GLOBALIZATION, CHINA AND DEMOCRACY IN DEVELOPING COUNTRIES

GLOBALIZATION, CHINA AND DEMOCRACY IN DEVELOPING COUNTRIES 20 October 2017 Tam Nguyen-Huu 1 Abstract Using a panel of around 100 developing countries during 1978-2014, this paper firstly shows that a closer trade relation with China has negative impacts on Democracy in developing countries but only since China started its go-out policy in 2000s. Secondly, impacts of Globalization on Democracy are not clear-cut during this period. Thirdly, while it is difficult to say with high confidence on positive and direct impacts of Globalization on Democracy, there is evidence that Globalization can work as a countervailing factor, albeit only to very small extent, to mitigate negative impacts of China on Democracy. Our findings are robust to different econometric methods (system GMM and IV estimations) with a strong focus on handling endogeneity issue. The comparative studies of Zimbabwe and Kenya consolidate the results of quantitative methods. Key words: China, Globalization, Democracy, democracy promotion, autocracy promotion, trade relation JEL Classifications: C26, F69, P33 (Rough draft. Please do not cite without permission.) 1 Hamburg University and Leuphana University of Lüneburg, Germany. Email: tam.nguyen_huu@leuphana.de i

1: Introduction It is widely accepted that the emergence and success of China over the last decade is among the most influential economic phenomenon. Being the biggest economy in the world, China is able to dominate the international economic order and becomes the most important external partner in many developing economies (Zhang 2010; Gilman 2015). While economic impacts of China have been thoroughly investigated, its political influences only attract attention recently. Economic success of China and its autocratic regime challenge supporters of Democracy in several ways. First, Chinese model questions the role of Democracy in improving standards of living, economic growth and reducing poverty (Gat 2007). Chinese model is often compared with Indian model regarding how Democracy and autocracy promote development. Secondly, as the biggest economy and biggest autocratic country in the world, with considerable capacity to affect the global order, China evolves as a potential threat for democratization in developing countries. Thirdly, China now aims at strengthening its soft power, materializing its dominance beyond economic realm while the traditional major powers in the West are experiencing economic difficulties and facing huge domestic challenges (Zhang 2010; Sun 2014; Yelery 2014). Using a large data sets and combining quantitative method with case studies, this article analyses impacts of China in relation with Globalization led by the West on democracy transformation in developing countries. This paper is structured in five parts. The second part reviews the theoretical and empirical literature. The third part presents the overall results from a large-n analysis of around 100 developing countries (as many developing countries as possible) over 1978-2014. The fourth part uses a comparative analysis, with the selection of the two most similar cases in Africa: Kenya and Zimbabwe. The two selected countries share a handful of similarities, but now are on very contradict political trends. We scrutinize these examples to answer the question, whether the different levels of Democracy in these countries could attribute to the Chinese involvement at corresponding extents. The conclusion part follows this. 2: Literature Review Democracy is determined by numerous factors. In an increasingly interdependent world, external determinants play an important role in reshaping political landscapes in the developing world through strong and comprehensive linkages among countries. International actors are often categorized into two opposite forces, enhancing or impeding democratization. Based on this division, theories on democracy promotion and autocracy promotion are developed to evaluate how negatively or positively the democratization of a country is affected by foreign forces. The focus in traditional literature is laid on the most active agents in international politics: powerful countries in the West and international organizations. Impacts of these actors are deemed as both positively and negatively, although positive sides seem more dominant. The main channels, by which most powerful actors in global politics exert their impacts on democracy in developing countries, are analysed in depth. Most 2

important channels are the diffusion of global ideas inspired by the democratic countries, the political and economic pressure of developed countries to require the political reforms in developing countries. The literature continues expanding rapidly to investigate whether democracy promotion works and if yes, in which conditions it can be facilitated or hindered. One factor that determines the effectiveness of democracy promotion is how strong the against-democracy forces are. To answer the question, a recent development in literature on external determinants of democracy is turned to the emerging countries in the developing world like China, Russia, Venezuela, Iran and Saudi Arabia. However, different from traditional actors, these newcomers, especially China, are mostly blamed for the backsliding of democratization in the developing world. This article investigates the two hypotheses, of which, the second hypothesis is examined only when there is evidence supporting the first hypothesis. Both the hypotheses are based on autocracy and democracy promotion theories. H1: Developing countries, which have closer relations with China, are less democratic over time. H2: Impacts of China on the democratization are smaller in the more globalized countries. The theory on autocracy promotion is developed based on the negative impacts of new powers on democratization in developing countries. Burnell (2010) theorizes the way that autocracy patrons negate democratization in its partners both unintentionally and purposely. The first channel is the diffusion of authoritarian values across borders and the borrowing of foreign models of authoritarian rule and their institutions (Burnell 2010: 6). In this way, the success of Chinese economic model could encourage the autocratic incumbent governments to follow and to reject the guidelines of the West, which seems not to work in developing countries for quite a long time. The second channel is to consolidate the authoritarian tendency through hard or soft pressure. The third is to assist diplomatically in international forums to eliminate the efforts to isolate autocratic governments from international donors. The fourth way is to put pressure on domestic public policies. Finally, doing business as usual can keep status-quo of the autocratic regime in recipient country, thereby, indirectly relieving the autocratic regime out of pressure to democratize from Democracy promoters. Fulfilling this set of criteria, an autocratic China s emergence as a leader of the world economy, even only through the story on its economic miracle, could be a real challenge for efforts to promote democracy internationally. Similarly to democracy promotion, the effectiveness of autocracy promotion is determined by how strong pro-democracy supporters are. A question at stake is why China attempts to promote autocracy overseas or how China can benefit from a world of less democratic countries. A prime example is often taken from Africa, the most vulnerable region. Firstly, from economic perspective, economic growth of China requires a lot of natural resources, which are important assets that developing countries export. At the same time, China needs a large market for its cheap goods, which are in need in poor economies and high population. Last but not least, more autocratic partners could be China s allies in international politics and China expects to reach goals on political 3

