Democracy Does Cause Growth Part 4

Table A1: Transitions to democracy in our sample. Country Year Propensity score Country Year Propensity score Country Year Propensity score Albania 1992 0.1687 Guinea-Bissau 2005 0.0669 Pakistan 1972 0.0158 Albania 1997 0.0169 Greece 1975 0.0126 Pakistan 1988 0.0351 Argentina 1973 0.0279 Grenada 1984 0.0117 Pakistan 2008 0.0523 Argentina 1983 0.0411 Guatemala 1966 0.0194 Panama 1994 0.0595 Armenia 1991 n.a. Guatemala 1986 0.0283 Peru 1963 n.a. Armenia 1998 0.0129 Guyana 1992 0.0725 Peru 1980 0.0160 Azerbaijan 1992 n.a. Honduras 1982 0.0462 Peru 1993 0.1107 Burundi 2003 0.0195 Croatia 2000 0.0453 Philippines 1987 0.0195 Benin 1991 0.1196 Haiti 1990 n.a. Poland 1990 n.a. Burkina Faso 1977 0.0149 Haiti 1994 n.a. Portugal 1976 0.0180 Bangladesh 1991 0.0975 Haiti 2006 0.0505 Paraguay 1993 0.1052 Bangladesh 2009 0.0167 Hungary 1990 0.0669 Romania 1990 0.0836 Bulgaria 1991 0.1115 Indonesia 1999 0.1128 Russia 1993 0.1532 Belarus 1991 n.a. Kenya 2002 0.0386 Sudan 1965 0.0292 Bolivia 1982 0.0498 Kyrgyz Republic 2005 0.0434 Sudan 1986 0.0439 Brazil 1985 0.0263 Kyrgyz Republic 2010 0.0449 Senegal 2000 0.0467 Bhutan 2008 0.0410 Cambodia 1993 n.a. Serbia & Montenegro 2000 n.a. Central African Rep. 1993 0.1439 Korea 1988 0.0200 Solomon Islands 2004 0.0361 Chile 1990 0.0513 Lebanon 2005 0.0426 Sierra Leone 1996 0.0553 Côte d’Ivoire 2000 0.0514 Liberia 2004 0.0689 Sierra Leone 2001 0.0267 Congo, Republic of 1992 0.0758 Lesotho 1993 0.1022 El Salvador 1982 0.0823 Comoros 1990 0.0866 Lesotho 1999 0.0909 São Tomé & Prı́ncipe 1991 n.a. Comoros 1996 0.0561 Lithuania 1993 n.a. Suriname 1988 0.0592 Comoros 2002 0.0383 Latvia 1993 0.2413 Suriname 1991 0.0755 Cape Verde 1991 0.0868 Moldova 1994 0.2090 Slovak Republic 1993 0.1690 Cyprus 1974 n.a. Madagascar 1993 0.1503 Slovenia 1992 n.a. Czech Republic 1993 n.a. Mexico 1997 0.0395 Taiwan 1992 n.a. Djibouti 1999 0.1158 Macedonia, FYR 1991 n.a. Thailand 1974 0.0143 Dominican Republic 1978 0.0531 Mali 1992 0.0866 Thailand 1978 0.0473 Ecuador 1979 0.0443 Mongolia 1993 0.1734 Thailand 1992 0.0454 Spain 1978 0.0529 Mozambique 1994 0.1031 Thailand 2008 0.0485 Estonia 1992 0.0955 Mauritania 2007 0.0131 Turkey 1961 n.a. Ethiopia 1995 0.0191 Malawi 1994 0.0973 Turkey 1973 0.0275 Fiji 1990 0.0642 Niger 1991 0.1173 Turkey 1983 0.0266 Georgia 1995 0.1025 Niger 1999 0.0958 Uganda 1980 n.a. Ghana 1970 0.0193 Niger 2010 0.0581 Ukraine 1994 0.1402 Ghana 1979 0.0453 Nigeria 1979 0.0539 Uruguay 1985 0.0356 Ghana 1996 0.0435 Nigeria 1999 0.1001 South Africa 1994 0.0890 Guinea 2010 0.0564 Nicaragua 1990 0.1258 Zambia 1991 0.1177 Guinea-Bissau 1994 0.0900 Nepal 1991 0.0955 Zimbabwe 1978 0.0888 Guinea-Bissau 1999 0.1559 Nepal 2006 0.0394 Notes: This table summarizes all democratization events in our sample. Democratizations are identified as transitions from nondemocracy to democracy using our dichotomous measure. For each democratization we report the country and the year in which it took place. The table also reports the estimated propensity score of each event based on lags of GDP and our model in Column 4, in the top panel of Table A11. Here, n.a. indicates insufficient GDP data to estimate the propensity score. The overall sample probability of a democratization following a period of nondemocracy is 0.0184. A18 Table A2: Reversals to nondemocracy in our sample. Country Year Propensity score Albania Argentina Armenia Azerbaijan Burkina Faso Bangladesh Bangladesh Belarus Brazil Central African Rep. Chile Côte d’Ivoire Congo, Republic of Congo, Republic of Comoros Comoros Comoros Djibouti Ecuador Ethiopia Fiji Fiji Ghana Ghana Gambia, The Guinea-Bissau Guinea-Bissau Greece Grenada Guatemala Haiti Haiti Haiti Kyrgyz Republic Cambodia South Korea 1996 1976 1996 1993 1980 1974 2007 1995 1964 2003 1973 2002 1963 1997 1976 1995 1999 2010 1961 2010 1987 2006 1972 1981 1994 1998 2003 1967 1979 1974 1991 2000 2010 2009 1995 1961 0.0252 0.0365 0.0777 n.a. 0.3021 0.1664 0.0189 0.0268 0.0393 0.0592 0.0459 0.0261 n.a. 0.0251 n.a. 0.0484 0.0654 0.0354 n.a. 0.0984 0.0224 0.0140 0.2532 0.0721 0.0344 0.0842 0.0927 0.0289 n.a. 0.0858 n.a. 0.0462 0.0608 0.0970 n.a. n.a. Country Year Propensity score Lebanon Lesotho Madagascar Myanmar Mauritania Niger Niger Nigeria Nigeria Nepal Pakistan Pakistan Panama Peru Peru Peru Philippines Russia Sudan Sudan Solomon Islands Sierra Leone Sierra Leone Somalia Suriname Suriname Thailand Thailand Thailand Turkey Turkey Uganda Uruguay Venezuela, Rep. Bol. Zimbabwe 1975 1998 2009 1962 2008 1996 2009 1966 1984 2002 1977 1999 1968 1962 1968 1992 1965 2004 1969 1989 2000 1967 1997 1969 1980 1990 1976 1991 2006 1971 1980 1985 1972 2009 1987 n.a. 0.0537 0.1156 n.a. 0.0286 0.1383 0.1274 0.1026 0.1212 0.0696 0.1151 0.0365 0.0626 n.a. 0.0934 0.0143 0.0758 0.0050 0.1589 0.1178 0.0237 0.2412 0.0449 n.a. 0.0657 0.0276 0.1459 0.0207 0.0100 0.0340 0.0526 n.a. 0.0408 0.0090 0.1505 Notes: This table summarizes all reversal events in our