Context Matters

While a number of studies focus on the specific role of legislatures, or their effects that are independent of political parties, few examine how legislative assemblies operate in different autocratic contexts.Footnote ²³ Recent research, however, demonstrates that differentiating among autocratic contexts can help explain important outcomes such as authoritarian longevity and conflict behavior in non-democracies.Footnote ²⁴

Early studies of authoritarianism argued that personalist regimes differ from other autocracies, such as military and party-based dictatorships, in the methods by which they rule.Footnote ²⁵ Some posited that personalism is best conceptualized as a continuous trait that is a secondary feature of institutional settings, but others observe that dictators create and maintain formal political institutions with little real power – often in lieu of empowering military or delegative organizations.Footnote ²⁶ The domination of political channels by an individual leader characterizes what others have termed ‘neo-patrimonial rule’, a type of authority considered the most salient for many non-OECD countries and an attribute of politics in many African countries.Footnote ²⁷

In personalist regimes the autocratic leader has consolidated power over organizations, such as the military or the support party, through which members of the support coalition could ‘prevent the leader from taking personal control of policy decisions and the selection of regime personnel’.Footnote ²⁸ Autocratic regimes lacking powerful formal institutions are not necessarily unstable, but they are often long lasting.Footnote ²⁹ Both monarchies and personalist regimes share longer tenures and may together constitute a common ‘hybrid’ form of government.Footnote ³⁰ In such regimes, formal institutions may be unlikely to constrain a dictator’s opportunistic behavior. Consistent with the expectation that the influence of formal political institutions differs across autocracies, Wright finds that the cross-national correlation between legislatures and economic outcomes, such as growth and domestic investment, varies by regime type.Footnote ³¹ This finding is broadly consistent with Gandhi’s study of autocratic institutions, but suggests that in some regimes – in which the leader has consolidated personal power over the support party and the military – formal institutions do not positively influence growth and investment.Footnote ³²

Svolik articulates a logic for understanding why political institutions are unlikely to enhance power sharing – and potentially constrain elite expropriation – when the dictator has successfully consolidated personal power over other elites.Footnote ³³ He argues that formal political institutions, such as legislative bodies, aid power sharing by lowering the costs of monitoring the power-sharing arrangement for regime elite.

Importantly, his model shows that institutions are only likely to facilitate power sharing when the dictator’s power vis-à-vis the ruling coalition is not excessively high. Repeated, successful power grabs by the leader can give rise to an ‘established dictatorship’ in which the leader has acquired sufficient control to restrict institutionalized avenues for deposing him.Footnote ³⁴ When the dictator has enough power, institutions cease to work as monitoring devices because his commitment to refrain from further opportunistic power grabs is no longer credible. In short, when power sharing is infeasible because one bargaining partner dominates the interaction, institutions cannot augment power sharing. Institutions in dictatorships where the leader has consolidated power over the military and political parties – namely personalist regimes – are therefore unlikely to ‘bind the grabbing hand’.

Although formal political institutions may lack the ability to constrain the leader in personalist contexts, they can still have instrumental value for the leader by, for example, serving as venues in which the dictator distributes patronage and identifies potential opponents. Indeed, Migdal notes that an important tactic for political survival in weak states is to ‘shuffle’ strong – and thus potentially threatening – local notables through the state bureaucracy to prevent them from organizing and independent power base.Footnote ³⁵ For example, while Rafael Trujillo developed a stranglehold on domestic institutions in the Dominican Republic, he used the legislature to manipulate and subdue elites.Footnote ³⁶ Legislative assemblies used principally for this purpose may not necessarily serve as conduits of information for anyone other than the leader, and thus do not facilitate the monitoring that enables credible power sharing, as proposed by Svolik.Footnote ³⁷

Gehlbach and Keefer posit an alternative logic that nonetheless points to personalist rule as a context in which we should not expect political institutions to constrain autocratic leaders.Footnote ³⁸ Party institutionalization, they argue, facilitates collective action among regime elites so they can credibly threaten the leader with removal and thus deter opportunistic behavior – particularly asset expropriation.

