- Email: [email protected]
Income and democracy: Evidence from nonlinear estimations
Economics Letters 118 (2013) 489–492
Contents lists available at SciVerse ScienceDirect
Economics Letters journal homepage: www.elsevier.com/locate/ecolet
Income and democracy: Evidence from nonlinear estimations Jess Benhabib a , Alejandro Corvalan b,∗ , Mark M. Spiegel c a
New York University, 19 W. 4th Street, 6FL, New York, NY, 10012, United States
b
Universidad Diego Portales, Santiago de Chile, Chile
c
Federal Reserve Bank of San Francisco, Economic Research, 101 Market St., San Francisco, CA, 94105, United States
article
info
Article history: Received 5 October 2012 Received in revised form 10 December 2012 Accepted 12 December 2012 Available online 20 December 2012
abstract We test the relation between income and democracy during the postwar period. We employ panel estimation methods that explicitly allow for the fact that the primary measures of democracy are censored with substantial mass at the boundaries. We find that the statistically significant positive income–democracy relationship is robust to the inclusion of country fixed effects. © 2012 Elsevier B.V. All rights reserved.
JEL classification: D72 E21 Keywords: Income Democracy Modernization theory Censoring Dynamic panel
1. Introduction The income–democracy correlation is the cornerstone of the ‘‘modernization theory’’. The theory claims a causal relation: democratic regimes are created and consolidated in affluent societies (e.g. Lipset, 1959, Przeworski and Limongi, 1997 and Epstein et al., 2006). Higher incomes can reduce conflict and support redistributive policies (e.g. Benhabib and Rustichini, 1996, Benhabib and Przeworski, 2006). However, a well-known recent paper by Acemoglu et al. (AJRY) (2008) argues that the income–democracy correlation is spurious. Using a cross-country panel between 1960 and 2000, they demonstrate that the significance of income in predicting democracy is not robust to the inclusion of country fixed effects. They interpret these specific effects as historical institutional arrangements, which are causing both democracy and higher income in the long run. This note reexamines the effect of income on democracy, concentrating on the continuous measures often used in the literature.1 From Barro (1999) on, studies dealing with either Freedom House or Polity measures of democracy have been mostly using
∗
Corresponding author. Tel.: +56 2 6762814. E-mail addresses: [email protected] (J. Benhabib), [email protected] (A. Corvalan), [email protected] (M.M. Spiegel). 1 For binary measures of democracy, see Acemoglu et al. (2009) and Benhabib et al. (2011). 0165-1765/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2012.12.015
linear models in their estimations. AJRY (2008) employs a dynamic linear panel estimation using GMM methods to evaluate the determinants of changes in democracy measured by the same indices. We confront here the issue that measures of democracy are ordinal, censored, and with substantial mass at the boundaries, violating the maintained assumptions under OLS and general linear models. We respond to this problem by interpreting democracy as a latent variable, as in Treier and Jackman (2008), and by using either a two-sided Tobit specification or the double-censoring specification of Alan et al. (2008).2 Additionally, we employ the Wooldridge (2005) method, which generalizes the Chamberlain (1980) estimator and parameterizes the fixed effects as well as the initial conditions in a dynamic panel. We also introduce a continuous and one-side censored measure of democracy – the Vanhanen (2000) Index of Democratization –. For all the three measures, there exists a statistically significant contribution from income to democracy, even in the presence of country fixed effects.3 2 See Epstein et al. (2006) for the use of a Tobit estimator with Polity 2 data. Also Barro (1999) suggests that the use of nonlinear estimation would improve his approach. 3 There are also several recent works that recover the significance of income under linear specifications using alternative econometric techniques (see Bobba and Coviello, 2007, Treisman, 2011, Moral-Benito and Bartolucci, 2011, and Heid et al., 2012).
490
J. Benhabib et al. / Economics Letters 118 (2013) 489–492
2. Data and econometric methods 2.1. Data We consider two measures of income. First, we use the PWT (7.0) data set, which has coverage for 189 countries, and whose times series extend farther for most countries than previous versions.4 The Penn World Tables have not been exempt from criticism. These criticisms point to the substantial degree of variability across different versions of the Penn World Tables (e.g. Ponomareva and Katayama, 2010 and Johnson et al., forthcoming) and to misgivings about purchasing power parity adjustments (e.g. Bergeijk, 1998 and Johnson et al., forthcoming). We note however that our results using the PWT (7.0) data set are similar to the ones using the PWT 6.3 data set in Benhabib et al. (2011). We also get similar results using the Maddison data set of per Capita GDP for 202 countries, which has even greater coverage per country than the Penn World Tables, facilitating the estimation of withincountry variations. We study several continuous measures of democracy: First, we use the composite Polity2 (PI2) index, which measures the difference between the Polity IV Democracy and Autocracy indices. A notable feature of the PI2 index, which may be problematic for linear statistical inference, is the substantial share of observations at either extreme. Almost 1/5 of the observations are coded as full democracies. This leaves the distribution of this variable bimodal. Moreover, a majority of these full democracies are high income countries which remain full democracies throughout our sample. We also use the Freedom House Political Rights Index (FH), which allocates higher scores to countries that are closer to a set of characteristics that would be associated with a functioning democracy. In the spirit of maximizing the availability of within-country data, we follow Barro (1999) and AJRY (2008) in augmenting the index with the Bollen (2001) data for 1950–1965. The FH data is also bimodal, either with or without the inclusion of the Bollen data, with a substantial share of the countries in the sample designated as full democracies. Additionally, we also use the Vanhanen’s (2000) Index of Democratization, which is a composite measure of voter participation rates and the intensity of election contestation. Unlike the PI2 and FH measures, the Vanhanen Index is not censored on the right hand side, and so the data is better able to track and identify an existing income–democracy relationship. All indices are normalized to range between 0 and 1. We also control for additional covariates typically used in the literature, such as education and the log of population. Education is measured as average total years of schooling in the population age 25 and over, and population is measured as country inhabitants over 25 years old. Data is from Barro and Lee (2010). The sample includes all independent countries during the postwar, with observations taken every fifth year from 1960 to 2000.5 We choose the period to allow for comparability with previous works. As some countries received independence during this period, our base sample is unbalanced. 2.2. Specification Consider the following linear specification, dit = α dit −1 + β log Y it −1 +δt + θi + γ Xit −1 + εit
(1)
where dit and log Yit −1 are democracy and the log of GDP per capita for country i at period t respectively, δt and θi represent time and country fixed effects respectively, Xit −1 are additional covariates 4 AJRY 2008 use the PWT 6.1 data set, which includes 168 countries. 5 Five year periods are used to mitigate serial correlation problems; results are robust for averages. Our 1960 observation uses lagged data from 1955.