legitimacy and sufficient security and stability to continue its commercial activities (Bader (2011); Hanauer & Morrissun (2014: 1); Sun (2014)). Employing rational-choice model, Bader et al. (2010) shows that autocracy promotion can originate from the preference of autocracy promoter towards convergence of political system given the cost of instability of political transition. This is illustrated by the engagement and supports of China to elites and autocratic regimes in Cambodia and Myanmar, for both China s economic and political purposes, from both its domestic demand of elites and foreign policy in region. Thus, not only unintentionally, China has reasons to promote autocracy abroad purposely. Several articles have empirically investigated whether China negates the institutional quality, which is often closely associated to democratization, such as survival of autocratic regimes, political violence and governance quality, in developing countries. However, the findings are vague in general. Qualitative work by Tull (2006) provides a causal relationship between China s involvement in Africa and resulting political situation. This work comes to conclude that the belief that China will make a constructive contribution to support transitions to Democracy in Africa s fragile states appears far-fetched (Tull (2006:473)). Dividing African nations into three groups: transition countries, mineral-rich countries, and post-conflict states, this paper analyses impacts of the involvement of China on political developments in each group. However, impacts of China on each group show significant differences. In terms of methodology, the classification of countries seems to fail to reach a sufficient coverage for all African states. Moreover, this analysis overlooks many cases in each group. From different perspectives, Bader (2011) and Bader et al. (2010) reveal that the engagement of China in its partners in Asia, namely Mongolia, Cambodia and Myanmar, has supported the survival and stability of autocratic regimes. In exchange for China s supports, the target countries have given China the chance to access natural resources, to achieve geo-political interests and to continue the goal of isolating Taiwan in international arena. Bader (2011) consolidates these findings on impacts of China on autocracy survival by conducting quantitative analysis. While case studies provide a detailed description of causal mechanism in specific countries, quantitative works use a rather large sample to generalize this mechanism. Bader (2015a), deploying a sample of more than 100 countries since 1993 to 2008 with the widest range of variables indicating China s involvement so far, shows that export dependence on China could reinforce the survival for autocratic regimes but has no positive impacts on the stability of democratic partner countries. In contrast, Melnykovska et al. (2012) using panel data for a sample of 24 post-communist countries from 1986 to 2007 claims that China's economic involvement in Central Asia may have positive influences in terms of improving governance and undermining autocratic structures. Moreover, this study goes further by comparing such impacts with those of Russia, finding that impacts of China are much better than that of Russia. However, the article ignores the necessary robust checks for their institutional quality variables in econometric models. Another drawback is that the selection of Central Asia could reduce the generality of the findings. Examining China s impacts on human rights protection in its partners, Bader & Daxecker (2015) compares impacts of oil export dependence from China and the US on the resulting human rights record of the partner countries from 4

1992 to 2010. Surprisingly, the results show that states that selling more oil to the USA have worse human rights achievements than those selling to China. At the middle stand, some works present the necessary conditions that China s involvement might or might not affect its partners political dimensions. Bader (2011) shows that close economic tie with China enhances the survival of autocratic regimes while political linkages seem not to work. Bader (2015b) using a sample of 69 non-democratic countries showed that the economic engagement of China damageable or not depends on the regime type of partner countries. Specifically, China s economic cooperation is accompanied with autocracy durability in party-based regimes while it enhances the regime collapse for non-party regimes. Besides, Bader (2015a) shows that the resulting institutions in recipient countries are determined by the type of linkages with China. For example, diplomatic or army relations, appear to have no significant impacts on its partner countries, although some impacts of economic relations have been affirmed. In similar vein, Teorell (2010) using a panel data from 1972 to 2006 concludes that closer trade relationship with China and Russia could exert negative impacts on Democracy level but these impacts are statistically insignificant, indicating that a more detailed model is necessary to reflect this complicated causality. Using both quantitative method for large N sample and case studies of Kenya and Zimbabwe, this paper aims to fill some gaps in literature on impacts of China on democracy in developing countries. Firstly, to the best of our knowledge, there is not yet any article that explore impacts of China directly on democracy. Secondly, the empirical works only focus on democracy or autocracy promotion separately, while both theories emphasize the countervailing effects of democracy and autocracy patrons. An interaction effect should be examined. The third issue of previous works is the ignorance of time-frame s role. Most articles (Teorell 2010; Bader (2011); Melnykovska et al. 2012; Bader & Daxecker 2015; Bader 2015a) attempt to use as long period as possible, however, ignore that China is a really important power only recently. In fact, China has begun its go-out policy since the end of 2002. Until 2001 Premier Zhu and later Vice Premier Wu Bangguo officially used the go-out word; at the official level, in November 2002, president Jiang Zemin in Report at 16th Party Congress states: Implementation of the strategy of "going out" is an important measure taken in the new stage of opening up. We should encourage and help relatively competitive enterprises with various forms of ownership to invest abroad (People s Daily 2001; Zemin 2002; Globalization Monitor 2009; Ernst & Young 2013; State Council 2016). Aside political statements, China s impacts on democracy at global level are only considerable when it really approaches the global market, which can be marked by China s membership of WTO in effect since 2002. Thus, considering the lagged effects, to evaluate impacts of China on democratization in developing countries, only period after 2003 can be taken into account. This means, to evaluate the impacts of China, it is more reasonable to begin at least since 2003, when the Chinese ambition and influences are materialized by implementing its specific policies and goals. The following part presents evidence from a quantitative analysis. 5

3: Quantitative Analysis Firstly, we show the description of data. Then we show the findings for full sample (years 1978-2014).This is followed by different estimation methods for go-out period (years 2002-2014). Finally, the findings are consolidated by external instrumental variable approach to handle more effectively reverse causality. 3a: Data and descriptive Statistics Following Barro (1999) & Acemoglu et al. (2005), dependent variable is Democracy level measured by Freedom House index, which ranges from 1 as most democratic to 7 as most autocratic. Despite of criticism in terms of conceptualization, coding procedure and aggregation (see Munck & Verkuilen 2002, Coppedge et al. 2011 for thorough overview), Freedom House index with its obvious advantages for econometric models is excellent option in point. For convenience of interpretation and further robustness check with other indexes of Democracy, we convert this 7-1 scale to 0-10 scale with 0 as autocratic and 10 as democratic 2. The independent variables of interest include past level of Democracy, Globalization and Trade relation with China. Independent variables of interest, Globalization and Trade relation with China, are lagged one year (following Teorell (2010); Acemonglu et al. (2008)). Moreover, the past level of Democracy is incorporated in dynamic models but removed in static ones. To proxy for Globalization, we use KOF Index of Globalization, which covers the 1970-2013 period for almost all countries in the world (208 countries in 2013) with 1-100 scale (higher rating means more globalized). This index, with three dimensions: economic, political and social, presents the multifaceted concept of Globalization better than economic indexes like trade openness or investment volume. It is especially suitable for the purpose of this paper, which focuses on political development as dependent variable. The KOF dataset has received wide recognition in literature with more than two thousand times of google scholar citations in its two versions. Potrafke (2015) discusses findings of more than one hundred published papers (until 2014) on impacts of Globalization proxied by KOF on numerous fields of social sciences. The most-studied aspects are: macroeconomic performance, distributional consequences, regulation and other institutional qualities. The number of papers using KOF index to measure Globalization continues increasing since then. To the best of my knowledge, until now, only Kollias & Paleologou (2016) uses KOF index to investigate impacts of Globalization on Democracy. Following Bader (2015a), we proxy China s involvement by the share of export of goods from a country to China out of total export of goods from that country to the world. Export data is chosen instead of total trade because export of developing countries reflects the dependence of the central government on specific partner more than total trade (including both import and export) does. This is suitable because our paper 2 Freedom House index data in 1982 is overlapped with data in 1981 and 1983. Namely, figure in 1981 covers January 1981 to August 1982 and figure in 1983 covers August 1982 to December 1983. To avoid the loss and interruption of data, data in year 1982 is calculated by taking weighted average of data in year 1981 (2/3) and year 1983 (1/3). 6