In their empirical analysis, the authors employ measures of party institutionalization that closely match concepts captured in the Geddes categorization of personalist regimes.Footnote ³⁹ Questions used to code the latter include: (1) whether the dictator lacks a support party (more personalist), (2) whether the party was created after the dictator gained power (more personalist), (3) whether the dictator chooses the party leaders (more personalist), (4) whether there is an established procedure for rotating the highest office (less personalist) and (5) whether the party faces multiparty competition or holds intraparty elections (less personalist). These characteristics, which define personalist regimes, are similar to the concepts in Gehlbach and Keefer’s proxies for party institutionalization: whether the party pre-dates the ruler and the party survives leader transitions (age of ruling party), regular (rather than irregular) leader transitions and legislative competitiveness.

Gehlbach and Keefer’s measures thus capture the same variation in autocracies as an indicator of personalist rule. For example, Figure A1 in the Appendix shows that: (1) ruling parties tend to be much older than individual leaders in party and military regimes but less so in personalist onesFootnote ⁴⁰ and (2) the share of regular leader turnover (as a fraction of all leader turnover) is lowest in personalist regimes. Thus independent measures of similar concepts are correlated in a way that suggests that the presence of a legislature, at least in this historical data, measures a different level of ‘constraint’ across personalist and non-personalist dictatorships.

Whereas Svolik concentrates on the informational role of legislatures in augmenting monitoring, Gehlbach and Keefer stress the logic of collective action. Both studies, however, point to the same group of autocracies in which formal political institutions should be weak. Our goal is not to disentangle these two logics empirically, but rather to show that the influence of ‘off-the-shelf’ cross-national measures of institutional constraint are likely to differ across distinct autocratic contexts. In doing so, we build on prior studies that point to a similar group of dictatorships in which institutions are unlikely to facilitate credible power-sharing constraints: personalist autocracies. We demonstrate that a leader’s ability to dominate formal institutions differentiates personalist regimes from other types of autocracy and modifies the relationship between authoritarian legislatures and expropriation risk.

Verifying and Extending Extant Research

To illustrate the importance of domestic political context, we revisit a central finding in JMW, which suggests that legislatures are not correlated with international investors’ perceptions about property protection. The dependent variable is a transformed measure of political risk provided by the Belgian political risk insurance agency (ONDD) for 2002. The measure is a seven-point rating for a ‘fifteen-year, forward looking insurance contract’, for which higher values represent a lower expropriation risk. According to JMW, the data ‘are representative of the political risk insurance ratings, as ONDD serves as a price leader in that industry’.Footnote ⁴¹ The main explanatory variable is a binary indicator for whether a legislature exists (Legislature).Footnote ⁴² The sample contains countries coded by Hadenius and Teorell as non-democratic in 2002.Footnote ⁴³

In this analysis, a positive coefficient indicates a correlation with greater perceived property protections (or lower political risk). The estimates in Column 1 show results identical to Column 1 of Table 1 in JMW: legislatures are not associated with greater perceived property protection. This result also holds when we restrict the sample to countries coded as autocracies in 2002 by Geddes, Wright and Frantz (hereafter GWF).Footnote ⁴⁴ When we add a dummy variable for personalist regimes and interact it with Legislature, however, the results show that the marginal effect of legislatures is different in non-personalist autocracies. In these autocracies, Legislature is associated with lower levels of expropriation risk.

Table 1 Replication of Jensen, Malesky and Weymouth (Reference Jensen, Malesky and Weymouth2014)

Note: JMW = Jensen, Malesky and Weymouth; GWF = Geddes, Wright and Frantz. The dependent variable is Risk of government expropriation from a Belgian political risk insurance agency. A positive coefficient indicates a correlation with greater perceived property protections (or lower political risk). Standard errors in parentheses. *p<0.1, **p<0.05, ***p<0.01