such as education and population, and εit is a disturbance term, clustered by country. Estimation of this specification under ordinary least squares is not suitable for our democracy measures which are censored and which have a substantial share of their samples falling on the boundaries. To account for these nonlinearities, we estimate using a two-sided Tobit specification. However, even in a Tobit specification our estimates may be inconsistent because of the incidental parameter problem: fixed effects cannot be omitted through differencing. The estimation of fixed effects is inconsistent for a given length of within-country time series, and in turn can bias the other parameter estimates. Alan et al. (2008) provide moment conditions for a two-sided censored panel in the presence of fixed effects. We therefore also report the results of this approach using their estimator, termed ‘‘Two-sides’’ below. Another source of concern is that our sample is a dynamic panel which includes lagged values of democracy. For the linear case, this problem can be treated with panel-GMM, but our nonlinear specification presents a more challenging problem, as taking the initial condition of the dependent variable as independent of unobserved heterogeneity may bias estimates in dynamic panels that are short in the time dimension. We therefore also follow Wooldridge (2005) in generalizing the Chamberlain (1980) approach by assuming that country fixed effects can be specified as a linear function of the mean sample value of the observable independent variables, the initial condition for the lagged dependent variables and country specific random effects. Akay (2009) indicates that this method can be used in unbalanced panels. 3. Results Table 1 summarizes our results for the three measures of democracy. With country fixed effects excluded, we obtain positive and statistically significant coefficient estimates for lagged income values using linear OLS. This is true for all three democracy measure samples. However, with fixed effects included, these results disappear for PI2 and FH. We thus verify the result in the literature that under OLS, the significance of lagged income on democracy levels is not robust to the inclusion of country fixed effects. However, the OLS results with the Vanhanen measure differ from the other censored democracy measures: the significance of lagged income is robust to the inclusion of country fixed effects. This is because the Vanhanen measure is censored only on the left, and is better able to capture the variability of democracy among countries with high measures of democracy. We next turn to nonlinear estimations. We find some discrepancies across democracy measures and data sets. For PI2, lagged income is statistically significant using both the PWT 7.0 and Maddison data sets, with point estimates relatively similar across different specifications. In the case of FH, income is not significant using PWT income data but is significant using the Maddison income data at a 5% confidence level for all estimations. Finally, our results using the Vanhanen index are robust, with confidence levels between 1% and 5% throughout. In addition, our point estimates indicate that substantial movements in income can have notable implications for measured democracy levels. While such spurts are exceptional, they do occur. In our balanced panel of 88 emerging market economies for which we have complete income data between 1960 and 2000, 18 countries, or 20.5% of the panel, experienced a growth spurt where income doubled within a 20 year period. To interpret the impact of such a growth spurt, consider that the mean coefficient for lagged income in all our specifications is 0.07. As lagged democracy is 0.5, the long run effect of income is 0.14. This effect implies that doubling income would result in an increase in measured democracy
J. Benhabib et al. / Economics Letters 118 (2013) 489–492
491
Table 1 Baseline results. OLS PWT 7.0 (1)
FE OLS Madd (2)
PWT 7.0 (3)
Tobit Madd (4)
Two sides
Wooldridge
PWT 7.0 (5)
Madd (6)
PWT 7.0 (7)
Madd (8)
PWT 7.0 (9)
Madd (10)
0.423*** (0.05) 0.052* (0.03)
0.445*** (0.05) 0.069** (0.03)
0.472*** (0.06) 0.064* (0.04)
0.506*** (0.06) 0.078* (0.04)
0.805*** (0.03) 0.078*** (0.03) −0.023 (0.03) 0.149*** (0.03)
0.573*** (0.04) 0.065** (0.03) 0.028 (0.03) 0.358*** (0.05)
998 158 28 196
1.067 156 38 190
998 158 28 196
1.067 156 38 190
998 158 28 196
1.067 156 38 190
0.501*** (0.06) 0.045 (0.04)
0.534*** (0.06) 0.106*** (0.04)
0.495*** (0.06) 0.024 (0.04)
0.514*** (0.06) 0.075** (0.04)
0.792*** (0.05) 0.049 (0.03) 0.033 (0.03) 0.262*** (0.06)