limits target countries to developing countries whose government revenue highly depends on export, especially raw materials, which are in high demand for the economic expansion of Chinese economy. This data is available from 1978 to 2014 extracted from IMF Direction of Trade Statistics database (IMF 2016). We also consider other economic, social and political indicators that might proxy for the relation with China. The main candidates include foreign direct investment from China compiled by UNCTAD (2016), the proportion of aid and other financial supports from China out of GDP compiled by aiddata.org (Tierney 2011), and the proportion of turnover of Chinese companies out of GDP, compiled by National Bureau of Statistics of China, military relations, measured by arm trade (provided by Stockholm International Peace Research Institute), the number of state visits or high-ranking officers (data from Chinavita.com). However, these data sets do not cover a long period or a high number of countries like IMF trade data. They will be used selectively in the qualitative analysis when only two countries Kenya and Zimbabwe are studied. We add several control variables in my regression models. The first is per thousand expenditure-side real GDP at chained PPPs (in 2011US$) provided by Penn World Table, version 9.0 (Feenstra et al. 2015). The GDP variable is set in log form to have better normal distribution as usual practices in literature. This control variable is among the most popular ones suggested by previous studies and is with the highest availability over years (see Barro 1999; Teorell 2010 for examples). Two time-invariant control variables are British colony (dummy variable takes value 1 if country is British colony and 0 otherwise) and legal systems. There are intense evidence that colonies of British have higher opportunity to democratize because the Democracy ideas spread freely to British colonies while colonies of other power do not have such privileges (Barro 1999). Moreover, British colonies have better access to English, the language of USA- the biggest Western power. Colony variable is recoded from Authoritarian Regime Dataset constructed by Hadenius & Teorell (2007). This dataset covers 211 countries in the world that were colonized since 1700 by Western countries. We do not take countries that are colony of both British and French as British colony because such countries were affected by not only Britain and the magnitude of British impacts is unclear. The second control time-invariant variable is the legal origin of company law or commercial law compiled by Porta et al. 1999. There are three groups of legal origins in our sample: English, French and Communist/ Socialist systems 3. This legal classification shows not only legal heritage, but also the political heritage that might reshape the current political situation. Both colony and legal variables are widely controlled in literature on Democracy (see, among others, Barro (1999); Teorell (2010)). The tables 1 and present summary statistics of main variables and correlation among them. Our sample drops all too small and too big countries (with population of less than one or more than one hundred millions, respectively). Democracy in second sample is just a little higher than that of initial period (4.5 and 3.7 respectively). In contrast, Globalization, GDP and ExptoChina all show obvious upward tendency over 3 For countries that are not classified into one of three groups, we use the data on colonial history or the longest legal system that country follows. 7

time. The second sample (2002-2014) is more balanced mainly because of the collapse of Soviet and the merge of countries over time in 1990s. Table 1: Summary Statistics Variable Obs Mean SD Min Max Mean 1978-2002 Mean 2003-2014 DemocracyFH 3424 4.00 2.81 0 10 3.70 4.56 ExptoChina 2828 5.54 10.77 0 88.457 2.95 9.17 GlobalizationKOF 3230 40.66 12.55 10.56 79.188 36.35 49.38 GDP 3108 14.92 0.91 11.866 16.97 14.76 15.22 BritishColony 3424 0.27 0.44 0 1 EnglishLaw 3424 0.27 0.44 0 1 FranchLaw 3424 0.53 0.50 0 1 CommunistLaw 3424 0.20 0.40 0 1 DemocracyPOL 3177 5.03 3.30 0 10 4.46 6.10 DemocracyWB 1584 3.76 1.50 0.4314 7.334 3.76 GlobalizationDHL 738 35.57 12.63 6.13 68.147 35.57 The correlation table indicates the high correlation between British colony and legal variables, thus, our later analysis will not use these variables simultaneously in one model to avoid multicollinearity. Table 2: Correlation 1 2 3 4 5 6 7 8 9 10 11 1 DemocracyFH 1 2 ExptoChina -0.095* 3 GlobalizationKOF 0.47* 0.029 4 GDP 0.321* 0.023 0.688* 5 BritishColony 0.14* -0.032* '0.066* 0.004 6 EnglishLaw 0.187* -0.037* -0.003-0.1* 0.725* 7 FranchLaw -0.005-0.148* -0.012-0.02 '-0.448* -0.6435* 8 CommunistLaw -0.201* 0.215* 0.018 0.133* -0.244* -0.3047* -0.533* 9 DemocracyPOL 0.86* '-0.083* 0.498* 0.283* 0.026 0.1079* 0.012-0.13* 10 DemocracyWB 0.953* -0.165* 0.5* 0.276* 0.162* 0.1842* 0.008-0.198* 0.85* 11 GlobalizationDHL 0.223* -0.055 0.713* 0.584* 0.056 0.0539-0.126* 0.09* 0.19* 0.277* 1 *** p<0.01, ** p<0.05, * p<0.1. 3b: Full Sample (1978-2014) Analysis POLS (Pooled Ordinary Least Squares) is biased and inconsistent because country fixed effect Country i is correlated with explanatory variables (ExptoChina i,t-1). In contrast, FE (Fixed Effects Estimation) allows Country i to be correlated with explanatory variables. Moreover, in this case, FE might be more appropriate than RE (Random Effects Estimation) because it is more reasonable to assume un-observed effects are correlated with the observed variables. However, FE causes the loss of time-invariant variables. The basic econometric model (LSDV, Least Squares Dummy Variables, equivalent to FE) is as follows: DemocracyFH i,t = α + ω ExptoChina i,t-1 + Controls i,t-1 + Countryi + Year t + µ i,t Of which DemocracyFH denotes Freedom house Democracy level, ExptoChina denotes the proportion of export to China out of total export (goods). Country i is a fixed state effect that covers all time constant characteristics of each country affecting Democracy. Year t is a fixed year effect that covers the shocks affecting Democracy level in all countries, and µ it describes an error term that covers other omitted factors. 8

