Why do military dictatorships become presidential democracies? Mapping the democratic interests of autocratic regimes

Abstract

Recent data show that virtually all military dictatorships that democratize become presidential democracies. I hypothesize that the reason is that military interests are able to coordinate on status-preserving institutional change prior to democratization and prefer political institutions with strong veto players. Civilian interests are more likely to suffer from coordination failure by being more diverse and less cohesive, implying that most military democratizations are planned partially while most democratization events from civilian autocracy are unforeseen or poorly planned. Exploring the characteristics of 111 democratization episodes between 1950 and 2017 illustrates features broadly consistent with further theoretical predictions.

Appendix: Theory

As a first, normalizing the total supply of inelastic labor at 1, and setting total demand for labor at 1 as in (24), the equilibrium wage rate can be written as (25a), expressed as a function of labor demand in industry n in (23), or the less wieldy expression in (25b). It is easy to see that increasing the subsidy s will reduce the wage rate, w. As it also increases price levels, the overall welfare effect on ordinary citizens is unambiguously negative. At the national level, the effect also is negative because the positive effects on the profits of firms within the K-set cannot outweigh the negative effects for most citizens owing to the adverse effect on overall resource allocation when those firms come to employ too much labor reallocated from more efficient uses.

$$x_{n} = a_{n}^{{\frac{1}{1 - \alpha }}} \left[ {\frac{\alpha }{w}} \right]^{{\frac{\alpha }{1 - \alpha }}}$$

(23)

$$\mathop \sum \limits_{i = 1}^{n} \left[ {\frac{{sa_{i} \alpha \beta }}{{x_{n} w}}} \right]^{{\frac{1}{\alpha \beta }}} = 1$$

(24)

$$w = \frac{{x_{n} }}{\beta }\left[ {\mathop \sum \limits_{i \in K} \left( {sa_{i} } \right)^{{\frac{1}{\alpha \beta }}} + \mathop \sum \limits_{i \notin K} \left( {a_{i} } \right)^{{\frac{1}{\alpha \beta }}} } \right]^{ - \alpha \beta }$$

(25a)

$$w = a_{n} \frac{{\alpha^{\alpha } }}{{\beta^{1 - \alpha } }}\left[ {\mathop \sum \limits_{i \in K} \left( {sa_{i} } \right)^{{\frac{1}{\alpha \beta }}} + \mathop \sum \limits_{i \notin K} \left( {a_{i} } \right)^{{\frac{1}{\alpha \beta }}} } \right]^{{\alpha \beta \left( {\alpha - 1} \right)}}$$

(25b)

Rearranging (5) and (6) and differentiating, it is straightforward to show that the minimum size of the price subsidy, s, that will yield a positive effect on profits in the supported firms, is given by (26). Differentiating, the total effects of the price subsidy s on wages are given by (27). The marginal effect of raising s thus depends both on the relative size of the initial wage and subsidy, but also intuitively on the relative share of total profits—the second fraction—accruing to the supported sectors (the K-set).

$$s > \frac{{\left( {\frac{w}{\alpha \beta }} \right)^{{1 - \alpha \left( {1 - \beta } \right)}} a_{i}^{{\alpha \left( {1 - \beta } \right) - \beta }} }}{{\left( {1 + \alpha \left( {\beta - 1} \right) + {\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 {\alpha \beta }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\alpha \beta }$}}} \right)x_{n}^{{\left( {1 - \alpha } \right)\left( {1 - \beta } \right)}} }}$$

(26)

$$\frac{dw}{ds} = - \frac{w}{s}\frac{{\mathop \sum \nolimits_{k \in K} \left( {sa_{i} } \right)^{{\frac{1}{\alpha \beta }}} }}{{\mathop \sum \nolimits_{k \in K} \left( {sa_{i} } \right)^{{\frac{1}{\alpha \beta }}} + \mathop \sum \nolimits_{k \notin K} \left( {a_{i} } \right)^{{\frac{1}{\alpha \beta }}} }}$$

(27)

As such, any development that either affects productivity within a ‘favored’ industry, i.e., one in the subset K, or a development that affects the productivity of all sectors through the wage rate w, will increase the price of buying support from an industrial selectorate. In other words, for any given tax rate, economic development—whether balanced or not—can undermine the rationale for maintaining a civilian autocracy.

