Online Appendix: Institutional characteristics and regime survival: Why are semi-democracies less durable than autocracies and democracies? September 22, 2014 A Online Appendix This online appendix consists of four sections. The first presents analysis and discusses whether there are structural changes, over time, in the durability of autocracies, democracies and semi-democracies. The second section presents metrics and discussions on the performance of the different models presented in the paper. The third contains a number of models mentioned in the paper – but which are not reported in tables – and a selection of the many robustness tests that we have performed. The fourth section contains a table listing all the regimes that either logically could not liberalize or could not de-liberalize according to the main regime change definition applied in the paper. A.1 Identifying structural breaks in the durability of regime types: Change point models In the paper, we briefly discussed the possibility that the relationships between regime-type characteristics and regime durability may have changed over time. Although it is beyond the scope of the paper, or this appendix, to analyze what may cause such changes – changes in the international context and changes in the institutional structures of the “typical” democracy, semi-democracy and autocracy are two plausible types of explanatory factors – the descriptive statistics represented in Figure 1 provided strong suggestions that the expected durability of different regimes are time-contingent. Below, we briefly analyze whether such time-contingencies exist and, if yes, when changes likely occurred by investigating whether there are identifiable structural breaks in the time series of regime survival for democracies, autocracies and semi-democracies. As in the paper, we use the data from Gates et al. (2006); the sample runs from 1800–2000. Further, we employ the Bayesian change point i model developed by Park (2013).1 This is a simple linear model, and the posterior likelihood we sample from is 0 yt = xt βi + I(st = i)εt , i = 1, ..., k (1) where k is the number of states, or change points, I(st = i) is an indicator function that equals 1 in state t and 0 otherwise, xt βi is the set of independent variables, and εt is the error. The model is estimated using vague priors, and the results are represented in Figures A.1 (democracies and autocracies) and A.2 (semi-democracies). Figure A.1: Change points in democratic (left) and autocratic (right) regime duration ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● Posterior Regime Probability ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● 1.0 1.0 Posterior Regime Probability ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● 0.8 ● 0.8 ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●● ● ● ● ● ● ● ● ● 0.6 Pr(St= k |Yt) ● ● ● ● ● 0.4 0.6 ● ● 0.4 Pr(St= k |Yt) ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● 1850 1900 ● ● ●●●●●●●●●●●●●● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●●●●●●●●● ●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● 0.0 0.0 ● 1800 ● 0.2 0.2 ● ● 1950 1800 Time 1850 1900 1950 2000 Time Bayesian change point model. Probability of change (y-axis) across years (x-axis). Data source: Gates et al. (2006) The left panel in Figure A.1 shows the result for democracies and the right panel for autocracies. The x-axes represent the temporal dimension, and the y-axes show the probability of the relationship with regime survival holding a particular value. Hence, the declines and increases of the different lines identify where we observe structural breaks for the different regime types. Three change points are located for both democracies and autocracies, 1 Most often, political scientists assume that the relationship between a set of variables is constant over time. When taking the possibility of variation in relationships across time into consideration, this is most often modeled by using a standard regression framework and including a set of dummy variables (and interactions between these dummies and the relevant independent variable) marking points in time in which the relationship is expected to have changed. Change point models, in contrast, explicitly look for variation in the relationship between variables, and can be used to estimate both whether structural breaks in the time series have occurred and, if yes, how many and when these breaks are temporally located (Western and Kleykamp 2004). ii indicating that the 200 years can appropriately be divided into four time periods.2 For democracies the first period ended in 1830, and an intermittent period stretches from then to around 1890. These are periods with a small sample of democracies, and