Which Dictators Produce Quality of Government?

Although there is a large literature on the two-way causality between institutions and economic development, there is also a solid body of evidence showing causality from institutions to income (e.g., Acemoglu, Johson and Robinson (2001). For a review, see Pellegrini and Gerlagh 2008).

Rothstein and Teorell (2008) offer a review of the main literature on the consequences of quality of government. For a non-academic review, see The Economist 13-03-2008.

By Quality of Government we are referring in this paper to what others in the literature have termed “State Capacity” or “Administrative Capacity” (Bäck and Hadenius 2008)—that is, the capacity a state has to perform its activities in an efficient way and without corruption. Thus, following this literature we are primarily concerned with accounting for variation in public sector bureaucratic performance and corruption. We borrow the term “Quality of Government” (QoG) from Rothstein and Teorell (2008), because the term “Capacity” has been more extensively used to depict the size or the level of resources—or even the capacity to raise taxes—a state has while we are more interested in how the state takes advantage of the resources it manages—that is, in its “quality.” Nevertheless, as already mentioned, QoG could be interchangeable here by the standard definitions in the literature of state capacity or administrative capacity.

According to indicators such as the World Bank’s Government Effectiveness indexes (Kaufmann et al. 2007) and the PRS Group’s International Country Risk Guide index, these four sates all receive scores well over the average in recent years, signifying high QoG. For example, all four of these states outperform India, Peru, Uruguay, and Bulgaria, which all rank as highly competitive democracies in recent indexes of both Polity and Freedom House.

Due to space constraints, the following is admittedly a limited discussion of the pros and cons of the data sources to be employed in the analysis, for a more thorough discussion of debates on such sources of QoG, see Knack (2007) and Pellegrini and Gerlagh (2008).

However, we are doubtful that this is the case, as Kaufmann et al. (2007) show that there are insignificant differences in household perceptions compared with those of expert opinions catering to businesses.

Not all regimes headed by military officers are classified as military regimes by Geddes. What defines a military regime is that it is controlled by a group of senior military officers. Some regimes where the leader may wear a uniform—such as Trujillo in the Dominican Republic, Uganda’s Idi Amin, or Central African Republic’s Bokassa—are considered personalist because they are personal dictatorships of a single officer.

Again, it is difficult to distinguish between “real” single-party regimes, where the organization exercises some power over the leader and the career paths of officials, and “nominal” ones, where the leader himself maintains a near monopoly over policy and personnel (Geddes 1999a: 124). Geddes’ dataset—used also in this paper—includes within the single-party category such regimes as the Partido Revolucionario Institucional (PRI) in Mexico or the Leninist parties in Eastern Europe, but regards as personalist systems regimes such as Juan Peron’s in Argentina, since the single-party was more “nominal” than “real.”

As regards to heteroskedasticity, we used both a Breusch-Pagan (1979) and a Cook-Weisberg (1983) post-estimation test using the baseline model in Table 1. As regards to serial correlation, we employed a Woodridge test, as recommended by Drukker (2003) for panel data. In both cases, we found that the post-estimation tests rejected the null hypotheses that there was no heteroskedasticity or autocorrelation (AR 1).

For a thorough overview of this issue, see the debate between Kurtz and Schrank (2007) and Kaufmann, Kraay and Mastruzzi (2007).

Depending on model specification, roughly 15–20% of the available sample in the regressions includes four classifications of “hybrid regimes” (single party/military, single party/personalist, personalist/military, and single party/personalist/military). In assigning the hybrids to one of the four main regime types, we follow Wright’s grouping method (Wright 2008a, b), and thus, when we discuss military and single-party regimes, we also include the hybrid forms of these regimes, as well as triple-hybrid regimes. For example, single party/military, single party/personalist, and single party/personalist/military are coded as “Single party” while personalist/military is coded as “Military.” We checked the sensitivity of these coding with several alternatives and did not find the results to be altered in any significant way. In addition, we run models that test H1 and H2 with all hybrid state years dropped from the analysis to check for robustness, which we include in Tables 1 and 2.

We elect to use the Geddes/Wright classification due to the essential feature of what differentiates a “personalist” regime from other types of authoritarianism is that “although personalist regimes have parties and militaries, these organizations have not become sufficiently developed or autonomous to prevent the leader from taking personal control of policy decision and selection of regime personnel” (Geddes 2003). Thus, because of weaker institutions in these societies, clientalism tends to dominate more as a ruling method than in other authoritarian regimes (Wright 2008a, b). Quantitatively speaking, the dataset most known for this type of classification is by Geddes (1999a), which has been updated and expanded (also to include “monarchies”) by Wright 2008a, b), whereby four groups—the “personality,” the “single party,” the “military regimes,” and “monarchy” are distinguished. In the single party, “access to political office and control over policy is controlled by a single political party” while in military regimes “a group of officers decides who will rule and exercise some influence on policy” (Geddes 1999a: 121). Power in a monarchy however relies on hereditary passing of rule from one generation to the next. Gandhi and Przeworski (2006) and Cheibub, Gandhi and Vreeland (2010) base their coding on the “way in which governments are removed from power” (Cheibub, Gandhi and Vreeland 2010: 84). Their classification consists of two similar categories as the Wright and Geddes data (military and monarchy) and a “civilian” category that “create(s) a smaller body within a regime party—a political bureau—to co-opt potential rivals.” For practical purposes, the “civilian dictatorships” in the Cheibub, Gandhi and Vreeland are more or less equivalent to Geddes/Wrights’ single-party regimes, and then we apply our hypotheses regarding single-party countries to civilian dictatorships when using the Cheibub, Gandhi and Vreeland (2010) data.

Which Wright defines as the “predicted probability of authoritarian regime failure” (Wright 2008a, b: 972)

For a full description of this measure and logit estimations for all variables in constructing the measure, see Wright (2008a: 47)

For the sake of space in the model, we present only the results of the oil reserves and population consistently. For the results of the other variables, such as ethnic fractionalization, the year count and area dummies, please contact the authors.

In addition, we performed a post-regression t test between the GDP coefficient for single parties and personalist regimes and found the former to be significantly higher (p value = 0.03)

We follow the advice of Gujarati (2003: 181–182) on the interpretation of logged variables in the form of a one percentage increase.

The black dots represent the coefficients for GDP per capita for each of the regime types, while the line connecting the dots should not be interpreted. Monarchies are included in each box because they served as the comparison group in Table 1. The graph was produced with the codes from Brambor, Clark and Golder (2006) and combine with the “graph combine” command in STATA.

A baseline not run model here because the results would be the same as in Table 1

Again, we perform a post-regression t test between GDP in military and civilian regimes, with the two being significantly different (p value = 0.045)

In the time period during which the World Bank data are available, the range of the time horizon variable is even smaller among monarchies, with a max value of 0.061.

To make these effects more meaningful, we show a change of QoG based on min–max values within each regime type, using the “predict” command in STATA. Control variables are held at their mean values.