sample. Reversals are identified as transitions from democracy to nondemocracy using our dichotomous measure. For each reversal we report the country and the year in which it took place. The table also reports the estimated propensity score of each event based on lags of GDP and our model in Column 4, in the bottom panel of Table A11. Here, n.a. indicates insufficient GDP data to estimate the propensity score. The overall sample probability of a reversal following a period of democracy is 0.0121. A19 Table A3: Empirical process for GDP and democracy used in the alternative counterfactual for our long-run estimates. AR(1) models (1) AR(2) models (2) AR(4) models (3) AR(8) models (4) Panel A. Democracy equation: Propensity to democratize Democracy first lag 0.025 (0.003) 0.956 (0.005) Democracy second lag 0.025 (0.002) 0.903 (0.018) 0.055 (0.017) Democracy third lag Democracy fourth lag 0.025 (0.002) 0.900 (0.018) 0.002 (0.024) 0.019 (0.028) 0.041 (0.018) Democracy fifth lag Democracy first lag Democracy seventh lag Democracy eight lag 0.026 (0.003) 0.892 (0.019) 0.000 (0.026) 0.020 (0.029) 0.005 (0.027) 0.012 (0.037) 0.046 (0.029) -0.034 (0.024) 0.024 (0.014) Panel B. Estimation of Counterfactuals: Effect of democracy after 25 years Counterfactual likelihood of democracy Counterfactual growth Effect of democracy relative to counterfactual Observations Countries in sample 17.791 (5.627) 0.382 (0.026) 4.269 (1.384) 13.521 (4.300) 6,790 175 13.800 (5.528) 0.374 (0.026) 3.368 (1.370) 10.432 (4.194) 6,642 175 16.895 (5.275) 0.366 (0.026) 4.071 (1.292) 12.824 (4.042) 6,336 175 17.715 (5.430) 0.377 (0.030) 4.417 (1.376) 13.299 (4.132) 5,688 175 Notes: This table reports a joint estimation of the GDP equation and an equation for democracy. The top panel presents estimates of the GDP equation. The bottom panel presents estimates of a model with democracy as dependent variable and lags of democracy as explanatory variables. All these models include a full set of country and year fixed effects. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A20 Table A4: Effect of democracy on (log) GDP per capita. Estimates obtained by imposing the persistence of the GDP process to lie between 0.95 and 1. Imposed persistence ρ = P γj : ρ = 0.95 (1) ρ = 0.96 (2) 0.638 (0.247) 12.750 (4.943) 11.477 (4.455) 6,336 175 0.752 (0.228) 18.811 (5.712) 15.511 (4.735) 6,336 175 0.483 (0.575) 9.662 (11.509) 8.698 (10.367) 34.86 6,309 174 0.974 (0.527) 24.341 (13.182) 20.060 (10.915) 34.86 6,309 174 ρ = 0.97 (3) ρ = 0.98 (4) ρ = 0.99 (5) ρ=1 (6) 1.097 (0.223) 109.724 (22.342) 34.888 (7.393) 6,336 175 1.212 (0.239) . . 44.844 (9.346) 6,336 175 2.445 (0.567) 244.525 (56.709) 77.442 (18.849) 34.86 6,309 174 2.936 (0.635) . . 107.989 (24.908) 34.86 6,309 174 Panel A: Within estimates. Democracy Long-run effect of democracy Effect of democracy after 25 years Observations Countries in sample Democracy Long-run effect of democracy Effect of democracy after 25 years Exc. Instruments F-stat. Observations Countries in sample Notes: 0.867 0.982 (0.218) (0.216) 28.913 49.116 (7.255) (10.795) 20.574 26.927 (5.232) (6.071) 6,336 6,336 175 175 Panel B: 2SLS estimates. 1.464 (0.509) 48.806 (16.956) 34.683 (12.231) 34.86 6,309 174 1.955 (0.523) 97.735 (26.138) 53.448 (14.743) 34.86 6,309 174 This table presents estimates of the effect of democracy on GDP per capita, imposing the persistence level of the GDP process at the top of each column. The coefficient on democracy is multiplied by 100. Panel A presents within estimates controlling for four lags of GDP per capita. Panel B presents 2SLS estimates instrumenting democracy with four lags of regional democracy waves and the F statistic for the excluded instruments. In all specifications we control for a full set of country and year fixed effects and four lags of GDP per capita. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A21 Table A5: Summary of the Monte Carlo simulations for panels with different levels of persistence. Within estimator ρ = 0.98 (3) GMM estimator Assumed persistence of GDP: ρ = 0.963 (1) ρ = 0.97 (2) ρ = 0.99 (4) ρ = 0.963 (5) Average persistence of GDP (standard deviation) Relative bias (Nickell bias): 0.9558 (0.0016) -0.75% 0.9637 (0.0015) -0.65% 0.9750 (0.0012) -0.51% 0.9864 (0.0010) -0.36% 0.9586 (0.0019) -0.46% Average coefficient of democracy (standard deviation) 0.7772 (0.2228) 0.7772 (0.2228) 0.7771 (0.2228) 0.7769 (0.2230) 0.7889 (0.2355) Average long-run effect of democracy (standard deviation) 17.671 (5.266) Average persistence of GDP (standard deviation) Relative bias (Nickell bias): ρ = 0.97 (6) ρ = 0.98 (7) ρ = 0.99 (8) 0.9662 (0.0017) -0.39% 0.9771 (0.0014) -0.29% 0.9881 (0.0010) -0.19% 0.7888 (0.2355) 0.7886 (0.2355) 0.7884 (0.2355) 21.536 (6.460) 31.264 57.731 19.141 23.489 (9.537) (18.452) (5.908) (7.301) Panel B: Assuming stationary initial conditions. 