resulting large fluctuations in average regime duration (as a result of particular democracies entering or exiting the sample). The third period runs from 1890 to 1940, and observes, at first, a sharp decrease followed by an increase in mean duration. In parts of this period, mean longevity of democracies is the same as for semi-democracies. Finally, the fourth period starts around the time of the Second World War and continues throughout the remainder of the dataset’s time series. This is a period of democratic consolidation, with a steady increase in the longevity of democracies. In this period, democracies are also by far more durable than the other regime types. Hence, consistent with the argument made by Svolik (2013), we observe the strongest correlation between having a democratic regime-type and long regime duration in the post-war period characterized by steady economic growth. The robust finding from both this paper and the literature at large, namely that democracy is the most stable regime type, is mainly driven by the strong relationship in this period. In this regard, it is notable that the size of the democracy time ratio is higher in the Gates et al replication model run on the sample starting in 1972 (Model B1, Table 3) than in replication models employing longer time series. The right panel in Figure A.1 shows the result for autocracies. For these regimes, our model places the change points just before 1850, during the First World War, and in the 1950s. The period up until 1850 is one of consistent increase in autocratic durability. This trend ends, however, in 1850; from then until the First World War the average duration of autocracies is fairly constant. Following the First World War, the average survival time for autocracies falls sharply, before it levels of in the fourth and final period starting in the 1950s. The moderately high survival time of autocracies reported in our paper is basically a result of a weighted average over these four periods, but with earlier periods weighing heavier than for democracies as autocracies used to be more numerous. Interestingly, the size of the autocracy time ratio is lower in the Gates et al replication model run on the shorter sample starting in 1972 than when including more extensive time series, and the difference between the democracy and autocracy time ratios is much larger in this shorter sample. Figure A.2 shows the result of the change point model for semi-democracies. In contrast to what was the case for democracies and autocracies, the model only identifies two change points for semi-democracies. These change points are placed around 1870 and just before 1940. The period up until 1870 contains a considerable amount of fluctuation in the average duration of semi-democracies, with the period between 1870 and 1940 being much more stable. Starting around 1940 the average duration of semi-democracies drops by around 50 percent and then levels of and remains at a very low level throughout the time period covered by the dataset. Finally, we should note that the change in estimated covariates is much greater for autoc2 The bayesian change point model recovers the full posterior, meaning that the uncertainty around the exact location of the change points also is available. For all three regime types, however, the high probability density intervals around the estimated change points are so narrow that, for the present discussion, we ignore the uncertainty. iii Figure A.2: Change points in semi-democratic regime duration 1.0 Posterior Regime Probability ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● 0.8 ●● ● 0.6 ● ● ● 0.4 Pr(St= k |Yt) ● ● ● 0.2 ● ●● ● 0.0 ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● 1800 1850 1900 1950 2000 Time Bayesian change point model. Probability of change (y-axis) across years (x-axis). Data source: Gates et al. (2006) racies and democracies, indicating that pooling all country-years together and not accounting for structural breaks is less problematic for semi-democracies, and more problematic for democracies and autocracies. One consequence of the above-discussed temporal heterogeneity is that models estimated on the full time series, and that do not properly for the period-specificity of effects, may either under- or overestimate the longevity of democracies and autocracies. This is further evidenced by looking at the predictive performance of the different models presented in the paper, to which we now turn. iv A.2 Evaluating model performance In this section, we evaluate the relative performance of the various models presented in the paper in terms of explaining regime duration (see Ward, Greenhill and Bakke 2010). More precisely, we examine the predicted survival times