Fixed effects and first-differences models are identical for T=2. For T >=3 as in this case, FE and FD indicate different outcomes. FE is more efficient than FD (First Difference Estimation) when u i,t is serially uncorrelated. If ui,t is a random walk, the FD is more efficient because ui,t is serially uncorrelated. It is better to check both FD and FE results (Wooldridge 2006). For FD estimator, the model is rewritten as: DemocracyFH i,t = α + ω ExptoChina i,t-1 + Controls i,t-1 + Country i+ Year t + µ i,t Usually FD model eliminates time-invariant variables through first-differencing. However, following Spilimbergo (2008), besides keeping FD model with only time-variant variables, I still add Country variable after differencing. This control is to make the model quite demanding given that they (specifications) are in difference and with country specific effects (Spilimbergo (2008: 17). With this strong control variable, the conclusion is much more robust. We report both outcomes, with and without Country fixed effects in FD. Table 3: Impacts of China on Democracy in Full Sample (1978-2014) (1) (2) (3) (4) (5) (6) (7) (8) (9) OLS OLS OLS FE FE FD FD FE FE Dependent Variable is DemocracyFH ExptoChina (t-1) -0.0385-0.0211-0.0111 0.00581 0.00654-0.000565-0.000677 (0.0299) (0.0274) (0.0296) (0.0163) (0.0172) (0.00204) (0.00231) D.ExptoChina (t-1) -0.00486-0.00584 (0.00328) (0.00354) GDP (t-1) 0.779*** 0.895*** -0.0175-0.0494 (0.229) (0.224) (0.303) (0.0480) D. GDP (t-1) -0.0589 (0.174) BritishColony 0.327 (0.512) EnglishLaw - Frenc Law -0.767* (0.449) CommunistLaw -2.072*** (0.541) DemocracyFH (t-1) 0.857*** 0.855*** (0.0155) (0.0160) Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Country Fixed Effects Yes Yes Constant 2.895*** -8.128** -9.181*** 4.324*** 4.794 0.0606-0.185 0.559*** 1.387* (0.589) (3.476) (3.399) (0.243) (4.736) (0.308) (0.359) (0.0878) (0.752) Observations 2,729 2,486 2,486 2,729 2,486 2,516 2,292 2,729 2,486 R-squared 0.057 0.122 0.186 0.131 0.155 0.065 0.069 0.793 0.796 Number of countries 99 90 90 99 90 99 90 99 90 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. POLS in full sample shows a statistically insignificant association of ExptoChina and DemocracyFH. Regardless of estimation methods in use (POLS, FE, FD and Dynamic-FE), we cannot find significant association between ExptoChina and DemocracyFH. This result is suitable because China is only a big power quite recently, thus, its impacts cannot be seen in the whole period since 1978. The findings give us supports to separate the second period (from 2003 to 2014) to examine impacts of China on Democracy in 9

the developing world in the following part. However, because the time-frame is considerably shortened (from 37 years, 1978-2014, to 12 years, 2003-2014), estimation methods should be re-considered. Dynamic model, which is characterized by using lagged dependent variable as one regressor, is used because Democracy is highly persistent over time and Democracy level comes back to some equilibrium level (Acemonglu et al. (2008), Spilimbergo (2009) and Kollias & Paleolongou (2016)). This is more important in our case, when time length of samples is rather short. The following dynamic models with different estimation methods deal with problems arising from this approach. 3c: Second period (2003-2014) Analysis The static model is changed to: DemocracyFH i,t = α + πdemocracyfh i,t-1+ ω ExptoChina i,t-1 + Controls i,t-1 + Country i + Year t + µ i,t The selection of estimation methods depends on specific panel structure and no estimator can perform well in all circumstances (Judson & Owen (1999). The departure point to select estimation method for dynamic panel data is Judson & Owen (1999). Judson & Owen (1999) uses Monte Carlo simulation for samples with T between 10 and 40 and N with maximum equal 100, thus their findings can be a source of reference to my situation. Widely cited in literature, their article provides important instructions in practicing dynamic panel model. It concludes that for T <=10, LSDVC (for balanced panel) and GMM (Generalized Method of Moments) one step (for unbalanced panel) are preferred while T=20, LSDVC (for balanced panel) or GMM one step or AH (for unbalanced panel) should be chosen. For samples with T=30 (unbalanced panel), LSDV works just as well or better than other estimators (Judson & Owen (1999: 12). This consolidates Nickell (1981) s conclusion that the bias of LSDV reduces to zero when T approaches infinity. Moreover, for a sufficiently large N and T, the differences in efficiency, bias and RMSEs of the different techniques become quite small and all the estimators (except OLS) work better with larger dimensions of T and N (Judson & Owen (1999: 12-13)). However, their instructions should be treated with careful consideration in specific circumstances. Firstly, the assumptions seem to favour LSDVC. The suggested model of Judson & Owen supposes explanatory variables are uncorrelated with disturbance (strictly exogenous). GMM method can accommodate not strictly exogenous variables (predetermined or endogenous variables). Thus, in this article, we only pay attention to GMM method. Secondly, there are some important updates recently on GMM that might produce better results. GMM: Difference or System Firstly, it should be noted that since Judson & Owen (1999) published their findings, GMM has developed very quickly, leading to significant improvements. GMM estimator in Judson & Owen (1999) is based on Arellano & Bond (1991) (difference GMM), while this estimator had been developed further by Arellano & Bover (1995) and Blundell & Bond (1998) (system GMM). 10