The finding in Column 3 is nevertheless fragile, as there is little variation in legislatures in non-personalist regimes in 2002.Footnote ⁴⁵ Furthermore, the specification in Column 3 only estimates the average marginal effect in each group of autocracies. As a result, we estimate the model in Column 1 (that is, the full sample from JMW) using a kernel regularized least squares approach. Hainmueller and Hazlett explain that this method ‘constructs a flexible hypothesis space that uses kernels as radial basis functions and finds the best-fitting surface in this space by minimizing a complexity-penalized least squares problem’. They further note that this approach ‘avoids strong parametric assumptions, yet allows interpretation in ways analogous to generalized linear models while also permitting more complex interpretation to examine non-linearities and heterogeneous effects’.Footnote ⁴⁶ This estimator allows us to explore how the marginal effect of legislatures varies in different groups of regimes, which is similar to estimating interaction terms.

The top panel of Figure 1 shows how the estimated marginal effect of Legislature (vertical axis) varies across levels of GDP per capita (horizontal axis). This plot is similar to estimating an interaction term between Legislature and GDP Per Capita to examine heterogeneous effects; in effect, we are ‘exploring’ whether the marginal effect of Legislature varies by GDP Per Capita. Each point represents a single observation in the sample, and the solid dark line is a Lowess curve derived from these points. In relatively low-income autocracies, the marginal effect of legislatures is generally positive (associated with lower expropriation risk), but these estimates are much smaller (and even negative) in high-income autocracies. The bottom panel shows the same plot divided between personalist and non-personalist regimes, as well as countries not coded as autocracies by GWF in 2002. The plots show that the marginal effect of Legislature differs considerably between personalist and non-personalist autocracies. In the non-personalist sample, for example, all but one of the observations lie above zero, indicating that the marginal effect of Legislature in this group is almost always positive, irrespective of GDP per capita. Among wealthier personalist regimes, however, the marginal effect of Legislature is negative.

Fig. 1 Pointwise marginal effect of Legislature on expropriation risk, by subsample Note: kernal regression estimates replicating the analysis in Jensen, Malesky and Weymouth (Reference Jensen, Malesky and Weymouth2014). Each point represents the marginal effect of Legislature for a single observation in the sample. The top panel shows that in the full sample, the marginal effect of Legislature varies by level of development (GDP per capita). The bottom three panels divide the sample into personalist autocracies, non-personalist autocracies and non-autocracy. The pointwise marginal effect of Legislature is almost always above zero in the non-personalist dictatorships (bottom, middle plot).

As a second way to address the concern that the verification sample has only one non-personalist regime that lacks a legislature in 2002, we examine new data on expropriation risk for the period 2002–08.Footnote ⁴⁷ This strategy increases the number of non-personalist regimes in the sample without a legislature. The results from both approaches indicate a strong negative correlation between legislature and expropriation risk in non-personalist dictatorships; in personalist ones, the estimates for legislatures vary around zero and are never statistically significant. This evidence corroborates the findings from our verification of the JMW study that examines expropriation risk data for only one year: 2002.

Expropriation

To move beyond a ‘blunt theoretical discussion of property rights and investment risk’, we use expropriation data in 138 countries between 1960 and 2006 to examine the influence of autocratic legislatures on nationalizations.Footnote ⁴⁸ This approach differs from prior studies that either look at alternative economic outcomes such as growth and investment or the perceived risk of expropriation.Footnote ⁴⁹

Instead, we use data on oil expropriation to explore whether observed property rights protections are upheld by authoritarian assemblies and whether this differs by authoritarian regime. Examining oil expropriation allows us to more precisely identify the countries in which expropriation in a particular sector is most likely to occur, because we have excellent data on oil production across a range of autocracies. We should not expect legislatures – or any other measure of political constraint – to make much difference for oil sector expropriation in countries such as Mali, Uganda or Senegal, where oil production is limited. The data are not as precise for other sectors, however, which limits our ability to assess the extent to which autocratic economies exploit them.