for democracies, autocracies and semi-democracies from the different models, and evaluate how they correspond with actual survival times. To this end, the centre column of Table A.1 reports the observed median duration time of semi-democracies, autocracies, and democracies as defined by Gates et al. (2006) for the relevant samples of the different models. The left column reports the predicted median survival times for the three regime types for all the models discussed, starting with the Gates et al. (2006) replication. A well performing model makes predictions about median survival time that match, as closely as possible, the observed median survival time of the different regime types. Table A.1 also reports the Root Mean Squared Error (MSE; third column) of each model. The root MSE is the square root of the mean of the squared difference between the predicted and observed survival time for all regime types. This statistic is reported in years, indicating, for instance, that the average error in predicted survival time from the Gates et al. (2006) replication model (from Table 1) is 7 years. This column then does not show how well a model performs in explaining the duration of, respectively, semi-democracies, autocracies or democracies, but how well the model explains regime duration in general. Interestingly, the root MSE is the lowest in models incorporating the Geddes (1999) (7 years) and Hadenius and Teorell (2007) (9 years) regime dummies. These models do not perform particularly well for what they are intended to – explaining autocratic survival, as we discuss below – but they do very well in explaining regime survival in general. At this aggregated regime level, several of the models have more or less overlapping explanatory power, with most models falling within ± 5 of the median model (Coup d’Etats). This implies that at this level of aggregation the different models do not add much to each other. When disaggregating to independently consider semi-democracies, autocracies, democracies, however, the differences are starker. What is immediately clear from Table A.1 is that every model exaggerates the survival time of democracies. The average observed duration for democracies varies between 23 and 27 years (because of different samples), yet the predicted survival times range from 41 to 70 years in the models estimated on all types of regime breakdowns (i.e. not distinguishing liberalizing- and de-liberalizing changes, which makes comparison with observed average survival problematic). In the paper we focused on explaining the instability of semi-democracies, so it might not be too surprising that the models fail to account properly for what makes democracies endure or not. Nevertheless, since we are interested in the relative longevity of semi-democracies compared to democracies and autocracies, it is somewhat worrisome that none of the models produce reasonable expectations of democracies’ duration. There is a large literature investigating the determinants of democratization or the breakdown of authoritarian regimes. The literature investigating what factors make democracies endure is, however, much scarcer – with notable exceptions such as Przeworski et al. (2000)’s finding that high levels of economic development stabilizes democracies (see also Boix and Stokes v Table A.1: Predicted Survival Time Median predicted survival (years) Gates et al Replication (Mod.2, Tab.1) Semi-democracy 6 Autocracy 9 Democracy 46 Recent Instability – Riots (Mod.3, Tab.1) Semi-democracy 5 Autocracy 9 Democracy 41 Recent Instability – Coup d’Etats (Mod.4, Tab.1) Semi-democracy 5 Autocracy 10 Democracy 49 Ruling Coalitions (Mod.5, Tab.1) Semi-democracy 9 Autocracy 20 Democracy 70 Regime Types, Geddes et al. (Mod.4, Tab.4) Semi-democracy 5 Autocracy 9 Democracy 45 Regime Types, Hadenius and Teorell (Mod.3, Tab.4) Semi-democracy 7 Autocracy 15 Democracy 50 Floor and Ceiling – Liberalization (Mod.2, Tab.6) Semi-democracy 8 Autocracy 11 Democracy 62 Floor and Ceiling – De-liberaliz. (Mod.3, Tab.6) Semi-democracy 20 Autocracy 65 Democracy 175 Floor and Ceiling – Dummy controls (Mod.4, Tab.6) Semi-democracy 5 Autocracy 8 Democracy 59 Observed survival Root MSE (model) 6 13 24 7 5 13 24 11 5 21 25 9 7 15 26 15 4 15 24 7 5 17 27 9 6 13 23 15 6 13 24 16 6 13 24 13 2003; Svolik 2008), and Houle (2009)’s result that low income inequality enhances democratic survival. One possible implication of the results from Table A.1 is that future research should pay more attention to studying and modeling the determinants of democratic survival. Furthermore, the models above – except for the ruling coalition model – underestimate the longevity of authoritarian regimes. Most models predict authoritarian regimes to