0.812*** (0.05) 0.076** (0.03) 0.03 (0.04) 0.172*** (0.05)
1.111 184 166 232
1.070 160 187 173
1.111 184 166 232
1.070 160 187 173
1.111 184 166 232
1.070 160 187 173
0.308*** (0.05) 0.050** (0.02)
0.331*** (0.06) 0.067** (0.03)
0.312*** (0.06) 0.066** (0.03)
0.328*** (0.06) 0.082** (0.04)
0.555*** (0.05) 0.053*** (0.02) −0.009 (0.02) 0.391*** (0.06)
0.530*** (0.05) 0.069*** (0.02) 0 (0.03) 0.375*** (0.06)
1.109 180 363 0
1.082 157 424 0
1.109 180 363 0
1.082 157 424 0
1.109 180 363 0
1.082 157 424 0
Dependent variable is polity 2 measure of democracy dem(t − 1) inc(t − 1)
0.785*** (0.03) 0.034*** (0.01)
0.795*** (0.03) 0.038*** (0.01)
0.435*** (0.05) 0.009 (0.03)
0.459*** (0.05) 0.017 (0.03)
0.76 998 158
0.763 1.067 156
0.82 998 158
0.822 1.067 156
inc(mean) dem(t = 0) R-squared Observations Countries Left censored Right censored
Dependent variable is freedom house measure of democracy dem(t − 1) inc(t − 1)
0.776*** (0.03) 0.043*** (0.01)
0.767*** (0.03) 0.055*** (0.01)
0.403*** (0.05) 0.01 (0.03)
0.428*** (0.05) 0.046 (0.03)
0.736 1.111 184
0.73 1.070 160
0.809 1.111 184
0.797 1.070 160
inc(mean) dem(t = 0) R-squared Observations Countries Left censored Right censored
Dependent variable is Vanhanen measure of democracy dem(t − 1) inc(t − 1)
0.815*** (0.03) 0.031*** (0.01)
0.804*** (0.03) 0.039*** (0.01)
0.330*** (0.05) 0.052*** (0.02)
0.356*** (0.05) 0.059*** (0.02)
0.809 1.109 180
0.804 1.082 157
0.87 1.109 180
0.861 1.082 157
inc(mean) dem(t = 0) R-squared Observations Countries Left censored Right censored
Note: Estimation results using PWT 7.0 and Maddison income data as indicated, with PI2, FH and Vanhanen democracy measures. Estimation by ordinary least squares, with country fixed effects excluded and included, in columns (1)–(4). Tobit, Two Sides and Wooldridge regressions in columns (5)–(10), coefficients beta are reported. Standard errors clustered by countries in parentheses. Country and time fixed effect estimates suppressed. * p < 0.1. ** p < 0.05. *** p < 0.001.
of about 0.10.6 Using the most recent 2008 Polity data, one could go from the democracy level found in Malaysia to the one found in Belgium. Finally, we test the robustness of our estimation to the inclusion of additional covariates, including education and the log of population. These estimations put a high threshold to the significance of income, given its close correlation with education (0.75 for PWT 7.0). The results displayed in Table 2, however, confirm the positive effect of income on democracy. For PI2 and the Vanhanen index, lagged income is statistically significant using both the PWT 7.0 and the Maddison data sets, although statistical significance is weaker. Only for FH is the effect not preserved, except in one case. All in all, the positive relationship between income and democracy is upheld. 6 The effect calculated is beta times log(2) = 0.7.
4. Conclusion While we observe some sensitivity by data source, the bulk of our results imply that when an appropriate nonlinear estimator is used to account for the censoring of democracy measures, we can reestablish a statistically significant positive relationship between income and democracy in the presence of country fixed effects. Moreover, our average coefficient point estimates indicate that prolonged episodes of economic development, which are not uncommon among emerging market economies in our sample, can result in substantive predicted changes in democracy levels. Acknowledgments Israel Malkin provided excellent research assistance. We are grateful for comments from an anonymous referee, Stephane Bonhomme, William Easterly, Chris Flynn, Bryan Graham, Bo Honoré and Adam Przeworski. The views expressed are not necessarily
492
J. Benhabib et al. / Economics Letters 118 (2013) 489–492
Table 2 Robustness checks. Polity2 PWT 7.0 (1)
Freedom house
Vanhanen
Madd (2)
PWT 7.0 (3)
Madd (4)
PWT 7.0 (5)
Madd (6)
0.444*** (0.053) 0.066** (0.033) 0.001 (0.018) −0.653*** (0.204)
0.493*** (0.051) 0.041 (0.033) 0.008 (0.025) −1.353*** (0.240)
0.501*** (0.052) 0.065* (0.035) 0.027 (0.023) −1.150*** (0.190)
0.339*** (0.057) 0.048* (0.025) 0.004 (0.014) 0.027 (0.140)
0.346*** (0.057) 0.054* (0.029) 0.017 (0.015) −0.034 (0.146)
0.801*** (0.030) 0.051* (0.031) 0.013 (0.015) 0.036 (0.127) −0.032 (0.033) 0.014 (0.015) −0.015 (0.128) 0.120*** (0.029) 941 131 31 190
0.865*** (0.039) 0.043 (0.037) 0.042** (0.019) −0.326* (0.168) 0.016 (0.039) −0.024 (0.019) 0.355** (0.170) 0.153*** (0.039) 939 142 130 201
0.769*** (0.054) 0.052 (0.036) 0.039** (0.017) −0.616*** (0.159) 0.020 (0.040) −0.017 (0.018) 0.651*** (0.162) 0.178*** (0.055) 935 131 144 169
