I run four separate regressions, testing historical averages of the four alternative institutions. Each regression includes the standard controls and Democracy History.

Figure 2 graphically displays the estimated coefficients (with 95% confidence intervals) predicting Infant Mortality, Literacy, Literacy Equality, and Civil Liberties. As discussed, EA History produces superior outcomes for all four measures. Results are virtually identical for Multiparty Legislature. However, the effects of legislatures and parties by themselves are opposite in direction for all four outcomes, albeit inconsistently significant. Thus, legislatures and parties without multiparty competition are associated with higher infant mortality and lower civil liberties.

OPEN IN VIEWER

Figure 2.

Effects of autocratic institutions on human development.

This result conflicts with theories, such as Gandhi’s (2008b), that legislatures in isolation can induce political concessions. Rather, legislatures and parties by themselves may only serve to increase regime power by co-opting elites who would otherwise pressure the government. Multiparty elections, in contrast, motivate regimes to promote citizen welfare. This calls into question many of the empirical findings on legislatures and parties, which may be driven by the large share of regimes with these institutions that additionally feature multiparty competition.