live around 10 years – compared to observed median survival times around 15 years. As noted, the exception is the model in Table 1 including a measure of the duration of a regime’s vi ruling coalition. This model actually overestimates the median duration of autocracies by 5 years. Again, though, this model grossly overestimates the longevity of democracies (by 44 years). Perhaps more surprisingly, the model incorporating the non-democratic regime type dummies from Geddes, Wright and Frantz (2014) does a poor job in predicting authoritarian survival times. This model underestimate the median survival time by 6 years, whereas the replication model misses by 4 years. In contrast, the model including regime dummies from Hadenius and Teorell (2007) improves on the explanatory power of the Gates et al. (2006) model in this respect, under-predicting the median duration of authoritarian regimes only by 2 years. In contrast to what is the case for autocracies and particularly democracies, most models do an excellent job in predicting median survival times for semi-democracies. The exception, quite naturally, is the model that only investigates regime changes in an autocratic direction. Among the models investigating regime changes in “both directions”, all except two are off by one year or less. The two remaining models – the ruling coalition and Hadenius and Teorell models – over-predict the median survival time of semi-democracies by (only) two years. Independent of specification, the models employed in the paper are seemingly very good at predicting the duration of semi-democratic regimes. vii A.3 Alternative model specifications and selected robustness tests Table A.2: A selection of robustness tests for Model A3, Table 1; different estimation techniques and regime measures Autocracy Democracy Cox model Baseline Instability ∗∗∗ 0.540 0.551∗∗∗ (-5.95) (-5.75) ∗∗∗ 0.252 0.248∗∗∗ (-8.97) (-8.98) Polity Polity2 Polity Baseline Instability 1.029∗∗∗ (6.32) 1.011∗∗∗ (4.75) 1.030∗∗∗ (6.53) 1.011∗∗∗ (4.97) SIP SIP2 SIP Baseline Instability 0.00200∗∗∗ (-8.06) 1266.4∗∗∗ (9.31) 0.00209∗∗∗ (-8.18) 1257.2∗∗∗ (9.57) 1.261∗∗∗ (4.77) 1.170∗∗∗ (4.86) 1.014 (1.64) 0.384∗∗ (-3.28) 1.404 (1.59) 1424.2 -701.1 0.651 5426 587 431 0.546 (-1.46) 1.620 (1.78) 0.654∗∗ (-3.29) 1.274∗∗∗ (5.06) 1.167∗∗∗ (4.84) 1.012 (1.43) 0.420∗∗ (-2.99) 1.514∗ (1.98) 1412.3 -692.2 0.641 5426 587 431 logit(SIP) logit(SIP2 ) Pressure to Democratize 0.792∗∗∗ (-4.90) 0.890∗∗∗ (-4.43) 0.978∗∗∗ (-3.40) 3.082∗∗∗ (4.73) 0.697∗ (-2.37) 5417.4 -2699.7 2.039 (1.64) 0.575∗ (-2.10) 1.410∗∗ (3.03) 0.781∗∗∗ (-5.04) 0.884∗∗∗ (-4.63) 0.979∗∗ (-3.15) 2.869∗∗∗ (4.40) 0.661∗∗ (-2.74) 5408.7 -2692.3 6520 687 516 6520 687 516 SIP change Past instability GDP p.c. GDP p.c. squared GDP p.c. growth Neighboring regimes First polity AIC ll Gamma N Polities Failures 1.226∗∗∗ (3.60) 1.190∗∗∗ (4.76) 1.021∗ (2.18) 0.263∗∗∗ (-4.87) 1.672∗ (2.43) 1568.8 -773.4 0.650 6340 666 495 0.513 (-1.79) 1.462 (1.69) 0.645∗∗∗ (-3.41) 1.248∗∗∗ (3.99) 1.188∗∗∗ (4.82) 1.018∗ (1.99) 0.287∗∗∗ (-4.63) 1.838∗∗ (3.07) 1556.6 -764.3 0.635 6340 666 495 ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 t statistics in parentheses. Time-period dummies are omitted from table viii logit(SIP) Baseline Instability 1.239∗∗∗ (5.66) 1.107∗∗∗ (5.29) 1.236∗∗∗ (3.96) 1.176∗∗∗ (4.78) 1.015 (1.61) 0.311∗∗∗ (-3.67) 1.425 (1.53) 1448.0 -713.0 0.672 5426 587 431 1.247∗∗∗ (5.98) 1.107∗∗∗ (5.21) 0.520 (-1.57) 1.611 (1.72) 0.657∗∗ (-3.13) 1.247∗∗∗ (4.21) 1.174∗∗∗ (4.80) 1.013 (1.42) 0.337∗∗∗ (-3.47) 1.548 (1.88) 1437.1 -704.6 0.661 5426 587 431 Table A.3: A selection of robustness tests for Model A3, Table 1; different sets of control variables Autocracy Democracy Standard Baseline 1.847∗∗∗ (5.83) 3.806∗∗∗ (8.84) 1.290∗∗∗ (5.47) 1.179∗∗∗ (5.64) 1.017∗ (2.01) 0.348∗∗∗ (-4.06) 1.676∗∗ (2.58) controls Instability 1.850∗∗∗ (5.94) 3.856∗∗∗ (9.19) 0.561 (-1.61) 1.572 (1.85) 0.659∗∗∗ (-3.59) 1.309∗∗∗ (5.86) 1.178∗∗∗ (5.67) 1.015 (1.82) 0.381∗∗∗ (-3.80) 1.792∗∗ (3.04) 1.018∗ (2.22) 0.256∗∗∗ (-5.15) 1.695∗ (2.54) 1.016∗ (2.04) 0.279∗∗∗ (-4.96) 1.818∗∗ (3.04) 1687.9 -833.0 0.645 6520 687 516 1674.5 -823.3 0.631 6520 687 516 1715.1 -847.5 0.664 6520 687 516 1702.6 -838.3 0.651 6520 687 516 Pressure to Democratize SIP change Past instability GDP p.c. GDP p.c. squared GDP p.c. growth Neighboring regimes First polity Drop squared GDP Baseline Instability 1.851∗∗∗ 1.866∗∗∗ (5.70) (5.85) 4.224∗∗∗ 4.307∗∗∗ (9.21) (9.71) 0.557 (-1.67) 1.610∗ (2.05) 0.663∗∗∗ (-3.39) 1.173∗∗ 1.196∗∗∗ (3.26) (3.72) Drop neighboring regimes Baseline Instability 2.023∗∗∗ 2.013∗∗∗ (6.70) (6.79) 3.775∗∗∗ 3.847∗∗∗ (8.55) (8.92) 0.563 (-1.60) 1.574 (1.87) 0.641∗∗∗ (-3.83) 1.307∗∗∗ 1.325∗∗∗ (5.61) (5.99) 1.211∗∗∗ 1.206∗∗∗ (6.71) (6.70) 1.016 1.014 (1.87) (1.69) 1.696∗ (2.57) 1.816∗∗ (2.99) 1702.0 -841.0 0.653 6520 687 516 1686.7 -830.3 0.638 6520 687 516 Latin America W. Eur., N. Am., Oceania Eastern Europe M.East and N. Afr. Western Africa Eastern and Central Africa Southern Africa South and Central Asia AIC ll Gamma N Polities Failures Add region dummies Baseline Instability 1.844∗∗∗ 1.865∗∗∗ (5.85) (6.06) 3.901∗∗∗ 3.969∗∗∗ (8.27) (8.66) 0.562 (-1.65) 1.563 (1.94) 0.672∗∗ (-3.24) 1.343∗∗∗ 1.349∗∗∗ (4.70) (4.86) 1.178∗∗∗ 1.173∗∗∗ (5.25) (5.21) 1.018∗ 1.016 (2.12) (1.95) ∗∗∗ 0.400 0.423∗∗∗ (-3.50) (-3.37) ∗ 1.575 1.697∗∗ (2.25) (2.70) 1.636∗ 1.442 (2.16) (1.60) 1.065 0.965 (0.31) (-0.18) 0.961 0.918 (-0.14) (-0.30) 1.159 1.175 (0.53) (0.60) 1.180 1.096 (0.66) (0.38) 1.052 0.918 (0.21) (-0.37) ∗∗ 1.901 1.661∗ (2.76) (2.20) 0.945 0.997 (-0.22) (-0.01) 1688.0 1677.0 -825.0 -816.5 0.636 0.624 6520 6520 687 687 516 516 ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Exponentiated coefficients; t statistics in parentheses. Time dummies are omitted from table. Reference category for region dummies is Eastern and South-East Asia. ix Table A.4: Models employing continuous SIP and including various regime categories – Regime Survival 1975–2005 SIP SIP squared G Monarchy G Single-party G Personalist G Military (1) Geddes Regime type 0.00397∗∗∗ (-9.42) 1006.9∗∗∗ (10.61) 1.553 (1.23) 1.615∗∗ (2.94) 1.317 (1.48) 0.951 (-0.26) (2) H.T. Regime type 0.00829∗∗∗ (-5.63) 489.0∗∗∗ (6.42) 3.172∗∗ (2.58) 0.722 (-1.16) 2.248∗ (2.14) 0.879 (-0.52) 0.742 (-0.65) HT Monarchy HT Military HT One-party HT Multiparty HT No-party Hegemonic Competitive GDP p.c. GDP p.c. squared GDP p.c. growth Neighboring regimes First polity AIC ll Gamma N Polities Failures (3) Competitive author. 0.0107∗∗∗ (-5.71) 391.0∗∗∗ (6.66) 1.288∗∗∗ (5.39) 1.137∗∗∗ (4.35) 1.019∗ (2.19) 0.335∗∗∗ (-4.14) 1.443∗ (1.99) 1582.2 -777.1 0.649 5997 629 474 1.159∗ (2.52) 1.130∗∗∗ (3.53) 1.034∗∗∗ (3.43) 0.294∗∗∗ (-3.87) 1.760∗∗ (3.01) 1008.2 -490.1 0.621 4070 470 317 ∗ 1.381 (1.46) 1.156 (0.76) 1.266∗∗∗ (4.36) 1.168∗∗∗ (4.40) 1.030∗∗ (2.84) 0.293∗∗∗ (-3.62) 2.006∗∗∗ (3.31) 962.4 -470.2 0.658 3708 445 292 p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Time ratios and t statistics (in parentheses) are reported. Time dummies are omitted from table. Hadenius and Teorell No-party and Other regimes are collapsed. x Table A.5: Testing first alternative explanation when disaggregating regime dimensions; only polities with open and competitive executive recruitment Participation Executive Constraints Participation*Constraints (1) Gates et al. 1900–2000 1.246∗ (1.98) 0.884 (-1.47) 1.235∗∗∗ (4.27) (2) Gates et al. 1919–2000 1.265∗ (1.96) 0.925 (-0.93) 1.232∗∗∗ (4.22) Past instability Pressure to Democratize SIP change (3) Instability (4) Coups (5) Ruling Coalition (6) Transitions (7) Shared frailty 1.249 (1.91) 0.953 (-0.58) 1.220∗∗∗ (3.96) 0.595∗ (-2.29) 0.708 (-0.79) 1.027 (0.10) 1.307∗∗ (2.91) 0.983 (-0.22) 1.187∗∗∗ (3.66) 1.140∗ (2.23) 1.024 (0.51) 1.062∗ (2.09) 1.275∗ (2.04) 0.967 (-0.41) 1.204∗∗∗ (3.82) 1.265∗ (2.17) 0.921 (-1.01) 1.231∗∗∗ (4.69) 0.0636∗∗∗ (-6.45) 1.363∗∗∗ (3.45) 1.154∗∗ (2.67) 1.010 (0.43) 0.432 (-1.73) 1.215 (0.45) 444.6 -211.3 0.620 2775 207 128 1.349∗∗∗ (3.50) 1.177∗∗ (2.89) 1.015 (0.93) 0.473 (-1.53) 1.110 (0.29) 466.9 -222.5 0.636 2775 207 128 Coups last 10 years 0.846∗∗∗ (-3.52) Ruling Coal. Duration 1.719∗∗∗ (6.73) Polity transition (-88) GDP p.c. GDP p.c. squared GDP p.c. growth Neighboring regimes First polity AIC ll Gamma N Polities Failures ∗∗ 1.308 (3.13) 1.160∗∗ (2.93) 1.016 (0.67) 0.459 (-1.63) 1.071 (0.20) 516.0 -248.0 0.646 3031 218 144 ∗∗∗ 1.343 (3.31) 1.177∗∗ (3.02) 1.015 (0.62) 0.477 (-1.50) 1.122 (0.28) 465.1 -222.6 0.646 2775 207 128 ∗∗∗ 1.379 (3.69) 1.176∗∗ (3.03) 1.010 (0.46) 0.490 (-1.52) 1.187 (0.42) 462.8 -218.4 0.625 2775 207 128 1.411 (4.05) 1.136∗ (2.48) 1.004 (0.12) 0.479 (-1.60) 0.995 (-0.02) 410.1 -194.1 0.620 2440 186 113 ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Time ratios and t statistics (in parentheses) are reported. xi ∗∗∗ ∗∗ 1.147 (2.65) 1.058∗ (2.27) 1.012 (1.14) 0.597∗ (-2.25) 1.080 (0.53) 461.3 -219.7 0.379 2874 207 135 Table A.6: Testing first alternative explanation when disaggregating regime dimensions; only polities with closed executive recruitment Participation Executive Constraints Dual executive Participation*Constraints Dual*Participation (1) Gates et al. 1900–2000 0.879 (-1.94) 0.938 (-1.39) 0.845 (-0.96) 1.058∗ (2.01) 1.247∗ (1.99) (2) Gates et al. 1919–2000 0.925 (-1.08) 0.897∗ (-2.08) 0.864 (-0.72) 1.080∗ (2.50) 1.205 (1.49) Past instability Pressure to Democratize SIP change (3) Instability (4) Coups (5) Ruling Coalition (6) Transitions (7) Shared frailty 0.902 (-1.51) 0.904∗ (-2.11) 0.872 (-0.69) 