There are some advantages of system GMM in comparison with difference GMM. Firstly, difference GMM suffers seriously from weak instruments problem, especially when dependent variable (Democracy in my case) is very persistent over time. This is because lagged levels only present limited prediction on the subsequent changes. Secondly, GMM system can incorporate time-invariant variables. In Monte Carlo simulation, Blundell & Bond (1998) investigates the performance of Difference GMM, System GMM and System GMM exploiting the homoscedasticity restrictions for N=100, 200 and 500 and T=4 and T=11. For T=11, N=100 (approximate T=13, N=120 in second period of my case), π=0.9 (similar with persistency of Democracy in my case), results based on 1000 Monte Carlo replications show mean of Difference GMM far lower (0.6455) while System GMM and System GMM exploiting heterokedasticity perform well (0.9259 and 0.9302, respectively). Considering other criteria RMSE and SD, two System GMM models are similar (around 0.052 and 0.045, respectively) and also much better than Difference GMM (0.3 and 0.16 respectively). Finally, Blundell & Bond (1998:1) concludes dramatic improvement in performance of the proposed estimators (system GMM) compared to the usual first-differenced GMM estimator, and compared to non-linear GMM. Thus, in my specific case, system GMM is obviously recommended. However, it should be noted that system GMM works only under arguably special circumstances, which makes it where offer the most hope, it may offer the least help (Roodman 2009b: 144, 156). Namely, in the context that prefers System GMM over Difference GMM, the assumption on non-correlation between control variables (x), which include fixed-effects information, and error is difficult to be fulfilled. To ensure the validity of using System GMM, Roodman (2009b) requests users of System GMM to report Differencein-Hansen test for all the system GMM instruments for the level equation and also those based only on dependent variable, which are expected to have p-value higher than conventional level of 10%. GMM: one-step or two-step In addition, Judson & Owen (1999) reports that the one-step GMM estimator outperforms the two-step estimation. However, Windmeijer (2005) proposes corrected variance estimate for the two-step efficient GMM. His Monte Carlo simulation finds that the Windmeijer correction in two step model works better than one-step one (with lower bias and standard errors in two-step model). This innovation has been incorporated in xtabond2 STATA procedure developed by Roodman (2009a). GMM: Too many weak instruments Another problem in applying GMM is weak instrument proliferation, which is analysed in details by Roodman (2009b). Too many weak instruments can over fit instrumented variables, leads to imprecise estimates of the optimal weighting matrix and invalidate Hansen test of weak instruments. Roodman (2009b) suggests several techniques to reduce the instrument counts: using only limited numbers of lags, collapsing instruments and imposing new moment conditions. As usual practice, I combine lag restriction to 1 and 2 lags with the collapsing of instruments, which is the most restrictive solution to limit instrument counts. 11

The findings in the second period of table 8 show that in all specifications, the negative impacts of China on political development in developing countries are obvious, statistically significant and lie in a narrow range. Table 4: Impacts of China on Democracy since China s go-out policy in 2003 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. The POLS model tends to have upward bias while FE has downward one. DemocracyFH t-1 in POLS model is 0.979 (model 1, table 4) while in FE is 0.678 (model 2, table 4), thus the appropriate estimator will produce the coefficient of DemocracyFH t-1 within the interval of (0.678; 0.979). All models in table 8 report the coefficients satisfying this initial requirement. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) POLS FE FD GMM GMM GMM GMM GMM GMM GMM GMM Dependent Variable is DemocracyFH DemocracyFH (t-1) 0.979*** 0.678*** 0.880*** 0.896*** 0.896*** 0.895*** 0.896*** 0.910*** 0.908*** 0.901*** (0.00640) (0.0406) (0.0557) (0.0574) (0.0580) (0.0564) (0.0532) (0.0568) (0.0576) (0.0556) D. DemocracyFH (t-1) -0.102* (0.0516) ExptoChina (t-1) -0.00213-0.00188-0.00807* -0.00911* -0.00921** -0.00920** -0.0481** -0.0421** -0.0424** -0.0390* (0.00155) (0.00270) (0.00416) (0.00467) (0.00470) (0.00461) (0.0191) (0.0176) (0.0183) (0.0210) D.ExptoChina (t-1) -0.00632 (0.00412) GlobalizationKOF -0.0110 0.00191 0.00154 0.00301 (0.0143) (0.0115) (0.0116) (0.0115) GlobalizationKOF*ExptoChina 0.000870** 0.000772** 0.000777* 0.000697 (0.000394) (0.000379) (0.000399) (0.000477) GDP (t-1) -0.0316-0.0296 0.00956-0.0361-0.0309-0.0163 (0.102) (0.101) (0.106) (0.0951) (0.0969) (0.0987) BritishColony 0.0302-0.0128 (0.0917) (0.0876) FrenchLaw -0.0642-0.0459 (0.104) (0.106) CommunistLaw -0.133-0.0650 (0.147) (0.145) Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Country Fixed Effects Yes Sargan test 0.299 0.773 0.775 0.641 0.677 0.671 0.694 0.66 Hansen test 0.106 0.3 0.306 0.204 0.726 0.516 0.503 0.397 AR(2) 0.56 0.84 0.84 0.833 0.556 0.826 0.827 0.825 Number of Instruments 17 20 21 22 23 27 28 29 Difference-in-Hansen tests All system GMM instruments 0.163 0.209 0.215 0.241 0.784 0.355 0.35 0.354 Those based on lagged democracy only 0.807 0.776 0.786 0.657 0.892 0.738 0.708 0.506 Number of countries 99 99 99 99 90 90 90 97 90 90 90 The coefficients of ExptoChina are negative and statistically significant, lying between -0.009 in models 4 to 7 and -0.04 in models 8 to 11. The results are almost remained when other control variables are added, both time-variant variables (GDP), and time-invariant variables (British Colony and Legal variables). Moreover, all GMM models meet the requirement that there is no second autocorrelation (test of AR (2) with p-value far higher than conventional level of 10% in all specifications). One of the most important concerns on system GMM is too many weak instruments. All GMM models are designed to reduce the instruments counts. We use both instruments collapse and lag constrains to first and second lag. The instrument counts (between 17 and 29) are well lower than the number of countries (around 100). In addition to this, both Hansen test and Difference-in-Hansen tests (including all system and those based on lagged dependent variable) consolidate the validity of instrumental variables (p-value far higher than conventional levels, indicating the null hypothesis of exogeneity of instruments cannot be 12