0.607*** (0.052) 0.048** (0.023) 0.012 (0.011) 0.297*** (0.101) −0.030 (0.025) 0.006 (0.012) −0.274*** (0.103) 0.271*** (0.056) 944 141 287 0
0.546*** (0.048) 0.060** (0.024) 0.017 (0.012) 0.101 (0.098) −0.031 (0.028) 0.003 (0.012) −0.066 (0.101) 0.290*** (0.057) 945 130 339 0
Estimation: Tobit dem(t − 1) inc(t − 1) educ(t − 1) pop(t − 1)
0.424*** (0.054) 0.057* (0.031) −0.017 (0.019) −0.693*** (0.260) Estimation: Wooldridge
dem(t − 1) income(t − 1) educ(t − 1) pop(t − 1) inc(mean) educ(mean) pop(mean) dem(t = 0) Observations Countries Left censored Right censored
0.774*** (0.031) 0.056* (0.032) 0.012 (0.016) 0.264* (0.144) −0.031 (0.033) 0.013 (0.016) −0.245* (0.146) 0.119*** (0.031) 893 134 18 196
Note: Estimation results using PWT 7.0 and Maddison income data as indicated, with PI2, FH and Vanhanen democracy measures in different columns. Estimation by unconditional Tobit with country dummies in the first part of the table; Wooldridge regressions are in the second part. Coefficients beta are reported. Standard errors clustered by countries in parentheses. Country and time fixed effect estimates suppressed. * p < 0.1. ** p < 0.05. *** p < 0.001.
those of the Federal Reserve Board of Governors or the Federal Reserve Bank of San Francisco. Corvalan thanks FONDECYT for support through grant 11121307. References Acemoglu, Daron, Johnson, Simon, Robinson, James A., Yared, Pierre, 2008. Income and democracy. American Economic Review 98 (3), 808–842. Acemoglu, Daron, Johnson, Simon, Robinson, James A., Yared, Pierre, 2009. Reevaluating the modernization hypothesis. Journal of Monetary Economics 56, 1043–1058. Akay, Alpaslan, 2009. The Wooldridge method for initial values problem is simple: what about performance? IZA Discussion Paper No. 3943. January. Alan, Sule, Honoré, Bo E., Leth-Petersen, Sören, 2008. Estimation of panel data models with two-sided censoring. Mimeo. November 9. Barro, Robert, 1999. Determinants of democracy. Journal of Political Economy 107 (6), 158–183. Barro, Robert, Lee, Jong-Wha, 2010. A new data set of educational attainment in the world, 1950–2010. NBER Working Paper No. 15902. April. Benhabib, J., Corvalan, A., Spiegel, M.M., 2011. Reestablishing the income– democracy nexus, NBER Working Paper No. 16832. February. Benhabib, J., Przeworski, A., 2006. The political economy of redistribution under democracy. Economic Theory 29, 271–290. Benhabib, J., Rustichini, R., 1996. Social conflict and growth. Journal of Economic Growth 1, 125–142. Bergeijk, P.V., 1998. Did real world per capita income really grow faster in 1870–1913 than in 1973–1992? De Economist 146 (1), 143–159. Bobba, M., Coviello, D., 2007. Weak instruments and weak identification in estimating the effects of education on democracy. Economics Letters 96, 301–306.
Bollen, K., 2001. Cross-national indicators of liberal democracy, 1950–1990. In: 2ndICPSR Version. Chapel Hill, University of North Carolina, NC, 1998. Chamberlain, G., 1980. Analysis of covariance with qualitative data. Review of Economic Studies 47, 225–238. Epstein, D., Bates, R., Goldstone, J., Kristensen, I., O’Halloran, S., 2006. Democratic transitions. American Journal of Political Science 50 (3), 551–569. Heid, B., Langer, J., Larch, M., 2012. Income and democracy: evidence from system GMM estimates. Economics Letters 116, 166–169. Johnson, Simon, Larson, William, Papageorgiou, Chris, Subramanian, Arvind, 2012. Is newer better? Penn World Table revisions and their impact on growth estimates. Journal of Monetary Economics (forthcoming). Lipset, Seymour M., 1959. Some social requisites of democracy: economic development and political legitimacy. American Political Science Review 53 (1), 69–105. Moral-Benito, E., Bartolucci, C., 2011. Income and democracy: revisiting the evidence. Economics Letters 117 (3), 844–847. Ponomareva, N., Katayama, H., 2010. Does the version of the Penn World Tables matter? an analysis of the relationship between growth and volatility. Canadian Journal of Economics 43 (1), 152–179. Przeworski, Adam, Limongi, Fernando, 1997. Modernization: theories and facts. World Politics 49, 155–183. Treier, Shawn, Jackman, Simon, 2008. Democracy as a latent variable. American Journal of Political Science 52 (1), 201–217. Treisman, D., 2011. Income, democracy, and the cunning of reason. NBER Working Paper 17132. Vanhanen, Tatu, 2000. A new dataset for measuring democracy, 1810–1998. Journal of Peace Research 37 (2), 251–265. Wooldridge, Jeffrey M., 2005. Simple solutions to the initial conditions problem in dynamic, nonlinear panel data models with unobserved heterogeneity. Journal of Applied Econometrics 20, 39–54.