Further, because governments cannot choose whether their countries have oil reserves, investment in the oil sector is less likely to solely be a function of whether the incumbent government creates a friendly investment environment. Thus unlike investment in other sectors, such as manufacturing and services, the geological distribution of oil reserves can explain why we observe oil investment in markets that are potentially prone to expropriation in the first place. That is, oil investment can be explained, in part, by factors unrelated to governments’ political decisions. For example, historically, oil investment flowed to Iraq and not Jordan – two countries whose first rulers at ‘independence’, Faisal and Abdullah, came from the same Hashemite family and were imposed by the British at the end of the First World War – because oil reserves were found in the former but not the latter.Footnote ⁵⁰

A drawback of using oil expropriation data is that there are relatively few instances of observed expropriation in recent decades. In fact, from 1996 to 2002 – the period under consideration in the JMW study – there are no instances of oil sector expropriation in autocracies, according to the GKS data. Investors surveyed about their perceptions of expropriation risk during this period may not have been thinking about these types of expropriations simply because there were none. There were, however, a number of high-profile expropriations that occurred after the period of study in JMW – for example, in Russia and Venezuela in 2006. While the number of expropriations decreased substantially after 1980 – perhaps the result of the rising Washington consensus and the attendant era of neo-liberal economic policy makingFootnote ⁵¹ – seven such incidents have occurred in the last decade.Footnote ⁵² On its face, two factors may explain the recent increase in oil expropriations: the sharp rise in world oil prices starting in the late 1990s and the end of the Washington consensus.Footnote ⁵³

To mitigate concerns about the dearth of observed oil expropriations in the past couple of decades, we complement our analysis with a model that includes expropriation data in all sectors – including agriculture, manufacturing and mining. Expanding the analysis to all sectors increases the number of observed expropriations from six to nearly fifty in the 1990–2006 period. All the same, analyzing expropriations in all sectors cannot isolate those autocracies with economies that are reliant on a particular sector and which therefore have the strongest incentive to expropriate assets in that particular sector. For this reason, we view all expropriations in conjunction with oil expropriation specifically.

Autocratic Legislatures and Oil Expropriation

Our approach builds on the analysis by GKS of oil sector nationalization: the forced divestment of foreign-owned oil companies.Footnote ⁵⁴ They use a sample of democracies and autocracies between 1960 and 2006 to examine whether executive constraint and oil prices influence expropriation. Consistent with theoretical expectations, they find that higher levels of executive constraint are associated with less expropriation, and that expropriations occur when oil prices are high.

Over the same period, we restrict our analysis to autocracies defined by GWF. The sample therefore excludes democracies, provisional regimes, and states governed by warlord groups or failed states (for example, Afghanistan between Najibullah’s regime and Taliban rule, and Somalia after the fall of Siad Barre). The dependent variable is a binary indicator of oil nationalization in a particular country-year. The control variables include GDP per capita, population size, regime duration and oil rents per capita – all logged and lagged by one year.

Like GKS, we test whether executive constraint (the XCONST variable from Polity) is associated with expropriation risk. This variable is a seven-point scale in which higher values indicate more executive constraint.Footnote ⁵⁵ To examine the influence of legislatures (Legislature), we use a variable from Cheibub, Gandhi and Vreeland indicating whether an elected legislature existed on 31 December of the calendar year prior to the observation year.Footnote ⁵⁶ We view both of these variables as plausible proxies for executive constraint. Indeed, the mean level of political constraint differs by Legislature for the full sample of autocracies, in a sample of personalist regimes and in non-personalist autocracies.Footnote ⁵⁷

We estimate a linear probability model using country (ξ _i ) and year (τ _t ) fixed effects.Footnote ⁵⁸ Because world oil price only varies by calendar year, the year-fixed effects absorb this information. We first examine a sample that includes all autocracies, and then we test the same model on two subsamples that differentiate between personalist and non-personalist regimes.Footnote ⁵⁹ The model estimates the following equation with a linear link function:

(1) $$Y_{{i,t}} =\alpha _{0} {\plus}\beta X_{{i,t{\minus}1}} {\plus}\gamma Legislature_{{i,t{\minus}1}} {\plus}\xi _{i} {\plus}\tau _{t} {\plus}{\varepsilon}_{{i,t}}. $$