1.078∗∗ (2.64) 1.245 (1.81) 0.674∗∗ (-2.78) 0.707 (-0.75) 1.814∗ (2.22) 0.906 (-1.38) 0.880∗∗ (-2.60) 0.863 (-0.78) 1.070∗ (2.25) 1.210 (1.65) 0.881∗ (-2.58) 0.987 (-0.39) 0.855 (-1.16) 1.039 (1.87) 1.191∗ (1.99) 0.881 (-1.75) 0.889∗ (-2.39) 0.791 (-1.17) 1.077∗ (2.40) 1.256 (1.77) 0.926 (-1.16) 0.896∗ (-2.33) 0.865 (-0.83) 1.080∗∗ (2.68) 1.204 (1.66) 0.0722∗∗∗ (-7.33) 1.228∗∗ (3.02) 1.117∗∗ (2.72) 1.021∗ (2.29) 0.374∗∗ (-3.01) 1.642∗ (2.35) 1091.1 -532.6 0.605 3643 459 369 1.174∗ (2.49) 1.118∗∗ (2.77) 1.017∗ (2.28) 0.447∗ (-2.41) 1.816∗∗ (2.61) 1162.8 -568.4 0.620 3643 459 369 Coups last 10 years 0.925∗ (-2.09) Ruling Coal. Duration 1.293∗ (2.52) Polity transition (-88) GDP p.c. GDP p.c. squared GDP p.c. growth Neighboring regimes First polity AIC ll Gamma N Polities Failures ∗ 1.136 (2.32) 1.086∗ (2.31) 1.019∗ (2.32) 0.520∗ (-2.27) 2.042∗∗∗ (3.63) 1281.9 -628.9 0.619 4216 642 412 ∗ 1.174 (2.52) 1.118∗∗ (2.84) 1.017 (1.96) 0.449∗∗ (-2.60) 1.816∗∗ (2.77) 1160.8 -568.4 0.622 3643 459 369 ∗∗ 1.183 (2.78) 1.111∗∗ (2.79) 1.016 (1.90) 0.496∗ (-2.33) 1.994∗∗∗ (3.43) 1153.3 -561.7 0.606 3643 459 369 1.121 (1.90) 1.093∗ (2.43) 1.023∗ (2.49) 0.450∗∗ (-2.64) 1.890∗∗ (3.08) 1122.4 -548.2 0.619 3509 432 352 ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Time ratios and t statistics (in parentheses) are reported. xii ∗ 1.090 (2.00) 1.052 (1.77) 1.016∗ (2.45) 0.582∗ (-2.56) 1.682∗∗∗ (3.36) 1091.9 -532.9 0.475 3736 512 394 Table A.7: Dimensions of Authority by Regime Type Executive Average Proportion constraints participation dual Geddes et al. Regime Categories Single-Party 2.75 .83 (1.22) .10 Personalist 1.90 .87 (1.33) .06 Military 2.17 .70 (1.30) .09 Monarchy 1.95 .20 (.52) .52 Hadenius and Teorell Regime Categories Monarchy 1.76 .17 (.47) .68 Military 1.75 .44 (.99) .06 Multiparty 3.67 .11 (.37) .01 One-Party 2.21 2.42(1.24) .30 No-Party 2.11 .40 (1.02) .03 Gates et. al. Semi-Democracies 4.50 1.47 (1.30) .50 Autocracies 1.70 .14 (.40) .03 Democracies 6.73 3.51 (.45) 0 Note: The time series is 1946–2010 for the Geddes et al. categories and 1972– 2010 for the Hadeniues and Teorell categories. Standard deviation in parenthesis for participation. xiii Figure A.3: Regime types in the year 2000 Geddes et al. classification, 2000 Regime type Monarchy Military Party Personal Democracy Hadenius and Teorell classification, 2000 Regime type Monarchy Military One−Party Multi−party Other Democracy Map produced using CShapes (Weidmann, Dorussen and Gleditsch 2010). xiv Table A.8: Testing second alternative explanation when disaggregating regime dimensions; only polities with open executive recruitment Participation Executive Constraints. Participation*Constraints (1) Gates et al 1.372∗ (2.19) 1.074 (0.61) 1.163∗ (2.16) G Single-party G Personalist G Military (2) Geddes regimes 1.384∗∗ (2.60) 1.028 (0.23) 1.174∗ (2.49) 1.190 (0.36) 0.393∗ (-1.96) 2.796∗ (2.31) HT Military HT Multiparty HT No-party (3) H.T. regimes 1.414∗ (2.29) 0.996 (-0.03) 1.084 (0.94) 0.0735∗∗∗ (-4.85) 0.352∗ (-2.30) 0.195∗∗ (-3.24) (4) Geddes full sample 1.342∗∗ (2.79) 0.931 (-0.91) 1.228∗∗∗ (4.36) 1.446 (1.07) 0.497 (-1.92) 1.230 (0.65) Hegemonic Competitive GDP p.c. GDP p.c. squared GDP p.c. growth Neighboring regimes First polity AIC ll Gamma N Polities Failures 1.555∗∗∗ (3.55) 1.213∗ (2.53) 1.020 (0.82) 0.484 (-1.10) 1.993 (1.67) 248.6 -114.3 0.657 1611 133 60 1.456∗∗ (2.73) 1.263∗∗ (2.96) 1.021 (0.78) 0.493 (-1.03) 2.152∗ (2.20) 246.1 -110.1 0.630 1611 133 60 1.350∗ (2.17) 1.196∗∗ (2.67) 1.020 (0.81) 0.973 (-0.04) 1.732 (1.44) 213.5 -93.74 0.580 1611 133 60 ∗ 1.347∗∗ (3.13) 1.176∗∗ (3.13) 1.004 (0.14) 0.412 (-1.69) 1.118 (0.37) 437.3 -205.6 0.652 2523 190 117 (5) Competitive author. 1.362 (1.57) 0.932 (-0.47) 1.184 (1.84) 1.476 (0.47) 0.549 (-1.41) 1.501∗∗ (3.02) 1.219∗ (2.56) 1.019 (0.71) 0.579 (-0.78) 1.771 (1.42) 237.2 -106.6 0.657 1497 128 55 p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Time ratios and t statistics (in parentheses) are reported. Hadenius and Teorell No-party regimes and Other regimes are collapsed. G Monarchy and H.T. Monarchy and One-party dummies are dropped (perfect colin./no obs.). xv Table A.9: Testing second alternative explanation when disaggregating regime dimensions; only polities with closed executive recruitment Participation Executive Constraints Dual executive Participation*Constraints Dual*Participation (1) Gates et al 0.917 (-0.98) 0.897 (-1.81) 0.819 (-0.77) 1.029 (0.83) 1.318 (1.92) G Monarchy G Single-party G Personalist G Military (2) Geddes regimes 0.903 (-1.10) 0.946 (-0.67) 0.686 (-1.37) 1.032 (0.80) 1.444∗ (2.34) 3.137∗ (2.31) 1.404 (1.17) 1.366 (0.97) 1.280 (0.75) (3) H.T. regimes 0.907 (-1.07) 1.014 (0.22) 0.583 (-1.93) 1.020 (0.54) 1.621∗∗ (3.03) 19.35∗∗∗ (4.86) 3.227∗∗ (3.23) 7.615∗∗∗ (4.69) 2.453∗∗ (2.68) 3.129∗ (2.10) HT Monarchy HT Military HT One-party HT Multiparty HT No-party/Other (4) Geddes full sample 0.906 (-1.35) 0.895∗ (-2.04) 0.901 (-0.52) 1.089∗∗ (2.72) 1.227 (1.64) 1.672 (1.44) 1.618∗ (2.45) 1.428 (1.65) 0.988 (-0.05) Hegemonic Competitive GDP p.c. GDP p.c. squared GDP p.c. growth Neighboring regimes First polity AIC ll Gamma N Polities Failures 1.114 (1.39) 1.086 (1.94) 1.033∗∗ (3.17) 0.298∗∗ (-3.14) 2.503∗∗∗ (4.30) 767.3 -371.7 0.606 2459 337 257 1.068 (0.79) 1.051 (1.12) 1.034∗∗ (3.23) 0.301∗∗ (-3.10) 2.479∗∗∗ (4.07) 768.0 -368.0 0.595 2459 337 257 1.024 (0.32) 1.044 (1.01) 1.038∗∗∗ (3.77) 0.458∗ (-2.11) 1.956∗∗ (3.17) 726.7 -346.4 0.571 2459 337 257 ∗ 1.132 (1.96) 1.078∗ (2.00) 1.022∗ (2.45) 0.496∗ (-2.23) 1.898∗∗ (3.15) 1132.4 -550.2 0.612 3686 583 358 p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Time ratios and t statistics (in parentheses) are reported. Hadenius and Teorell No-party regimes and Other regimes are collapsed. xvi (5) Competitive author. 0.965 (-0.36) 0.870∗ (-2.25) 0.892 (-0.42) 1.036 (0.94) 1.273 (1.56) 1.188 (0.73) 1.026 (0.12) 1.125 (1.40) 1.101∗ (2.10) 1.032∗∗ (2.74) 0.276∗∗ (-3.17) 2.513∗∗∗ (4.21) 722.3 -347.2 0.623 2211 317 237 Table A.10: Testing third alternative explanation when disaggregating regime dimensions; polities with open and competitive executive recruitment (4 leftmost models) and polities with closed executive recruitment (4 rightmost models) Polities w. open and comp. exec. recr. Participation Executive Constr. Particip.*Constr. Baseline 1.246∗ (1.98) 0.884 (-1.47) 1.235∗∗∗ (4.27) Lib. only 1.701∗∗∗ (3.69) 0.765∗∗ (-2.71) 1.305∗∗∗ (4.08) Delib. only 0.978 (-0.15) 1.063 (0.46) 1.140 (1.77) Floor-Ceiling 1.250∗ (2.03) 0.889 (-1.39) 1.226∗∗∗ (4.09) 1.943∗∗∗ (5.52) 1.286∗∗ (3.12) 1.015 (0.48) 0.236∗∗ (-2.68) 1.372 (0.78) 378.1 -179.1 0.710 3031 218 81 2.350 (1.10) – – 1.281∗∗ (2.68) 1.148∗ (2.48) 1.016 (0.67) 0.472 (-1.52) 1.037 (0.10) 516.0 -247.0 0.654 3031 218 144 Dual executive Dual*Particip. Ceiling dummy Floor dummy GDP p.c. GDP p.c. sq. GDP p.c. gr. Neighb. regimes First polity AIC ll Gamma N Polities Failures 1.308∗∗ (3.13) 1.160∗∗ (2.93) 1.016 (0.67) 0.459 (-1.63) 1.071 (0.20) 516.0 -248.0 0.646 3031 218 144 0.882 (-0.78) 1.071 (0.85) 1.026 (1.23) 1.000 (-0.00) 0.658 (-0.83) 301.9 -141.0 0.711 3031 218 63 ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 Time ratios and t statistics (in parentheses) are reported. xvii Polities w. closed exec. recr. Baseline 0.879 (-1.94) 0.938 (-1.39) 1.058∗ (2.01) 0.845 (-0.96) 1.247∗ (1.99) Lib. only 1.085 (0.91) 0.973 (-0.51) 1.032 (0.92) 0.874 (-0.64) 1.065 (0.43) Delib. only 0.550∗∗∗ (-6.01) 0.794∗∗ (-3.13) 1.214∗∗∗ (4.35) 0.664 (-1.49) 1.644∗∗ (2.68) 1.136∗ (2.32) 1.086∗ (2.31) 1.019∗ (2.32) 0.520∗ (-2.27) 2.042∗∗∗ (3.63) 1281.9 -628.9 0.619 4216 642 412 1.011 (0.16) 1.075 (1.77) 1.031∗∗ (3.12) 0.446∗ (-2.35) 3.338∗∗∗ (5.98) 1056.5 -516.3 0.664 4216 642 287 1.394∗∗ (2.91) 1.066 (1.00) 1.002 (0.13) 1.438 (0.58) 0.777 (-0.77) 621.6 -298.8 0.760 4216 642 118 Floor-Ceiling 0.867∗ (-2.13) 0.913 (-1.82) 1.075∗ (2.41) 0.835 (-1.04) 1.256∗ (2.06) – – 0.726 (-1.79) 1.121∗ (2.04) 1.088∗ (2.43) 1.018∗ (2.24) 0.532∗ (-2.20) 1.916∗∗∗ (3.29) 1280.9 -627.4 0.615 4216 642 412 Table A.11: Estimated median survival times for different polities in models with/without past instability controls Open and competitive executive recruitment Share of population Executive constraints participating elections 1 (weak) 2-3 4-5 6-7 ≤ 1% 8.2/6.5 4.4/4.1 1.6/1.4 0.6/0.8 5% 5.0/4.0 3.6/3.3 2.8/1.6 2.8/2.8 10% 2.6/2.0 3.7/3.4 6.0/6.0 23.6/32.1 50% 1.9/1.8 6.1/6.2 16.5/17.2 74.2/100.5 Designated or ascribed executive Share of population Executive constraints participating elections 1 (weak) 2-3 4-5 6-7 ≤ 1% 10.3/11.5 7.9/7.9 4.0/3.2 2.6/2.3 5% 7.2/7.6 5.9/6.0 4.0/3.2 3.5/2.5 10% 3.6/3.7 4.0/4.3 4.5/4.7 6.1/6.6 50% 2.6/2.7 4.5/5.1 4.7/5.2 8.7/10.8 All covariates at means. Estimates are for: Baseline model/Model controlling for Pressure to Democratize, Past instability and SIP-change; see Appendix Tables A.5, A.6. xviii A.4 List of regime-observations that cannot register liberalizing or de-liberalizing regime changes Table A.12: List of regimes that cannot liberalize (Ceiling) or de-liberalize (Floor) Country Cuba Haiti Haiti Haiti Mexico Mexico Mexico Guatemala Guatemala Guatemala Guatemala Guatemala Guatemala Nicaragua Panama Ecuador Peru Peru Peru Peru Peru Peru Brazil Brazil Bolivia Bolivia Bolivia Bolivia Bolivia Paraguay Chile Argentina Uruguay Uruguay Ceiling Floor Start year Last year 0 1 1961 2000 0 1 1810 1915 0 1 1988 1989 0 1 1991 1994 0 1 1835 1846 0 1 1848 1863 0 1 1864 1867 0 1 1839 1847 0 1 1854 1871 0 1 1897 1898 0 1 1954 1958 0 1 1963 1966 0 1 1982 