34.714 (10.991) 67.082 (22.476) 0.9171 (0.0049) -4.7% 0.9238 (0.0047) -4.8% 0.9333 (0.0045) -4.8% 0.9485 (0.0042) -4.2% 0.9129 (0.0096) -5.2% 0.9175 (0.0096) -5.4% 0.9247 (0.0095) -5.6% 0.9539 (0.0075) -3.6% Average coefficient of democracy (standard deviation) 0.7688 (0.2144) 0.7672 (0.2141) 0.7642 (0.2136) 0.7628 (0.2133) 0.7499 (0.2386) 0.7455 (0.2383) 0.7394 (0.2376) 0.7558 (0.2376) Average long-run effect of democracy (standard deviation) 9.310 (2.686) 10.114 (2.929) 11.524 (3.361) 14.921 (4.404) 8.744 (3.032) 9.189 (3.226) 10.013 (3.587) 16.971 (6.438) Panel A: Assuming non-stationary initial conditions. Notes: This table presents the average estimates obtained from 1,000 Monte Carlo simulations of samples for GDP and democracy that satisfy the same empirical properties as in our dataset. The persistence of GDP is set to the level indicated in the top row. The top panel presents results in which we assume that the initial level of GDP is independent of its stationary level. The bottom panel presents results in which we assume that the initial level of GDP is given by its stationary level. Columns 1 to 4 present results for the within estimator, and columns 5 to 8 present results for the GMM estimator. All the estimates and t−statistics are obtained using the within estimator and its standard asymptotic limit. A22 Table A6: Effect of democracy on (log) GDP per capita. Estimates obtained with alternative dichotomous measures of democracy. Measure of democracy: Ours (1) Freedom House (3) PS (2) Polity IV (4) CGV (5) BMR (6) Panel A: Within estimates controlling for GDP dynamics. Democracy Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process Observations Countries in sample Democracy Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process Observations Countries in sample Democracy Observations Countries in sample 0.787 (0.226) 21.240 (7.215) 16.895 (5.297) 0.963 (0.005) 6,336 175 1.149 (0.554) 31.521 (17.425) 24.866 (12.978) 0.964 (0.005) 6,309 174 -10.112 (4.316) 6,934 175 0.785 (0.287) 21.457 (8.515) 16.967 (6.440) 0.963 (0.005) 5,736 153 Panel B: 2SLS 0.652 0.152 0.323 (0.222) (0.251) (0.259) 13.332 4.406 8.835 (4.577) (7.463) (7.437) 11.938 3.462 6.996 (4.040) (5.774) (5.774) 0.951 0.966 0.963 (0.006) (0.005) (0.005) 5,587 5,630 5,994 174 153 175 estimates controlling for GDP dynamics. 1.040 4.179 1.139 1.440 (0.424) (1.594) (0.537) (0.760) 28.605 72.043 34.515 40.413 (13.791) (30.453) (19.336) (23.993) 22.538 67.680 26.553 31.581 (10.090) (28.112) (13.588) (17.719) 0.964 0.942 0.967 0.964 (0.005) (0.007) (0.005) (0.005) 5,736 5,185 5,577 5,962 153 174 151 174 Panel C: Within estimates in levels ignoring GDP dynamics. -8.387 (6.746) 6,328 153 5.414 (3.150) 5,840 174 -11.377 (4.091) 6,179 154 -7.116 (4.713) 6,588 175 0.530 (0.271) 14.654 (7.910) 11.700 (6.128) 0.964 (0.005) 5,783 174 1.088 (0.668) 30.403 (20.649) 24.145 (15.639) 0.964 (0.005) 5,775 174 -4.225 (4.482) 6,372 174 Notes: This table presents estimates of the effect of democracy on GDP per capita, using alternative measures of democracy listed in the top row. PS stands for Papaioannou and Siourounis (2008), CGV stands for Cheibub, Gandhi, and Vreeland (2010), and BMR stands for Boix, Miller, and Rosato (2012). The coefficient of democracy is multiplied by 100. Panel A presents within estimates controlling for four lags of GDP per capita. Panel B presents 2SLS estimates instrumenting democracy with four lags of regional democracy waves and the F statistic for the excluded instruments. Panel C presents within estimates that do not control for GDP dynamics. In all specifications we control for a full set of country and year fixed effects. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A23 Table A7: Effect of democracy on (log) GDP per capita. Estimates obtained with alternative measures of GDP. Within estimates Arellano & Bond estimates HHK estimates 2SLS estimates HHK with IV estimates (1) (2) (3) (4) (5) Panel A. GDP per capita from the World Bank in constant local currency Democracy Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process Observations/Countries Democracy Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process Observation/Countries Democracy 0.858 (0.237) 21.110 (7.106) 17.307 (5.248) 0.959 (0.007) 6,377/181 1.202 0.712 (0.487) (0.387) 17.326 9.265 (8.027) (5.475) 16.006 8.570 (7.148) (4.977) 0.931 0.923 (0.012) (0.008) 6,195/181 6,195/181 Panel B. GDP per capita from the Penn 1.430 (0.680) 35.699 (20.321) 29.044 (15.459) 0.960 (0.007) 6,350/180 World Tables 9.0 0.565 (0.610) 10.782 (12.415) 8.675 (9.788) 0.948 (0.010) 6,350/180 0.473 1.141 1.268 0.497 1.139 (0.236) (0.528) (0.496) (0.727) (0.695) 11.638 18.436 19.609 12.869 22.931 (5.782) (8.987) (8.598) (19.308) (15.188) 8.863 16.128 16.827 9.776 18.000 (4.371) (7.664) (7.024) (14.535) (11.388) 0.959 0.938 0.935 0.961 0.950 (0.006) (0.008) (0.011) (0.006) (0.010) 7,204/183 7,021/183 7,021/183 7,159/182 7,159/182 Panel C. GDP per capita from the Penn World Tables 7.0 (PPP adjusted) 0.646 (0.280) 15.483 (6.781) 11.399 (4.858) 0.958 (0.007) 1.068 0.705 0.674 0.471 (0.604) (0.406) (0.787) (0.645) Long-run effect of democracy 15.897 8.529 17.025 8.530 (10.303) (5.149) (20.747) (11.820) Effect of democracy after 25 years 13.795 7.777 12.436 6.902 (8.601) (4.615) (14.895) (9.519) Persistence of GDP process 0.933 0.917 0.960 0.945 (0.012) (0.011) (0.006) (0.010) AR2 test p-value 0.38 Observations/Countries 6,990/181 6,809/181 6,809/181 6,945/180 6,945/180 Notes: This table presents estimates of the effect of democracy on log GDP per capita. Each panel presents results using a different measure of GDP. The reported coefficient on democracy is multiplied by 100. Column 1 presents results using the within estimator. Column 2 presents results using Arellano and Bond’s GMM estimator. Column 3 presents results using the HHK estimator. Columns 4 presents 2SLS estimates instrumenting democracy with four lags of regional democracy waves and the p-value of a Hansen overidentification test. Column 5 presents results using the HHK estimator instrumenting democracy with four lags of regional democracy waves. In all specifications we control for a full set of country and year fixed effects. Standard errors robust against heteroskedasticity and serial correlation at the country level are reported in parentheses. A24 Table A8: Effects of democracy on (log) GDP per capita. The estimates control for the influence of outliers. Democracy log GDP first lag log GDP second lag log GDP third lag log GDP fourth lag Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process Observations (1) (2) (3) (4) (5) 0.787 (0.226) 1.238 (0.038) -0.207 (0.046) -0.026 (0.028) -0.043 (0.017) 21.240 (7.215) 16.895 (5.297) 0.963 (0.005) 6,336 0.558 (0.178) 1.225 (0.015) -0.197 (0.022) -0.028 (0.018) -0.029 (0.010) 19.423 (7.039) 13.055 (4.338) 0.971 (0.003) 6,046 0.596 (0.173) 1.234 (0.016) -0.212 (0.022) -0.020 (0.016) -0.029 (0.010) 21.983 (7.418) 14.276 (4.334) 0.973 (0.003) 6,027 0.397 (0.143) 1.229 (0.011) -0.205 (0.017) -0.034 (0.014) -0.013 (0.009) 18.086 (7.019) 9.999 (3.672) 0.978 (0.002) 6,160 0.490 (0.171) 1.240 (0.009) -0.209 (0.015) -0.031 (0.014) -0.026 (0.009) 19.003 (6.919) 12.074 (4.249) 0.974 (0.002) 6,336 Notes: This table presents within estimates of the effect of democracy on log GDP per capita. The coefficient on democracy is multiplied by 100. Column 1 presents our baseline within estimates. Column 2 removes observations with a a standardized residual estimated above 1.96 or below -1.96. In Column 3 we remove observations with Cook’s distance above the rule of thumb value of four over the number of observations. Following Li (1985), in Column 4 we compute a robust regression estimator that assigns outliers a lower weight. In Column 5 we present a Huber M estimator, which is less sensitive to the presence of outliers. In all specifications we control for a full set of country and year fixed effects and four lags of GDP per capita. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A25 Table A9: Effect of democracy on (log) GDP per capita. GMM estimates that exploit alternative sets of moment conditions. Within estimator Democracy log GDP first lag log GDP second lag log GDP third lag log GDP fourth lag Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process AR2 test p-value Moments Observations Countries in sample Arellano & Bond, different set of moments Adding Ahn & Schmidt moments from columns 2 to 4 (1) (2) (3) (4) (5) (6) (7) (8) 0.787 (0.226) 1.238 (0.038) -0.207 (0.046) -0.026 (0.028) -0.043 (0.017) 21.240 (7.215) 16.895 (5.297) 0.963 (0.005) 0.875 (0.374) 1.204 (0.041) -0.193 (0.045) -0.028 (0.028) -0.036 (0.020) 16.448 (8.436) 14.713 (7.128) 0.947 (0.009) [0.51] 2,509 6,161 175 0.994 (0.554) 1.204 (0.047) -0.193 (0.047) -0.027 (0.028) -0.039 (0.020) 17.930 (11.679) 16.307 (10.191) 0.945 (0.011) [0.45] 1,266 6,161 175 1.034 (0.700) 1.176 (0.048) -0.183 (0.046) -0.026 (0.027) -0.038 (0.022) 14.526 (10.810) 13.885 (10.184) 0.929 (0.013) [0.53] 941 6,161 175 1.268 (0.607) 1.238 (0.051) -0.207 (0.049) -0.027 (0.028) -0.039 (0.017) 37.564 (30.953) 28.391 (18.483) 0.966 (0.015) [0.32] 231 6,161 175 1.107 (0.336) 1.230 (0.039) -0.202 (0.046) -0.029 (0.028) -0.039 (0.019) 27.928 (10.787) 22.743 (7.917) 0.960 (0.006) [0.46] 2,555 6,161 175 1.257 (0.508) 1.241 (0.043) -0.204 (0.047) -0.029 (0.029) -0.045 (0.020) 33.321 (17.133) 26.965 (12.562) 0.962 (0.008) [0.38] 1,312 6,161 175 1.461 (0.661) 1.237 (0.043) -0.203 (0.047) -0.030 (0.028) -0.045 (0.021) 36.386 (20.106) 30.193 (15.440) 0.960 (0.008) [0.39] 987 6,161 175 6,336 175 Notes: This table presents different GMM estimates of the effect of democracy on log GDP per capita. The coefficient on democracy is multiplied by 100. Column 1 presents our baseline within estimates. Columns 2-4 remove the country fixed effects by taking first differences of the data and estimates the model by GMM. Column 2 uses Arellano and Bond’s moment conditions, while columns 3 and 4 use different subsets of moment conditions described in the appendix. In Column 5 we remove fixed effects using forward orthogonal differences, and estimate the model using fewer moment conditions. In Columns 6-8 we add Ahn and Schmidt (1995) non-linear moment conditions to the models in columns 2-4. The AR2 row reports the p-value for a test of serial correlation in the residuals of the GDP series. The number of moments used by each estimator is reported below it. In all specifications we control for a full set of country and year fixed effects and four lags of GDP per capita. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A26 Table A10: Effect of transitions in and out of democracy on (log) GDP per capita. Within estimator (1) Arellano & Bond GMM estimator (2) Democratizations (3) (4) HHK estimator (5) (6) 0.803 0.894 1.470 0.846 0.947 1.132 (0.235) (0.256) (0.543) (0.524) (0.544) (0.488) Reversals -0.705 -0.853 -1.313 -1.123 -0.465 -0.753 (0.335) (0.376) (0.957) (0.860) (0.796) (0.743) Long-run effect of democracy 21.770 22.199 27.377 15.141 18.955 24.317 (7.635) (8.186) (12.982) (11.165) (13.581) (13.790) Effect of democracy after 25 years 17.283 17.855 24.617 13.471 16.204 20.492 (5.560) (5.743) (10.786) (9.370) (10.640) (10.269) Long-run effect of reversal -19.116 -21.200 -24.450 -20.089 -9.301 -16.187 (9.302) (9.785) (17.763) (15.466) (16.526) (16.403) Effect of reversal after 25 years -15.177 -17.051 -21.985 -17.872 -7.951 -13.642 (7.256) (7.587) (16.098) (13.627) (13.937) (13.617) Persistence of GDP process 0.963 0.960 0.946 0.944 0.950 0.953 (0.005) (0.007) (0.011) (0.012) (0.013) (0.013) Observations 6,336 5,688 6,161 5,513 6,161 5,513 Countries in sample 175 175 175 175 175 175 Number of GDP lags: 4 8 4 8 4 8 Notes: This table presents estimates of the effect of democracy on GDP per capita, allowing democratizations and reversals to have different effects. The coefficient on democratizations and reversals is multiplied by 100. Columns 1 and 2 present within estimates. Columns 3 and 4 present Arellano and Bond GMM estimates. Columns 5 and 6 present HHK estimates. Even columns add up to eight lags of GDP as controls. In all specifications we control for a full set of country and year fixed effects, as well as four lags of GDP. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A27 Table A11: Marginal effects of GDP lags on the propensity to democratize. (1) (2) (3) (4) (5) Panel A: Probability of a democratization. Change in GDP at t − 1 -0.126 (0.042) Change in GDP at t − 2 -0.086 (0.047) -0.121 (0.045) Change in GDP at t − 3 GDP level effect Observations -0.002 (0.003) 2,832 Change in GDP at t − 1 -0.002 -0.002 (0.003) (0.003) 2,752 2,706 Panel B: Probability -0.094 (0.044) -0.133 (0.046) 0.074 (0.062) -0.017 (0.002) 2,836 -0.017 (0.002) 2,741 Change in GDP at t − 2 Change in GDP at t − 3 GDP level effect Observations -0.017 (0.002) 2,882 -0.076 (0.050) -0.128 (0.048) -0.011 (0.049) -0.002 (0.003) 2,616 of a reversal. -0.075 (0.051) -0.129 (0.048) -0.013 (0.049) -0.100 (0.050) 0.080 (0.069) -0.077 (0.054) -0.017 (0.002) 2,552 -0.106 (0.064) 0.079 (0.091) -0.138 (0.062) 2,616 2,552 Notes: This table presents the estimated marginal effects derived from a Probit model of the propensity to democratize (top panel) or revert to nondemocracy (bottom panel) based on past dynamics of GDP. In the top panel, the sample comprises the countries that were nondemocracies at time t − 1. In the bottom panel, the sample comprises the countries that were democracies at time t − 1. For each sample we estimate the probability of a transition based on past levels of GDP and year effects. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A28 Table A12: Spatial patterns of diffusion for democracy, unrest, and GDP per capita. (1) (2) (3) (4) (5) (6) Panel A: Spatial diffusion patterns for democracy. Lagged democracy Lagged regional democracy 0.812 (0.015) 0.143 (0.022) Lagged distance-weighted democracy Lagged unrest Lagged regional unrest 6,799 174 0.291 (0.021) 0.103 (0.051) Lagged distance-weighted unrest Lagged GDP 7,027 174 0.972 (0.006) 0.007 (0.006) 0.810 (0.015) 0.150 (0.024) -0.029 (0.058) 0.811 (0.015) 0.147 (0.023) 0.024 -0.003 (0.013) (0.013) 6,730 6,730 6,700 6,700 174 174 173 173 Panel B: Spatial diffusion patterns for unrest. 0.284 (0.021) 0.284 (0.021) 0.211 (0.130) Lagged neighbors’ average unrest Observations Countries in sample 0.835 (0.013) 0.130 (0.056) Lagged neighbors’ average democracy Observations Countries in sample 0.837 (0.013) 0.283 (0.021) 0.079 (0.054) 0.142 (0.140) 0.283 (0.021) 0.101 (0.053) 0.007 -0.002 (0.021) (0.021) 6,730 6,730 6,708 6,708 174 174 173 173 Panel C: Spatial diffusion patterns for GDP. 0.972 (0.006) 0.970 (0.007) 0.007 (0.006) 0.001 (0.017) 0.970 (0.007) 0.007 (0.006) 0.283 (0.021) 0.080 (0.055) 0.176 (0.152) -0.014 (0.023) 6,708 173 0.970 (0.007) Lagged regional GDP 0.007 (0.006) Lagged distance-weighted GDP 0.003 -0.002 (0.017) (0.017) Lagged neighbors’ average GDP -0.002 -0.002 -0.002 (0.001) (0.001) (0.001) Observations 6,941 6,730 6,730 6,703 6,703 6,703 Countries in sample 174 174 174 173 173 173 Notes: This table reports estimates of the association between innovations to democracy and lagged regional democracy (by initial regime), lagged average democracy weighted by inverse distance and lagged neighbors’ democracy. Panel B and C present analogous estimates for unrest and GDP. All models include a full set of country and year fixed effects. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A29 0.972 (0.006) 0.810 (0.015) 0.150 (0.024) -0.027 (0.058) -0.001 (0.013) 6,700 173 Table A13: Effects of democracy on (log) GDP per capita. 2SLS estimates that also control for the influence of outliers. Robust second stage Democracy log GDP first lag log GDP second lag log GDP third lag log GDP fourth lag Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process Observations Robust first and second stage (1) (2) (3) (4) (5) (6) (7) (8) 1.149 (0.554) 1.238 (0.038) -0.205 (0.046) -0.029 (0.028) -0.040 (0.018) 31.521 (17.425) 24.866 (12.978) 0.964 (0.005) 6,309 0.869 (0.446) 1.228 (0.015) -0.195 (0.021) -0.034 (0.017) -0.027 (0.010) 30.743 (16.896) 20.547 (10.776) 0.972 (0.003) 6,015 0.813 (0.454) 1.235 (0.016) -0.207 (0.022) -0.032 (0.016) -0.022 (0.010) 31.227 (19.210) 19.755 (11.386) 0.974 (0.003) 6,000 0.836 (0.395) 1.231 (0.011) -0.204 (0.017) -0.039 (0.013) -0.009 (0.008) 39.697 (20.397) 21.298 (10.235) 0.979 (0.002) 6,133 1.098 (0.500) 1.332 (0.020) -0.307 (0.033) -0.023 (0.024) -0.032 (0.015) 36.859 (19.517) 27.861 (13.571) 0.970 (0.004) 5,967 0.716 (0.388) 1.244 (0.017) -0.219 (0.024) -0.029 (0.018) -0.021 (0.011) 28.677 (18.020) 17.691 (10.045) 0.975 (0.003) 5,612 0.507 (0.268) 1.232 (0.008) -0.206 (0.013) -0.038 (0.012) -0.009 (0.008) 23.529 (14.029) 12.844 (7.002) 0.978 (0.003) 6,309 0.843 (0.385) 1.242 (0.009) -0.209 (0.015) -0.035 (0.015) -0.022 (0.009) 33.757 (16.508) 21.002 (9.808) 0.975 (0.002) 6,309 Notes: This table presents 2SLS estimates of the effect of democracy on GDP per capita that instrument democracy with four lags of regional democracy. The coefficient on democracy is multiplied by 100. Column 1 presents our baseline 2SLS estimates. In Column 2 we remove observations with a standardized residual above 1.96 or below -1.96 in the second stage. In Column 3 we remove points with estimated Cook’s distance above the rule of thumb value of four over the number of observations in the second stage. In Column 4 we compute robust regression weights for the second stage following Li (1985), and re-estimate the model by 2SLS using these weights. In Column 5 we estimate the first and second stage manually excluding at each step countries with a standardized residual estimated above 1.96 or below -1.96. In Column 6 we estimate the first and second stage manually, excluding at each step countries with Cooks’ distance above 4 over the number of observations. In Column 7 we estimate each stage using a robust estimator following Li (1985). In Column 8 we estimate each stage using a Huber M estimator. In all specifications we control for a full set of country and year fixed effects and four lags of GDP per capita. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. Standard errors for our two step procedures in columns 5 to 8 are obtained following the adjustments proposed by Stefanski and Boos (2002) and Murphy and Topel (1985). We report the estimated persistence of the GDP process and the p−value for this being less than 1. We also report the estimated long-run effect of democracy and the p−value for this being different from 0. A30 Table A14: Effect of democracy on (log) GDP per capita. 2SLS estimates using alternative definitions of the regional democratization waves. Instrument construction: Initial regime measured at: Baseline Alternative Base 1960-65 All years Various Base 1960-65 All years Various (1) (2) (3) (4) (5) (6) (7) (8) Democracy 1.149 1.598 1.672 1.996 0.849 0.988 1.041 0.939 (0.554) (0.674) (0.552) (0.909) (0.512) (0.606) (0.547) (0.539) Long-run effect of democracy 31.521 44.573 46.118 56.717 23.028 26.926 28.027 25.646 (17.425) (22.706) (19.516) (32.291) (15.878) (18.381) (17.293) (16.425) Effect of democracy after 25 years 24.866 34.853 36.229 43.962 18.275 21.313 22.297 20.299 (12.978) (16.384) (13.743) (22.659) (11.880) (13.850) (12.819) (12.378) Persistence of GDP process 0.964 0.964 0.964 0.965 0.963 0.963 0.963 0.963 (0.005) (0.005) (0.005) (0.006) (0.005) (0.005) (0.006) (0.005) Exc. instruments F-stat. 254.5 167.5 302.8 121.9 28.7 20.4 28.1 16.1 Observations 6,309 6,270 6,330 5,906 6,309 6,270 6,330 5,906 Countries in sample 174 173 175 164 174 173 175 164 Notes: This table presents 2SLS estimates of the effect of democracy in GDP per capita using alternative constructions of the regional democracy instrument. The coefficient on democracy is multiplied by 100. In all models we