rejected). Thus, the consistent findings of different econometric techniques shows that negative impacts of China on political development in the developing world since 2003 is a real and significant concern. Finally, models 12 to 14 show that the negative association of Relation with China is mitigated by Globalization. Interaction variables between ExptoChina and GlobalizationKOF in models 8 to 11 have positive and statistically significant signs, albeit very small (coefficients around 0.0008), while impacts of GlobalizationKOF is statistically insignificant. It can be interpreted that, during 2003-2014, as a countervailing factor, GlobalizationKOF can reduce the negative impacts of China on political development. In other words, two countries with the same impacts of China, the more globalized country will suffer less from negative political effects from China. However, these impacts are very small. This might be due to short time of study (only 2003-2014), high persistence of democracy and too tightly controlled model (in dynamic specification). 3d: External Instrumental Variable Approach The previous part shows how internal instrumental variables can fix endogeneity problems in dynamic model. The following part presents the results with external instrumental variables approach. Although the following part emphasizes on handling weak instruments issue, it is difficult to have a satisfactory solution. It should be noted that our main independent variables of interest (ExptoChina) are highly variant over time, but most of my potential instruments for them are almost unchanged over time. Thus, the goal of this part is just to present another approach to handle reverse causality, consolidating findings on directions of causality. For this reason, interpretation of results will pay less attention on the efficiency of estimation. In general, valid instrumental variables must meet two requirements: exogeneity and relevance. Exogeneity means there is no correlation between instruments and error term. This requires arguments proposing a potential instruments must prove that: there is no direct impact of instruments on dependent variables, there is no impacts of dependent variable on instruments and there are convincing reasons why the instruments affect endogenous variables. However, the exogeneity condition cannot be tested, we must base on economic reasoning, convincing arguments or experiences in literature (Cameron & Trivedi (2009: 181), of course, natural or random experiments do not work in the situation of this article). Fortunately, the second condition, relevance requirement, can be tested through first stage regression. Relevance requires valid instruments to be highly associated with the endogenous variables given the control of exogenous independent variables (Murray (2006); Angrist & Pischke (2015); Schmidheiny (2016)). Exogeneity requirement of instrumental variables, unfortunately, cannot be tested because exogeneity requires Cov(instruments, error)=0 while error is unobservable. We can only test partially in overidentified cases, when more conditions than actually needed for identification (Verbeek (2013: 151)). Sargan-Hansen test of overidentifying restrictions will be implemented for cases with more than one instrumental variables. The J statistic is consistent in the presence of heteroscedasticity and autocorrelation while Sargan test for homokedasticity. In addition to overidentification test, Murray (2006; 2010) suggest several methods to 13

detect invalid instruments that might work in my situation: use alternative instruments, use intuition and reduced forms. On relevance requirement, a wide range of suggestions to detect weak instruments is introduced in literature. The purpose of tests is to evaluate how large the explanatory power of instruments is. A natural departure point is to check the pairwise correlation coefficients between instrumental variables and their instrumented variables. A further way is to check the magnitude and significance of instrumental variables in the first stage regression. Particularly, Bound et al. (1995) suggests to check partial R2 and the F statistic of the identifying instruments in the first stage estimation. More precisely speaking, Cragg and Donald (1993) statistics for relevance test of instruments is equal to first stage F statistics if only one endogenous regressor is included. A widely-used rule of thumb is F statistic must be at least 10 that means we allow the maximum bias in IV estimators to be less than 10%. However, this rule is ad hoc and may not sufficiently conservative when there are many overidentifying restrictions (Cameron & Trivedi (2009:196)), thus, I check the F statistics with Stock-Yogo weak ID test critical values reported for specific situations. However, one disadvantage of Stock-Yogo critical value is that it assumes conditional homokedastic and serially uncorrelated errors. As usual practice for heteroskedastic and autocorrelated errors, robust first stage F statistics is presented to check with Stock-Yogo critical values, however, the F values might be high even if instruments are weak (Ola & Pflueger (2013); Pflueger & Wang (2015)). Thus, I use so-called Effective F statistics, developed by Ola & Pflueger (2013) and implemented in STATA by weakivtest procedure, which can work when errors are heteroscedastic and serially correlated for both TSLS and LIML with single endogenous variable. (Ola & Pflueger (2013); Pflueger & Wang (2015)) Going further, I follow Murray (2010) to check whether it is required to correct for weak instruments by examining AR test developed by Anderson and Rubin (1949), Lagrange multiplier test, also called the score test, developed by Kleibergen (2002) and Conditional Likelihood Ratio (CLR) developed by Moreira (2003) and further developed by Andrews et al. (2007). This is implemented by comparing the constructed region (confidence sets of corresponding tests) and hypothesized values. If the hypothesized values lie outside the confidence region, the hypothesis would be invalid at conventional level (5%) (Cameron & Trivedi (2010: 202)). Using simulations, Andrews et al. (2007) finds that CLR test outperforms the AR test, therefore this is the most appropriate approach for our article. Finally, the selection of estimation strategy can be a solution for weak instruments. When model has more instrumental variables than endogenous variables, the endogenous variables are over-identified. In overidentifying model, two least squares (2SLS) estimator is effective when the instruments are strongly associated with regressors. However, in general, it is difficult to satisfy strong instruments, thus the 2SLS is rarely efficient. There are three main alternatives for 2SLS that will be considered. The first one is limited information maximum likelihood estimator (LIML for short); the second one is Fuller and the third one is GMM. In total, we have 4 estimation procedures, of which, GMM can handle non i.i.d error but both GMM and 2SLS are not appropriate for weak instruments- the main concern in this paper. Compared to 2SLS, 14

LIML is less impacted by finite sample bias, despite it produces higher standard errors (Angrist & Pischke 2015). Also opposing 2SLS, however, Murray (2010) recommends using Fuller. Thus, our priority is both LIML and Fuller, of which the Fuller estimators are conducted using parameter alpha equal to 1 and 4 (the most popular options). In exactly identified model (one regressor, one instrument in my case), LIML, Fuller, GMM and 2SLS are similar. Besides, to avoid challenges due to many weak instruments, a good solution is to use a single instrument for a single causal mechanism (Angrist & Pischke (2015: 145)). Another advantage of using single instrument is that a parsimonious usage of instrumental variables results in smaller small-sample bias because higher number of instrumental variables is accompanied with higher small-sample bias (Hahn & Hausman (2002)). Thus, our priority is to use the single instruments. Even though, we are well aware that using all available instrumental variables simultaneously has its own advantage: it leads to the most efficient estimator (Cameron (2009: 185); Verbeeck (2013: 154)). Moreover, different combinations of instrumental variables consolidate results for worries due to sensitivity to the choice of instruments. Thus, after using single instrumental variables, results for models with different combinations of instruments will be also reported. In general, following practical instructions of Murray (2010) and Angrist & Pischke (2015), there are several criteria for a good instrument variables regressions: Coefficient and significance levels of independent variables of interest in the first stage and the significance of coefficients of instruments in reduced form, Partial R2, F test (checked with Stock-Yogo critical values); effective F test, CLR confidence set test; alternative estimators (LIML, Fuller, GMM, 2SLS) and alternative instruments and their combination in different specifications. On instrumental variables for Trade Relation (Export of goods) with China, to the best of our knowledge, there is no such variable in literature. However, instrumental variables for international trade in literature in general is often motivated from gravity model. Thus, I believe that instrumental variables for exporting to China can be found using in the same reasoning. First instrument of relation with China is the distance between capital city of China and its partner from Gleditsch & Ward (2001). A closer distance to China promotes bilateral relations. This is a widely used proxy for international trade relation in general in literature.. The second instrument is population density. High population density is often associated with large population or small land area or both. Considering the fact that China imports mostly natural resources from developing countries. A larger land area can have more chance to have more natural resources and thus, trigger export relation with China. A smaller population means natural resources of the country can be exported because domestic demand is small. Thus, countries with larger land or small population, or at best- large land and small population (or low population density) can have better possibility to increase export to China. 15