1985 0 1 1981 1984 0 1 1968 1978 0 1 1972 1979 0 1 1821 1822 0 1 1835 1836 0 1 1836 1839 0 1 1879 1881 0 1 1882 1886 0 1 1968 1978 0 1 1889 1894 0 1 1934 1945 0 1 1836 1839 0 1 1864 1871 0 1 1876 1879 0 1 1964 1966 0 1 1969 1982 0 1 1936 1937 0 1 1973 1989 1 0 1983 2000 0 1 1828 1856 0 1 1860 1873 Continued on next page xix Table A.12 – continued from previous page Country Ceiling Floor Start year Last year Uruguay 0 1 1875 1890 Uruguay 0 1 1898 1903 Uruguay 1 0 1989 2000 Belgium 1 0 1949 2000 Luxembourg 1 0 1945 2000 Spain 0 1 1923 1930 Spain 0 1 1939 1975 Spain 1 0 1978 2000 Portugal 0 1 1823 1833 Portugal 1 0 1976 2000 Germany 1 0 1949 2000 Hungary 0 1 1919 1920 Czechoslovakia 1 0 1946 1947 Czechoslovakia 1 0 1990 1992 Czech Republic 1 0 1993 2000 Slovakia 1 0 1993 2000 Italy 1 0 1948 2000 Yugoslavia 0 1 1830 1838 Slovenia 1 0 1991 2000 Greece 0 1 1833 1844 Greece 0 1 1967 1974 Greece 1 0 1986 2000 Bulgaria 1 0 1990 2000 Rumania 0 1 1864 1866 Rumania 0 1 1941 1944 Rumania 1 0 1996 2000 Lithuania 1 0 1991 2000 Finland 1 0 1978 2000 Sweden 1 0 1968 2000 Denmark 1 0 1945 2000 Equatorial Guinea 0 1 1969 2000 Gambia 0 1 1994 1996 Mali 0 1 1968 1991 Benin 0 1 1965 1967 Benin 0 1 1972 1990 Niger 0 1 1974 1987 Niger 0 1 1996 1996 Guinea 0 1 1984 1993 Burkina Faso 0 1 1966 1977 Burkina Faso 0 1 1980 1991 Continued on next page xx Table A.12 – continued from previous page Country Ceiling Floor Start year Last year Liberia 0 1 1980 1985 Sierra Leone 0 1 1967 1968 Sierra Leone 0 1 1992 1996 Ghana 0 1 1966 1969 Ghana 0 1 1972 1978 Ghana 0 1 1981 1991 Togo 0 1 1967 1991 Togo 0 1 1993 1994 Nigeria 0 1 1966 1978 Nigeria 0 1 1983 1998 Cent. Af. Rep. 0 1 1960 1986 Cent. Af. Rep. 0 1 1986 1993 Chad 0 1 1975 1978 Chad 0 1 1985 1996 Congo 0 1 1997 2000 Zaire 0 1 1997 2000 Uganda 0 1 1971 1979 Uganda 0 1 1986 1993 Burundi 0 1 1966 1991 Burundi 0 1 1996 2000 Rwanda 0 1 1973 1991 Rwanda 0 1 1995 2000 Somalia 0 1 1969 1991 Ethiopia 0 1 1930 1935 Ethiopia 0 1 1942 1946 Lesotho 0 1 1986 1993 Comoros 0 1 1985 1990 Comoros 0 1 1999 2000 Algeria 0 1 1992 1995 Libya 0 1 1969 2000 Sudan 0 1 1958 1964 Sudan 0 1 1989 1996 Iraq 0 1 1936 1941 Iraq 0 1 1958 1979 Syria 0 1 1951 1954 Afghanistan 0 1 1973 1979 Afghanistan 0 1 1996 2000 South Korea 0 1 1961 1963 South Korea 1 0 1988 2000 Pakistan 0 1 1958 1962 Continued on next page xxi Table A.12 – continued from previous page Country Ceiling Floor Start year Last Pakistan 0 1 1977 Pakistan 0 1 1999 Bangladesh 0 1 1975 Bangladesh 0 1 1982 Bangladesh 0 1 1990 Myanmar 0 1 1962 Nepal 0 1 1846 Thailand 0 1 1958 Thailand 0 1 1971 Cambodia 0 1 1976 Cambodia 0 1 1997 Republic of Vietnam 0 1 1963 Indonesia 1 0 1999 xxii year 1985 2000 1978 1986 1991 1980 1951 1968 1973 1979 1998 1967 2000 References Boix, Carles, and Susan C. Stokes. 2003. “Endogenous Democratization.” World Politics 55(4): 517–549. Gates, Scott, Håvard Hegre, Mark P. Jones, and Håvard Strand. 2006. “Institutional Inconsistency and Political Instability: Polity Duration, 1800–2000.” American Journal of Political Science 50(4): 893–908. Geddes, Barbara. 1999. “What Do We Know About Democratization After Twenty Years?” Annual Review of Political Science 2: 115–144. Geddes, Barbara, Joseph Wright, and Erica Frantz. 2014. “Autocratic Breakdown and Regime Transitions: A New Data Set.” Perspectives on Politics 12(2): 313–331. Hadenius, Axel, and Jan Teorell. 2007. “Pathways from Authoritarianism.” Journal of Democracy 18(1): 143–156. Houle, Christian. 2009. “Inequality and Democracy: Why Inequality Harms Consolidation but Does Not Affect Democratization.” World Politics 61(4): 589–622. Park, Jong Hee. 2013. “A Change-point Approach to Intervention Analysis Using Bayesian Inference.” Seoul National University. Working Paper. Przeworski, Adam, Michael E. Alvarez, José Antonio Cheibub, and Fernando Limongi. 2000. Democracy and Development. Political Institutions and Well-Being in the World, 1950– 1990. Cambridge: Cambridge University Press. Svolik, Milan. 2008. “Authoritarian Reversals and Democratic Consolidation.” American Political Science Review 102(2): 153–168. Svolik, Milan. 2013. “Learning to Love Democracy: Electoral Accountability and the Success of Democracy.” American Journal of Political Science 57(3): 685–702. Ward, Michael D., Brian D. Greenhill, and Kristin M. Bakke. 2010. “The perils of policy by p-value: Predicting civil conflicts.” Journal of Peace Research 47(4): 363–375. Weidmann, Nils B, Hans Dorussen, and Kristian Skrede Gleditsch. 2010. “The Geography of the International System: The CShapes Dataset.” International Interactions 36(1): 86–106. Western, Bruce, and Meredith Kleykamp. 2004. “A Bayesian Change Point Model for Historical Time Series Analysis.” Political Analysis 12(4): 354–374. xxiii