Contents lists available at SciVerse ScienceDirect
Economics Letters journal homepage: www.elsevier.com/locate/ecolet
Income and democracy: Evidence from nonlinear estimations Jess Benhabib a , Alejandro Corvalan b,∗ , Mark M. Spiegel c a
New York University, 19 W. 4th Street, 6FL, New York, NY, 10012, United States
b
Universidad Diego Portales, Santiago de Chile, Chile
c
Federal Reserve Bank of San Francisco, Economic Research, 101 Market St., San Francisco, CA, 94105, United States
article
info
Article history: Received 5 October 2012 Received in revised form 10 December 2012 Accepted 12 December 2012 Available online 20 December 2012
abstract We test the relation between income and democracy during the postwar period. We employ panel estimation methods that explicitly allow for the fact that the primary measures of democracy are censored with substantial mass at the boundaries. We find that the statistically significant positive income–democracy relationship is robust to the inclusion of country fixed effects. © 2012 Elsevier B.V. All rights reserved.
JEL classification: D72 E21 Keywords: Income Democracy Modernization theory Censoring Dynamic panel
1. Introduction The income–democracy correlation is the cornerstone of the ‘‘modernization theory’’. The theory claims a causal relation: democratic regimes are created and consolidated in affluent societies (e.g. Lipset, 1959, Przeworski and Limongi, 1997 and Epstein et al., 2006). Higher incomes can reduce conflict and support redistributive policies (e.g. Benhabib and Rustichini, 1996, Benhabib and Przeworski, 2006). However, a well-known recent paper by Acemoglu et al. (AJRY) (2008) argues that the income–democracy correlation is spurious. Using a cross-country panel between 1960 and 2000, they demonstrate that the significance of income in predicting democracy is not robust to the inclusion of country fixed effects. They interpret these specific effects as historical institutional arrangements, which are causing both democracy and higher income in the long run. This note reexamines the effect of income on democracy, concentrating on the continuous measures often used in the literature.1 From Barro (1999) on, studies dealing with either Freedom House or Polity measures of democracy have been mostly using
∗
Corresponding author. Tel.: +56 2 6762814. E-mail addresses: [email protected] (J. Benhabib), [email protected] (A. Corvalan), [email protected] (M.M. Spiegel). 1 For binary measures of democracy, see Acemoglu et al. (2009) and Benhabib et al. (2011). 0165-1765/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2012.12.015
linear models in their estimations. AJRY (2008) employs a dynamic linear panel estimation using GMM methods to evaluate the determinants of changes in democracy measured by the same indices. We confront here the issue that measures of democracy are ordinal, censored, and with substantial mass at the boundaries, violating the maintained assumptions under OLS and general linear models. We respond to this problem by interpreting democracy as a latent variable, as in Treier and Jackman (2008), and by using either a two-sided Tobit specification or the double-censoring specification of Alan et al. (2008).2 Additionally, we employ the Wooldridge (2005) method, which generalizes the Chamberlain (1980) estimator and parameterizes the fixed effects as well as the initial conditions in a dynamic panel. We also introduce a continuous and one-side censored measure of democracy – the Vanhanen (2000) Index of Democratization –. For all the three measures, there exists a statistically significant contribution from income to democracy, even in the presence of country fixed effects.3 2 See Epstein et al. (2006) for the use of a Tobit estimator with Polity 2 data. Also Barro (1999) suggests that the use of nonlinear estimation would improve his approach. 3 There are also several recent works that recover the significance of income under linear specifications using alternative econometric techniques (see Bobba and Coviello, 2007, Treisman, 2011, Moral-Benito and Bartolucci, 2011, and Heid et al., 2012).
490
J. Benhabib et al. / Economics Letters 118 (2013) 489–492
2. Data and econometric methods 2.1. Data We consider two measures of income. First, we use the PWT (7.0) data set, which has coverage for 189 countries, and whose times series extend farther for most countries than previous versions.4 The Penn World Tables have not been exempt from criticism. These criticisms point to the substantial degree of variability across different versions of the Penn World Tables (e.g. Ponomareva and Katayama, 2010 and Johnson et al., forthcoming) and to misgivings about purchasing power parity adjustments (e.g. Bergeijk, 1998 and Johnson et al., forthcoming). We note however that our results using the PWT (7.0) data set are similar to the ones using the PWT 6.3 data set in Benhabib et al. (2011). We also get similar results using the Maddison data set of per Capita GDP for 202 countries, which has even greater coverage per country than the Penn World Tables, facilitating the estimation of withincountry variations. We study several continuous measures of democracy: First, we use the composite Polity2 (PI2) index, which measures the difference between the Polity IV Democracy and Autocracy indices. A notable feature of the PI2 index, which may be problematic for linear statistical inference, is the substantial share of observations at either extreme. Almost 1/5 of the observations are coded as full democracies. This leaves the distribution of this variable bimodal. Moreover, a majority of these full democracies are high income countries which remain full democracies throughout our sample. We also use the Freedom House Political Rights Index (FH), which allocates higher scores to countries that are closer to a set of characteristics that would be associated with a functioning democracy. In the spirit of maximizing the availability of within-country data, we follow Barro (1999) and AJRY (2008) in augmenting the index with the Bollen (2001) data for 1950–1965. The FH data is also bimodal, either with or without the inclusion of the Bollen data, with a substantial share of the countries in the sample designated as full democracies. Additionally, we also use the Vanhanen’s (2000) Index of Democratization, which is a composite measure of voter participation rates and the intensity of election contestation. Unlike the PI2 and FH measures, the Vanhanen Index is not censored on the right hand side, and so the data is better able to track and identify an existing income–democracy relationship. All indices are normalized to range between 0 and 1. We also control for additional covariates typically used in the literature, such as education and the log of population. Education is measured as average total years of schooling in the population age 25 and over, and population is measured as country inhabitants over 25 years old. Data is from Barro and Lee (2010). The sample includes all independent countries during the postwar, with observations taken every fifth year from 1960 to 2000.5 We choose the period to allow for comparability with previous works. As some countries received independence during this period, our base sample is unbalanced. 2.2. Specification Consider the following linear specification, dit = α dit −1 + β log Y it −1 +δt + θi + γ Xit −1 + εit
(1)
where dit and log Yit −1 are democracy and the log of GDP per capita for country i at period t respectively, δt and θi represent time and country fixed effects respectively, Xit −1 are additional covariates 4 AJRY 2008 use the PWT 6.1 data set, which includes 168 countries. 5 Five year periods are used to mitigate serial correlation problems; results are robust for averages. Our 1960 observation uses lagged data from 1955.