The left panel of Figure 2 shows the main result for XCONST.Footnote ⁶⁰ In the full sample of autocracies, executive constraint has a small average effect, and the estimate is not statistically different from zero. In the subsample analysis, however, executive constraint is associated with a lower expropriation risk in non-personalist regimes but a higher risk in personalist regimes. The right panel repeats the analysis but employs the legislature variable in lieu of executive constraints. The average effect in the sample of all autocracies, while negative, is not distinguishable from zero. Again, however, this average estimate varies across the subsamples; it is negative and different from zero in non-personalist autocracies, and positive and different from zero in personalist regimes.Footnote ⁶¹ Thus, according to both measures of political constraint – XCONST and Legislature – a consistent pattern emerges. Institutional variables are associated with a lower risk of expropriation in non-personalist dictatorships, but a higher risk of expropriation in personalist regimes. This suggests that standard cross-national measures of political constraint may not provide good measures of this concept in personalist autocracies.

Fig. 2 Political constraints and oil expropriation in autocracies Note: the left panel depicts the marginal effect of a two-standard-deviation increase in XCONST, from the full sample and subsample tests. The right panel depicts the marginal effect of a one-unit change in Legislature from 0 to 1, from the full sample and subsample tests. Years: 1960–2006.

Kernal Regression Estimates

The subsample analysis in Figure 2 estimates the average marginal effect of legislatures within each group of autocracies. This is similar to estimating a model with an interaction term, except that in addition to estimating the heterogeneous effect of legislatures, the subsample approach allows the estimates of the covariates to vary by group of regimes. Nevertheless, even this approach may not identify the best model specification, given the data. The results in Figure 2 rely on a linear model to avoid separation issues that arise when non-linear estimators drop countries (and years) in which no expropriation is observed. However, a linear model can easily produce heteroskedastic errors with a binary dependent variable, and often yields non-sensical (unbounded) predicted values. One approach that addresses both concerns is a kernel regularized least squares estimator,Footnote ⁶² which estimates the pointwise marginal effect of covariates for each observation in the sample.

We estimate the specification in Equation 1 but exclude country and year dummies to obtain convergence, replacing the latter with a flexible quadratic time trend. To model unit fixed effects using kernel regression, we add the unit means of the explanatory variables to the right-hand side of the equation. This approach, which is similar to the correlated random effects estimator discussed in the Appendix, accounts for unobserved, time-invariant country-specific factors by ‘controlling for’ unit means.Footnote ⁶³ Thus instead of including a dummy variable for each country, we add the in-sample country mean for each explanatory variable.

Using a kernel regularized least squares estimator for the full sample of all autocracies allows us to examine whether there is heterogeneity in the marginal effects of the main variable of interest, Legislature. We do this in two ways. First, we examine the difference in the mean values of the pointwise derivatives for Legislature, by personalism. We find that the average marginal effect is roughly zero in personalist regimes but −0.029 in non-personalist regimes.Footnote ⁶⁴ Second, we do this for oil rents and find that the marginal effects for Legislature vary considerably by the level of oil rents. This should not be surprising, as we would not expect legislatures to influence the risk of oil sector expropriation in countries with little oil income. The marginal effect of legislatures thus varies by both regime type and the size of oil rents.

Figure 3 plots the pointwise marginal effects against logged oil rents, by subsample. Each circle represents an individual observation. The solid horizontal line marks zero, while the dashed line indicates the average marginal effect in each subsample. In personalist autocracies, the average marginal effect of Legislature is roughly zero, while in non-personalist regimes the average effect is −0.029. Consistent with the results in Figure 2, legislatures in non-personalist regimes are associated (on average) with a lower expropriation risk. The thick dark solid line in each plot is a local area estimate of the relationship between oil rents and the marginal effect of Legislature. In countries with no oil rents, the pointwise marginal effects are roughly zero in each subsample. As oil rents increase, however, the local area estimate increases in personalist regimes but decreases in other dictatorships. This suggests that not only does the average marginal effect of legislature differ by regime type, but that these differences get stronger as oil rents increase.