instrument democracy using four lags of the alternative instruments. In columns 1-4, we use the baseline construction of the instrument. In columns 5-8 we use the alternative instruments described in the appendix. In columns 1 and 5 we use the baseline definition of initial regimes. In columns 2 and 6 we define initial regimes based on whether they were democratic during 1960-1964. We consider countries that were not independent as nondemocratic. In columns 3 and 7 we define initial regimes based on whether they were democratic throughout the sample. In columns 4 and 8 we use a richer set of initial regimes described in the text to construct the instrument. In all specifications we control for a full set of country and year fixed effects and four lags of GDP per capita. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. We report the estimated persistence of the GDP process and the p−value for this being less than 1. We also report the estimated long-run effect of democracy and the p−value for this being different from 0. The F statistic for the excluded instruments is reported below each estimate. A31 Table A15: Heterogeneous effects of democracy on (log) GDP per capita (additional estimates). Interaction with: Measured at: Democracy Interaction Long-run effect of democracy Effect of democracy after 25 years Persistence of GDP process Observations Countries in sample Share with primary: Share with tertiary: 1960 (1) 1970 (2) 1980 (3) Lagged (4) 1960 (5) 1970 (6) 1980 (7) Lagged (8) 0.573 (0.271) 0.008 (0.007) 17.730 (9.493) 12.952 (6.460) 0.968 (0.005) 5,300 138 0.537 (0.279) 0.008 (0.007) 16.561 (9.667) 12.115 (6.628) 0.968 (0.005) 5,300 138 0.537 (0.268) 0.010 (0.007) 16.488 (9.302) 12.099 (6.370) 0.967 (0.005) 5,300 138 0.443 (0.257) 0.016 (0.008) 13.481 (8.693) 9.936 (6.041) 0.967 (0.006) 5,300 138 0.531 (0.252) 0.182 (0.099) 16.532 (8.592) 12.041 (5.914) 0.968 (0.005) 5,300 138 0.507 (0.253) 0.136 (0.070) 15.746 (8.558) 11.480 (5.925) 0.968 (0.005) 5,300 138 0.537 (0.260) 0.073 (0.046) 16.624 (8.882) 12.141 (6.109) 0.968 (0.005) 5,300 138 0.660 (0.269) 0.031 (0.042) 20.037 (9.081) 14.804 (6.307) 0.967 (0.006) 5,300 138 Notes: This table presents within estimates of the effect of democracy on log GDP per capita and its interaction with other country characteristics. The column labels specify the variable interacted with democracy in each model. The reported coefficients on democracy and the interaction are multiplied by 100. We report main effects and long-run effects evaluated at the 25th percentile of the interacted variable. In all specifications we control for a full set of country and year fixed effects and four lags of GDP per capita. Standard errors robust against heteroskedasticity and serial correlation at the country level are in parentheses. A32 Figure A1: Different measures of democracy averaged across regions and worldwide. Democracy 1960-2010 in Africa 0 .2 Democracy index average .3 .4 .5 .6 Democracy index average .2 .4 .6 .7 Democracy 1960-2010 in the World 1960 1970 1980 1990 2000 2010 1960 1970 1980 Year 1990 2000 2010 Year Our democracy measure CGV democracy measure Polity IV index PS democracy measure BMR democracy measure Freedom House index Our democracy measure CGV democracy measure Polity IV index Democracy 1960-2010 in Eastern Europe and Central Asia 0 0 Democracy index average .2 .4 Democracy index average .2 .4 .6 .8 .6 Democracy 1960-2010 in East Asia and the Pacific PS democracy measure BMR democracy measure Freedom House index 1960 1970 1980 1990 2000 2010 1960 1970 1980 Year 1990 2000 2010 Year Our democracy measure CGV democracy measure Polity IV index PS democracy measure BMR democracy measure Freedom House index Our democracy measure CGV democracy measure Polity IV index Democracy index average .4 .6 .8 .2 .85 Democracy index average .9 .95 1 Democracy 1960-2010 in Latin American and the Caribbean 1 Democracy 1960-2010 in Western Europe and Offshoots PS democracy measure BMR democracy measure Freedom House index 1960 1970 1980 1990 2000 2010 1960 1970 1980 Year 1990 2000 2010 Year Our democracy measure CGV democracy measure Polity IV index PS democracy measure BMR democracy measure Freedom House index Our democracy measure CGV democracy measure Polity IV index Democracy 1960-2010 in South Asia 0 .2 Democracy index average .1 .2 Democracy index average .4 .6 .8 1 .3 Democracy 1960-2010 in Middle East and North Africa PS democracy measure BMR democracy measure Freedom House index 1960 1970 1980 1990 2000 2010 1960 1970 1980 Year Our democracy measure CGV democracy measure Polity IV index 1990 2000 2010 Year PS democracy measure BMR democracy measure Freedom House index Our democracy measure CGV democracy measure Polity IV index PS democracy measure BMR democracy measure Freedom House index Notes: The figures present the evolution over time of the several democracy measures for each of the seven regions used in the paper, as well as for the