such as education and population, and εit is a disturbance term, clustered by country. Estimation of this specification under ordinary least squares is not suitable for our democracy measures which are censored and which have a substantial share of their samples falling on the boundaries. To account for these nonlinearities, we estimate using a two-sided Tobit specification. However, even in a Tobit specification our estimates may be inconsistent because of the incidental parameter problem: fixed effects cannot be omitted through differencing. The estimation of fixed effects is inconsistent for a given length of within-country time series, and in turn can bias the other parameter estimates. Alan et al. (2008) provide moment conditions for a two-sided censored panel in the presence of fixed effects. We therefore also report the results of this approach using their estimator, termed ‘‘Two-sides’’ below. Another source of concern is that our sample is a dynamic panel which includes lagged values of democracy. For the linear case, this problem can be treated with panel-GMM, but our nonlinear specification presents a more challenging problem, as taking the initial condition of the dependent variable as independent of unobserved heterogeneity may bias estimates in dynamic panels that are short in the time dimension. We therefore also follow Wooldridge (2005) in generalizing the Chamberlain (1980) approach by assuming that country fixed effects can be specified as a linear function of the mean sample value of the observable independent variables, the initial condition for the lagged dependent variables and country specific random effects. Akay (2009) indicates that this method can be used in unbalanced panels. 3. Results Table 1 summarizes our results for the three measures of democracy. With country fixed effects excluded, we obtain positive and statistically significant coefficient estimates for lagged income values using linear OLS. This is true for all three democracy measure samples. However, with fixed effects included, these results disappear for PI2 and FH. We thus verify the result in the literature that under OLS, the significance of lagged income on democracy levels is not robust to the inclusion of country fixed effects. However, the OLS results with the Vanhanen measure differ from the other censored democracy measures: the significance of lagged income is robust to the inclusion of country fixed effects. This is because the Vanhanen measure is censored only on the left, and is better able to capture the variability of democracy among countries with high measures of democracy. We next turn to nonlinear estimations. We find some discrepancies across democracy measures and data sets. For PI2, lagged income is statistically significant using both the PWT 7.0 and Maddison data sets, with point estimates relatively similar across different specifications. In the case of FH, income is not significant using PWT income data but is significant using the Maddison income data at a 5% confidence level for all estimations. Finally, our results using the Vanhanen index are robust, with confidence levels between 1% and 5% throughout. In addition, our point estimates indicate that substantial movements in income can have notable implications for measured democracy levels. While such spurts are exceptional, they do occur. In our balanced panel of 88 emerging market economies for which we have complete income data between 1960 and 2000, 18 countries, or 20.5% of the panel, experienced a growth spurt where income doubled within a 20 year period. To interpret the impact of such a growth spurt, consider that the mean coefficient for lagged income in all our specifications is 0.07. As lagged democracy is 0.5, the long run effect of income is 0.14. This effect implies that doubling income would result in an increase in measured democracy
J. Benhabib et al. / Economics Letters 118 (2013) 489–492
491
Table 1 Baseline results. OLS PWT 7.0 (1)
FE OLS Madd (2)
PWT 7.0 (3)
Tobit Madd (4)
Two sides
Wooldridge
PWT 7.0 (5)
Madd (6)
PWT 7.0 (7)
Madd (8)
PWT 7.0 (9)
Madd (10)
0.423*** (0.05) 0.052* (0.03)
0.445*** (0.05) 0.069** (0.03)
0.472*** (0.06) 0.064* (0.04)
0.506*** (0.06) 0.078* (0.04)
0.805*** (0.03) 0.078*** (0.03) −0.023 (0.03) 0.149*** (0.03)
0.573*** (0.04) 0.065** (0.03) 0.028 (0.03) 0.358*** (0.05)
998 158 28 196
1.067 156 38 190
998 158 28 196
1.067 156 38 190
998 158 28 196
1.067 156 38 190
0.501*** (0.06) 0.045 (0.04)
0.534*** (0.06) 0.106*** (0.04)
0.495*** (0.06) 0.024 (0.04)
0.514*** (0.06) 0.075** (0.04)
0.792*** (0.05) 0.049 (0.03) 0.033 (0.03) 0.262*** (0.06)