Table 5 shows that a closer relationship with China has negative impacts on Democracy in all models but in different magnitudes. In the models 1 and 2, I use instruments separately then in combination in the following models in dynamic context, so that the method using external instrumental variables can be compared with those using internal instrumental variables in previous parts. Table 5: External IV approach for investigating impacts of China on Democracy (1) (2) (3) (4) (5) (6) (7) (8) (9) IVa IVb IVc IVd IVe IV2a IV2b IV2c IV2d Dependent Variable is DemocracyFH ExptoChina (t-1) -0.0412* -0.188*** -0.233*** -0.184*** -0.00461-0.124*** -0.0977*** -0.0899*** -0.00404 (0.0219) (0.0216) (0.0305) (0.0238) (0.00415) (0.0145) (0.0132) (0.0121) (0.00259) DemocracyFH (t-1) 0.976*** 0.974*** (0.00702) (0.00766) GDP 0.409*** 0.547*** 0.00700 0.434*** 0.565*** 0.00545 (0.118) (0.106) (0.0184) (0.0929) (0.0919) (0.0200) FrenchLaw -1.006*** -0.0206 0.820*** -0.0120 (0.244) (0.0420) (0.217) (0.0465) CommunistLaw -1.655*** -0.0278 (0.289) (0.0506) EnglishLaw 1.751*** -0.00433 (0.257) (0.0562) Britishcol 0.460* 0.532*** (0.256) (0.206) Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Constant 4.969*** 5.099*** -0.819-1.983 0.162 5.526*** -1.190-3.915*** 0.0669 (0.309) (0.354) (1.825) (1.619) (0.276) (0.315) (1.442) (1.467) (0.313) F statistics 88.46 168.88 100.34 130.33 123.92 145.24 161.58 186.45 183.48 Sargan statistic exactly exactly exactly exactly exactly identified identified identified identified identified 0 0 0 0.6113 Observations 984 1,172 1,070 1,070 1,070 984 897 897 897 R-squared 0.023-0.530-0.986-0.525 0.956-0.188-0.097-0.022 0.954 First stage PopulationDensity -3.199*** -3.6697*** -4.78*** -4.829*** -4.882*** (-0.34) (.3137) (.3215) (.3139) (.314) DistancetoChina -8.083*** -7.07*** -10.224*** -10.49*** -9.281*** -8.275*** -11.932*** -12.523*** (-0.622) (-0.706) (.895) (.942) (.6819) (.74) (0.93) (.981) Lfh_status10.3014* (0.16) Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Partial R-squared 0.0833 0.1272 0.0868 0.11 0.1053 0.2301 0.2679 0.2972 0.2941 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. First-stage regressions indicate that two instrumental variables (DistancetoChina and Populationdensity) are highly significant and have expected signs (negative), in all specifications. Values of partial R2 and F statistics of models are rather high. DistancetoChina is better instrument, the resulting coefficient of ExptoChina increases considerably, from -0.04 in model using PopulationDensity as instrument to -0.2 when DistancetoChina is used. The findings of F test is consolidated by Effective F test whose values exceed the critical values of conventional levels of 10% for different estimators to reject the null hypothesis of weak instruments, presenting that concern on weak instruments not serious. The models 6 to 9 use the combination of two instrumental variables. Although Sargan statistics raises concerns on overidentification in models 6 to 8, the last model with dynamic context satisfies this criteria. Moreover, findings (provided as request) using different estimation techniques (2SLS, GMM, LIML, 16

Fuller(1) and Fuller(4) show rather consistent results with coefficient of ExptoChina is in very narrow range, from -0.0075 to -0.0072. This indicates that weak instruments problem, which is considered to be mitigated more effectively by LIML or Fuller, is not a serious problem of my specifications. Finally, I check conditional LR test, Anderson-Rubin and Score (LM) tests. The findings show that interval confidence of LIML specifications is almost the same with confidence sets of conditional LR, Anderson-Rubin and Score (LM test) at 5% significance level. This consolidates the arguments that weak instruments are not a serious problem in my specifications. In short, there is strong and consistent evidence on negative impact of China on Democracy in developing countries. The findings overcome the most important tests on validity of model 4. 3e: Robustness Check To consolidate our findings, several robustness checks are implemented. I follow Acemonglu et al. (2008), Spilimbergo (2009) and Kollias & Paleolongou (2016) to use an alternative index of DemocracyFH: Polity Index (DemocracyPOL, Marshall 2014) and World Bank Governance Sub indicator of Voice and Accountability. Table 5: Robustness Check for DemocracyFH (1) (2) (3) (4) (5) (6) (7) (9) GMM GMM IV IV IV IV IV IV ExptoChina (t-1) -0.00369-0.00629** -0.0285** -0.189*** -0.108*** -0.134*** -0.0738*** -0.00234* (0.00346) (0.00264) (0.0120) (0.0208) (0.0118) (0.0154) (0.00797) (0.00124) DemocracyPOL (t-1) 0.868*** (0.0664) DemocracyWB (t-19 0.837*** 0.985*** (0.0840) (0.00543) Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes F statistics 88.46 205.34 168.88 158.53 145.54 123.27 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Dependent Variable is DemocracyPOL and DemocracyWB exactly identified exactly identified exactly identified 0.0004 0 0.4755 Sargan statistic Sargan test 0.532 0.001 Hansen test 0.574 0.103 AR(2) 0.513 0.011 Number of Instruments 17 19 Difference-in-Hansen tests All system GMM instruments 0.576 0.142 Those based on lagged democracy only 0.33 0.074 Observations 994 1,078 984 1,084 1,172 912 984 984 R-squared 0.036-0.308-0.516-0.102-0.183 0.976 Number of countries 92 99 First stage PopulationDensity -3.198*** -3.464*** -3.669*** -3.61*** (0.34) (0.306) (.313) (.315) DistancetoChina -8.72*** -8.083*** -9.897*** -9.281*** -8.8989*** (0.61) (0.622) (0.667) (0.682) (0.725) DemocracyWB (t-1) -.4209 (0.27) Year Fixed Effects Yes Yes Yes Yes Yes Yes Partial R-squared 0.0833 0.1609 0.1272 0.2605 0.2301 0.2025 4 However, we do not examine interaction effects of GlobalizationKOF and ExptoChina on Democracy because the inefficiency of using time-invariant instruments for ExptoChina and the uncertainty of exogeneity of GlobalizationKOF can provide incorrect interpretation. 17