For the firm choices, the first-order conditions of Eq. (7) are given by (27)–(30).

$$\mu_{m = } \left( {1 - \mu } \right)\left( {\frac{\chi }{m} - \frac{\beta }{a}} \right)$$

(28)

$$\mu_{s} = \left( {1 - \mu } \right)\left( {\frac{{\left( {1 - \chi } \right)\mathop \sum \nolimits_{k \in K} {\raise0.7ex\hbox{${d\Pi _{k} }$} \!\mathord{\left/ {\vphantom {{d\Pi _{k} } {ds}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${ds}$}}}}{{\mathop \sum \nolimits_{k \in K}\Pi _{k} }} - \frac{\beta }{a}} \right)$$

(29)

$$\mu_{r} = \left( {1 - \mu } \right)\frac{\beta }{a}$$

(30)

For the voting outcomes in (13) and (14), total income is irrelevant. Because of the assumption of median voter politics, the income of the median voter is the relevant measure. Expression (31) provides the income of the median voter, given that she receives profits from a firm outside of the K-set. Equation (32) provides the median income for a voter receiving dividends from a firm within of the K-set, i.e., a firm that provides support for a civilian autocracy/party A and thus also receives the price subsidy.

$$y_{m, k \notin K} = w + \frac{{a^{1 + \alpha \beta } \alpha \beta }}{{x_{n}^{\beta } }}w^{ - 1} - \left( {\frac{{a_{i} \alpha \beta }}{{x_{n} }}} \right)^{1 - \alpha \beta } w^{1 - \alpha \beta }$$

(31)

$$y_{m, k \in K} = w + s^{2} \frac{{a^{1 + \alpha \beta } \alpha \beta }}{{x_{n}^{\beta } }}w^{ - 1} - \left( {\frac{{sa_{i} \alpha \beta }}{{x_{n} }}} \right)^{1 - \alpha \beta } w^{1 - \alpha \beta } - c_{i}$$

(32)

The sensitivity of median incomes to raising the subsidy, i.e., of buying direct support from industrial special interests, is given by (33) and (34), where R is defined as the relative share of total profits accruing to the firm associated with the median voter. Equations (33) and (34) represent the relevant sensitivities in the policy decisions in (16) and (17). While the sensitivity of y_m is unambiguously negative when the median voter is not associated with a firm in the K-set, the sensitivity when the median voter is so may under specific circumstances become positive. However, as can be seen in (34), that requires a sufficiently large subsidy, s, as well as a profit share R larger than αβ/(1 − αβ).

$$\frac{{dy_{m, k \notin K} }}{ds} = - \frac{{R_{k \notin K} }}{s}\left[ {w + \frac{{a^{1 + \alpha \beta } \alpha \beta }}{{x_{n}^{\beta } }} + \left( {1 - \alpha \beta } \right)\left( {\frac{{a_{i} \alpha \beta }}{{x_{n} }}} \right)^{1 - \alpha \beta } w^{1 - \alpha \beta } } \right]$$

(33)

$$\frac{{dy_{m, k \in K} }}{ds} = \frac{{R_{k \in K} }}{{R_{k \notin K} }}\frac{{dy_{m, k \notin K} }}{ds} + 2\frac{s}{w}\frac{{a^{1 + \alpha \beta } \alpha \beta }}{{x_{n}^{\beta } }} - \left( {1 - \alpha \beta } \right)s^{ - \alpha \beta } \left( {\frac{{a_{i} \alpha \beta }}{{x_{n} }}} \right)^{1 - \alpha \beta } w^{1 - \alpha \beta }$$

(34)