0.812*** (0.05) 0.076** (0.03) 0.03 (0.04) 0.172*** (0.05)
1.111 184 166 232
1.070 160 187 173
1.111 184 166 232
1.070 160 187 173
1.111 184 166 232
1.070 160 187 173
0.308*** (0.05) 0.050** (0.02)
0.331*** (0.06) 0.067** (0.03)
0.312*** (0.06) 0.066** (0.03)
0.328*** (0.06) 0.082** (0.04)
0.555*** (0.05) 0.053*** (0.02) −0.009 (0.02) 0.391*** (0.06)
0.530*** (0.05) 0.069*** (0.02) 0 (0.03) 0.375*** (0.06)
1.109 180 363 0
1.082 157 424 0
1.109 180 363 0
1.082 157 424 0
1.109 180 363 0
1.082 157 424 0
Dependent variable is polity 2 measure of democracy dem(t − 1) inc(t − 1)
0.785*** (0.03) 0.034*** (0.01)
0.795*** (0.03) 0.038*** (0.01)
0.435*** (0.05) 0.009 (0.03)
0.459*** (0.05) 0.017 (0.03)
0.76 998 158
0.763 1.067 156
0.82 998 158
0.822 1.067 156
inc(mean) dem(t = 0) R-squared Observations Countries Left censored Right censored
Dependent variable is freedom house measure of democracy dem(t − 1) inc(t − 1)
0.776*** (0.03) 0.043*** (0.01)
0.767*** (0.03) 0.055*** (0.01)
0.403*** (0.05) 0.01 (0.03)
0.428*** (0.05) 0.046 (0.03)
0.736 1.111 184
0.73 1.070 160
0.809 1.111 184
0.797 1.070 160
inc(mean) dem(t = 0) R-squared Observations Countries Left censored Right censored
Dependent variable is Vanhanen measure of democracy dem(t − 1) inc(t − 1)
0.815*** (0.03) 0.031*** (0.01)
0.804*** (0.03) 0.039*** (0.01)
0.330*** (0.05) 0.052*** (0.02)
0.356*** (0.05) 0.059*** (0.02)
0.809 1.109 180
0.804 1.082 157
0.87 1.109 180
0.861 1.082 157
inc(mean) dem(t = 0) R-squared Observations Countries Left censored Right censored
Note: Estimation results using PWT 7.0 and Maddison income data as indicated, with PI2, FH and Vanhanen democracy measures. Estimation by ordinary least squares, with country fixed effects excluded and included, in columns (1)–(4). Tobit, Two Sides and Wooldridge regressions in columns (5)–(10), coefficients beta are reported. Standard errors clustered by countries in parentheses. Country and time fixed effect estimates suppressed. * p < 0.1. ** p < 0.05. *** p < 0.001.
of about 0.10.6 Using the most recent 2008 Polity data, one could go from the democracy level found in Malaysia to the one found in Belgium. Finally, we test the robustness of our estimation to the inclusion of additional covariates, including education and the log of population. These estimations put a high threshold to the significance of income, given its close correlation with education (0.75 for PWT 7.0). The results displayed in Table 2, however, confirm the positive effect of income on democracy. For PI2 and the Vanhanen index, lagged income is statistically significant using both the PWT 7.0 and the Maddison data sets, although statistical significance is weaker. Only for FH is the effect not preserved, except in one case. All in all, the positive relationship between income and democracy is upheld. 6 The effect calculated is beta times log(2) = 0.7.
4. Conclusion While we observe some sensitivity by data source, the bulk of our results imply that when an appropriate nonlinear estimator is used to account for the censoring of democracy measures, we can reestablish a statistically significant positive relationship between income and democracy in the presence of country fixed effects. Moreover, our average coefficient point estimates indicate that prolonged episodes of economic development, which are not uncommon among emerging market economies in our sample, can result in substantive predicted changes in democracy levels. Acknowledgments Israel Malkin provided excellent research assistance. We are grateful for comments from an anonymous referee, Stephane Bonhomme, William Easterly, Chris Flynn, Bryan Graham, Bo Honoré and Adam Przeworski. The views expressed are not necessarily
492
J. Benhabib et al. / Economics Letters 118 (2013) 489–492
Table 2 Robustness checks. Polity2 PWT 7.0 (1)
Freedom house
Vanhanen
Madd (2)
PWT 7.0 (3)
Madd (4)
PWT 7.0 (5)
Madd (6)
0.444*** (0.053) 0.066** (0.033) 0.001 (0.018) −0.653*** (0.204)
0.493*** (0.051) 0.041 (0.033) 0.008 (0.025) −1.353*** (0.240)
0.501*** (0.052) 0.065* (0.035) 0.027 (0.023) −1.150*** (0.190)
0.339*** (0.057) 0.048* (0.025) 0.004 (0.014) 0.027 (0.140)
0.346*** (0.057) 0.054* (0.029) 0.017 (0.015) −0.034 (0.146)
0.801*** (0.030) 0.051* (0.031) 0.013 (0.015) 0.036 (0.127) −0.032 (0.033) 0.014 (0.015) −0.015 (0.128) 0.120*** (0.029) 941 131 31 190
0.865*** (0.039) 0.043 (0.037) 0.042** (0.019) −0.326* (0.168) 0.016 (0.039) −0.024 (0.019) 0.355** (0.170) 0.153*** (0.039) 939 142 130 201
0.769*** (0.054) 0.052 (0.036) 0.039** (0.017) −0.616*** (0.159) 0.020 (0.040) −0.017 (0.018) 0.651*** (0.162) 0.178*** (0.055) 935 131 144 169