whole world. A33 -.2 -.5 Free elections dummy 0 .5 Constraints on the executive dummy 0 .2 .4 .6 1 .8 Figure A2: Institutional changes that follow an episode of democratization. -5 0 5 10 Years around democratization 15 20 -15 -10 -5 0 5 10 Years around democratization 15 20 -15 -10 -5 0 5 10 Years around democratization 15 20 -15 -10 -5 0 5 10 Years around democratization 15 20 0 .2 Democracy persistence .4 .6 .8 1 -.2 -.2 0 -.1 Civil liberties 0 Inclussive politics dummy .2 .4 .1 .2 -10 .6 -15 -15 -10 -5 0 5 10 Years around democratization 15 20 Notes: The figures plots the behavior of different components of democracy around a democratization (relative to continuing nondemocracies). Time (in years) relative to the year of democratization runs on the horizontal axis. See the text for a detailed explanation of how we measure these components separately from Polity IV and Freedom House raw data. A34 0 0 .01 .02 Fraction .02 Fraction .04 .03 .06 .04 Figure A3: Results of our Monte Carlo simulations, non-stationary initial conditions. 4 Coefficient of democracy t-stat 6 8 0 10 20 30 Estimate for permanent democratization 40 0 .5 1 Coefficient of democracy 1.5 2 Fraction 0 0 .02 .02 Fraction .04 .04 .06 .06 2 .08 0 .94 .945 .95 .955 .96 Estimate for persistence of GDP .965 .97 Notes: These figures plot the histograms and smoothed densities that we obtained for several estimates in our Monte Carlo simulation. The dashed red line indicates the population paremeters, and the solid red line indicates the average estimate over 1,000 simulations. In this case, the simulations assume that the initial GDP in each country is independent of the level implied by its GDP process. A35 .06 .04 Fraction .02 0 0 .02 Fraction .04 .06 Figure A4: Results of our Monte Carlo simulations, stationary initial conditions. 4 Coefficient of democracy t-stat 6 8 0 .5 1 Coefficient of democracy 1.5 .06 Fraction .04 .02 0 0 .02 Fraction .04 .06 .08 2 .08 0 0 5 10 15 Estimate for permanent democratization 20 .9 .905 .91 .915 .92 .925 .93 .935 .94 .945 .95 .955 .96 .965 .97 Estimate for persistence of GDP Notes: These figures plot the histograms and smoothed densities that we obtained for several estimates in our Monte Carlo simulation. The dashed red line indicates the population paremeters, and the solid red line indicates the average estimate over 1,000 simulations. In this case, the simulations assume that the initial GDP in each country is given by the level implied by its GDP process. A36 Figure A5: Semi-parametric estimates of the over-time effects of a reversal to nondemocracy on the log of GDP. Regression adjustment. 20 -40 -40 Change in GDP per capita log points -30 -20 -10 0 10 Change in GDP per capita log points -30 -20 -10 0 10 20 Inverse propensity score reweighting. -15 -10 -5 0 5 10 15 Years around reversal 20 25 30 -15 -10 -5 0 5 10 15 Years around reversal 20 25 30 -40 Change in GDP per capita log points -30 -20 -10 0 10 20 Doubly-robust estimates. -15 -10 -5 0 5 10 15 Years around reversal 20 25 30 Notes: These figures plot semi-parametric estimates of the effect of a reversal to nondemocracy on GDP per capita in log points. The solid line plots the estimated average effect on GDP per capita (in log points) on countries that reverted, together with a 95% confidence interval in dashed lines. Time (in years) relative to the year of reversal runs on the horizontal axis. A37 −20 Change in GDP per capita log points −10 0 10 20 30 40 Figure A6: Semi-parametric estimates of the over-time effects of a democratization on the log of GDP. Doubly-robust estimates for the average treatment effect. −15 −10 −5 0 5 10 Years around democratization 15 20 Notes: This figure plots semi-parametric estimates of the effect of democratizations on GDP per capita in log points, using the doubly-robust estimator. The solid line plots the estimated average effect on GDP per capita (in log points), together with a 95% confidence interval in dashed lines. Time (in years) relative to the year of democratization runs on the horizontal axis. A38 0 Estimated density 10 20 30 Figure A7: Smoothed density for the estimated propensity to democratize or revert to nondemocracy. .1 .2 .3 .4 .5 Propensity to democratize Propensity to democratize among countries that democratized Propensity to democratize among countries that remained nondemocratic 0 10 Estimated density 20 30 40 50 0 0 .1 .2 .3 Propensity to revert Propensity to revert among countries that reverted Propensity to revert among countries that remained in democracy .4 Notes: These figures plots the smoothed density of the estimated propensities to democratize (top figure) and revert (bottom figure). The black line plots the density for democratizers and countries experiencing reversals, respectively, while the gray line plots the density for the control countries in each case, which experienced no regime change. We smooth the densities using a standard Epanechnikov kernel. A39