Fig. 3 Pointwise marginal effect of Legislature on oil expropriation, by subsample Note: kernal regression estimates from a model with unit means of all explanatory variables to proxy for fixed effects. Years: 1960–2006.

Robustness Tests

Table 2 shows the estimates associated with Legislature, by subsample, from robustness tests. In all changes to the estimator or specification, the main pattern persists: legislatures are associated with a lower risk of expropriation in non-personalist regimes. While many robustness tests indicate that legislatures are positively correlated with observed expropriation in personalist autocracies, we do not interpret this as evidence that legislatures increase the risk of expropriation in these dictatorships. Rather, we stress that the findings indicate that the effect of legislatures in personalist contexts is different from that in non-personalist dictatorships. The Appendix discusses two robustness tests in detail: correlated random effects and 2SLS-IV. In the latter, we use Inherited Legislature for a particular leader as an excluded instrument to model exogenous variation in Legislature, an approach similar to JMW.Footnote ⁶⁵

Table 2 Robustness Tests

Note: bold estimates are not statistically significant at the 0.10 level or less. Standard errors in parentheses. Linear probability model, unless otherwise noted. All specifications include country, year fixed effects, oil rents per capita (log), GDP per capita (log), population (log) and regime duration, unless otherwise noted. No control variables $\,\equiv\,$ no added variables except country and year effects. Oil price varies by year, so these specifications exclude year effects. Oil shock years are a binary indicator for the years 1973, 1980 and 1998. Regime fixed effects substitute regime-specific dummies for country dummies. (A non-democratic spell can have multiple regimes, for example Iran has two regimes in the sample period: pre-1980 and post-1979.) Pure regime types specification drops all hybrid regimes from the sample. See the Appendix for details of correlated random effects, 2SLS-IV and all expropriations tests.

Expropriations from All Sectors

To examine all expropriation acts, we obtained data from Hajzler, who collected information on expropriation acts from Kobrin and Minor and updated the dataset through 2006, creating a list of nationalizations from 1960 to 2006.Footnote ⁶⁶ This data includes expropriation acts in the petroleum sector as well as other economic sectors such as agriculture, manufacturing, mining and utilities. In the main sample of autocratic regimes, both personalist and non-personalist (N = 2,886), the number of expropriation observations rises from sixty-one in the petroleum sector to 207 in all sectors. As before, we report estimates from a specification with country and year fixed effects as well as controls for oil rents per capita, GDP per capita, population size and regime duration. The first row of Table 3 reports this result. The next estimate drops the oil rents variable to increase the sample size. The third estimate drops all control variables, save regime duration, to maximize the sample size. A consistent pattern remains: the estimates for Legislature are negative and different from zero in non-personalist autocracies, while the estimates in personalist dictatorships are small and never different from zero. Finally, because the Hajzler data contains a slightly different list of oil sector expropriations than the list used by GKS, we estimate the main oil expropriation model with both linear and non-linear link functions using oil expropriation data from Hajzler.Footnote ⁶⁷ These estimates are reported in the final row and are consistent with results reported throughout.

Table 3 All Expropriations

Finally, we estimated a kernal regression and constructed a plot of the marginal effect of Legislature, by personalism and over time. Figure 4 shows that the average marginal effect is over twice as large in non-personalist regimes. Further, there is a strong non-linear time trend in the marginal effect. Over time, the (absolute) marginal effect of legislatures is diminishing in personalist regimes but growing in non-personalist ones. For example, in the post-1989 period, these estimates indicate that non-personalist legislatures are associated with a 10 per cent reduction in the likelihood of asset expropriation, while personalist legislatures have almost no empirical relationship with this outcome. Robustness tests confirming this result using a 2SLS estimator are reported in the Appendix.

Fig. 4 Pointwise marginal effect of Legislature, by subsample Note: kernal regression estimates from a model of all expropriations, with unit means of all explanatory variables to proxy for fixed effects. Years: 1960–2006.