0.607*** (0.052) 0.048** (0.023) 0.012 (0.011) 0.297*** (0.101) −0.030 (0.025) 0.006 (0.012) −0.274*** (0.103) 0.271*** (0.056) 944 141 287 0
0.546*** (0.048) 0.060** (0.024) 0.017 (0.012) 0.101 (0.098) −0.031 (0.028) 0.003 (0.012) −0.066 (0.101) 0.290*** (0.057) 945 130 339 0
Estimation: Tobit dem(t − 1) inc(t − 1) educ(t − 1) pop(t − 1)
0.424*** (0.054) 0.057* (0.031) −0.017 (0.019) −0.693*** (0.260) Estimation: Wooldridge
dem(t − 1) income(t − 1) educ(t − 1) pop(t − 1) inc(mean) educ(mean) pop(mean) dem(t = 0) Observations Countries Left censored Right censored
0.774*** (0.031) 0.056* (0.032) 0.012 (0.016) 0.264* (0.144) −0.031 (0.033) 0.013 (0.016) −0.245* (0.146) 0.119*** (0.031) 893 134 18 196
Note: Estimation results using PWT 7.0 and Maddison income data as indicated, with PI2, FH and Vanhanen democracy measures in different columns. Estimation by unconditional Tobit with country dummies in the first part of the table; Wooldridge regressions are in the second part. Coefficients beta are reported. Standard errors clustered by countries in parentheses. Country and time fixed effect estimates suppressed. * p < 0.1. ** p < 0.05. *** p < 0.001.
those of the Federal Reserve Board of Governors or the Federal Reserve Bank of San Francisco. Corvalan thanks FONDECYT for support through grant 11121307. References Acemoglu, Daron, Johnson, Simon, Robinson, James A., Yared, Pierre, 2008. Income and democracy. American Economic Review 98 (3), 808–842. Acemoglu, Daron, Johnson, Simon, Robinson, James A., Yared, Pierre, 2009. Reevaluating the modernization hypothesis. Journal of Monetary Economics 56, 1043–1058. Akay, Alpaslan, 2009. The Wooldridge method for initial values problem is simple: what about performance? IZA Discussion Paper No. 3943. January. Alan, Sule, Honoré, Bo E., Leth-Petersen, Sören, 2008. Estimation of panel data models with two-sided censoring. Mimeo. November 9. Barro, Robert, 1999. Determinants of democracy. Journal of Political Economy 107 (6), 158–183. Barro, Robert, Lee, Jong-Wha, 2010. A new data set of educational attainment in the world, 1950–2010. NBER Working Paper No. 15902. April. Benhabib, J., Corvalan, A., Spiegel, M.M., 2011. Reestablishing the income– democracy nexus, NBER Working Paper No. 16832. February. Benhabib, J., Przeworski, A., 2006. The political economy of redistribution under democracy. Economic Theory 29, 271–290. Benhabib, J., Rustichini, R., 1996. Social conflict and growth. Journal of Economic Growth 1, 125–142. Bergeijk, P.V., 1998. Did real world per capita income really grow faster in 1870–1913 than in 1973–1992? De Economist 146 (1), 143–159. Bobba, M., Coviello, D., 2007. Weak instruments and weak identification in estimating the effects of education on democracy. Economics Letters 96, 301–306.
Bollen, K., 2001. Cross-national indicators of liberal democracy, 1950–1990. In: 2ndICPSR Version. Chapel Hill, University of North Carolina, NC, 1998. Chamberlain, G., 1980. Analysis of covariance with qualitative data. Review of Economic Studies 47, 225–238. Epstein, D., Bates, R., Goldstone, J., Kristensen, I., O’Halloran, S., 2006. Democratic transitions. American Journal of Political Science 50 (3), 551–569. Heid, B., Langer, J., Larch, M., 2012. Income and democracy: evidence from system GMM estimates. Economics Letters 116, 166–169. Johnson, Simon, Larson, William, Papageorgiou, Chris, Subramanian, Arvind, 2012. Is newer better? Penn World Table revisions and their impact on growth estimates. Journal of Monetary Economics (forthcoming). Lipset, Seymour M., 1959. Some social requisites of democracy: economic development and political legitimacy. American Political Science Review 53 (1), 69–105. Moral-Benito, E., Bartolucci, C., 2011. Income and democracy: revisiting the evidence. Economics Letters 117 (3), 844–847. Ponomareva, N., Katayama, H., 2010. Does the version of the Penn World Tables matter? an analysis of the relationship between growth and volatility. Canadian Journal of Economics 43 (1), 152–179. Przeworski, Adam, Limongi, Fernando, 1997. Modernization: theories and facts. World Politics 49, 155–183. Treier, Shawn, Jackman, Simon, 2008. Democracy as a latent variable. American Journal of Political Science 52 (1), 201–217. Treisman, D., 2011. Income, democracy, and the cunning of reason. NBER Working Paper 17132. Vanhanen, Tatu, 2000. A new dataset for measuring democracy, 1810–1998. Journal of Peace Research 37 (2), 251–265. Wooldridge, Jeffrey M., 2005. Simple solutions to the initial conditions problem in dynamic, nonlinear panel data models with unobserved heterogeneity. Journal of Applied Econometrics 20, 39–54.