Together with Freedom House Democracy Index, Polity Index (DemocracyPOL) is also very popular in literature on Democracy and is often used as another option of Freedom House index. World Bank Governance Indicators include six components: Voice and Accountability, Political Stability, Government Effectiveness, Control of Corruption, Rule of Law and regulatory Quality (Kaufmann et al. 2010). However, according to Baird (2012), only Voice and Accountability can reflect elements of Democracy while the last four components might present governance infrastructure. Following this, we just use Voice and Accountability (DemocracyWB) as an alternative for Freedom House Index. Both alternative indexes are converted to 0-10 scale for convenience of comparison. Table 5 replicates main specifications relevant to DemocracyFH, the results are almost qualitatively similar with the original specifications using Freedom house Democracy Index. The second sensitivity check is use an alternative for GlobalizationKOF. Global Connectedness Index 2016 (GlobalizationDHL) developed by Ghemawat & Altman (2017) is a comprehensive index to measure globalization from different aspects: international flows of products and services, capital, information and people. The findings with this index are highly consistent with those using GlobalizationKOF. Table 6: Robustness Check for GlobalizationKOF GMM GMM GMM GMM Dependent Variable is DemocracyFH DemocracyFH (t-1) 0.880*** 0.915*** 0.915*** 0.916*** (0.0860) (0.0838) (0.0820) (0.0766) GlobalizationDHL (t-1) -0.0163-0.00196-0.00196-0.00305 (0.0151) (0.0128) (0.0130) (0.0121) ExptoChina (t-1) -0.0382** -0.0310** -0.0310** -0.0310** (0.0174) (0.0148) (0.0152) (0.0143) GlobalizationDHL*ExptoChina (t-1) 0.000957** 0.000710* 0.000703* 0.000693** (0.000419) (0.000366) (0.000371) (0.000353) GDP (t-1) 0.170 0.166 0.225* (0.136) (0.141) (0.136) FrenchLaw -0.0919 (0.162) Communist Law -0.190 (0.209) BritishColony 0.0274 (0.133) Year Fixed Effects Yes Yes Yes Yes Sargan test 0.038 0.068 0.069 0.056 Hansen test 0.044 0.098 0.099 0.039 AR(2) 0.962 0.94 0.94 0.937 Number of Instruments 21 25 26 27 Difference-in-Hansen tests All system GMM instruments 0.358 0.089 0.09 0.087 Those based on lagged democracy only 0.86 0.974 0.986 0.828 Observations 663 654 654 654 Number of countries 74 73 73 73 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. 4: Qualitative Analysis The following part compares Zimbabwe and Kenya under most similar cases design. While statistical techniques are rather young and have developed strongly almost recently, this most similar approach has 18

long history in social sciences research, firstly in S. Mill s (1872) s System of Logic (Seawright & Gerring (2008); George & Bennett (2005)). George & Bennett (2005) and Seawright & Gerring (2008) provides a review and guidance on this approach in practice in relevance with corresponding statistical methods and other case selection alternatives. 4a: Most similar Analysis with Two Cases There are several case selection procedures: typical, diverse, extreme, deviant, influential, most similar, and most different cases (Seawright & Gerring (2008). Seawright & Gerring (2008: 304-305) analyses these procedures in comparison with corresponding statistical method, indicating that matching in statistical originates from the most similar design. If our dependent variable is Y, independent variables of interest is X1, most similar cases are defined as cases (two or more) are similar on specified variables other than X1 and/or Y. (Seawright & Gerring (2008:298)). The rationale of this procedure is described in the table 7. Table 7: Most Similar Analysis Design Case Variables X2 X1 Y "Most Similar" Different Different 1 + + + 2 + - - Note: Plusses and minuses represent the score demonstrated by a case on a particular dimension (variable), coded dichotomously. X1 = the variable of theoretical interest; X2 = the background/ control variable or vector; Y = the outcome. Source: Seawright & Gerring (2008:305) In this case, Zimbabwe and Kenya have the similar Democracy level and economic-political-social background before 2003 and even afterwards (the similar X2). However, currently, two countries show opposite trends, one (Zimbabwe) has been more and more autocratic while the other (Kenya) has improved markedly its political system (different Y). If China s involvement is significantly different between the two countries (Different X1) (while the two countries experience similar situations and face similar challenges in democratization process), it is concluded that China contributes to this difference, albeit not all (X1 is a cause of Y). To make the causal claim convincing, we should prove (qualitatively) that there is a clear causal mechanism running from China s involvement to resulting Democracy (from X1 to Y). This might not be implemented by statistical tools, because statistical method concerns only numbers but not socially causal linkages behinds such numbers, numbers go through quantitative method like going into Black box (Collier et al. 2004). It is difficult to find observations with exact matching, similarly, it is almost impossible to find cases with the absolutely identical backgrounds. While Zimbabwe and Kenya are selected in my design, I am well aware that it is too careless to say that China is the only source of the Democracy divergence. 19

The following part is structured as: firstly, comparing Zimbabwe and Kenya before (and during) the Chinese involvement to prove that both countries have the similar departure; secondly, explaining the possible mechanism that China may affect the democratization process of two countries to indicate a clear difference of intervention levels; thirdly, main assessments derived from comparing the consequences of linkages with China. 4b: Zimbabwe and Kenya: Most similar cases Zimbabwe and Kenya share numerous similarities in political, economic and social dimensions, especially before 2000s. Firstly, both countries show similar demography and geographical characteristics. Both countries are developing African states with middle-sized population: Zimbabwe with around 13 million people while Kenya with 31 million people. The total area in Zimbabwe is about 387000 sq. km compared to 570000 sq. km in Kenya (CIA Factbook). Younger population can expect more political improvements and male youth might be a source of social and political violence, destabilizing young democracies (Cincotta (2008); Urdal (2006); Lutz et al. (2010); Weber (2012)). Both Kenya and Zimbabwe have a young population of median 20 years old and differences in education opportunities by gender are similar in Zimbabwe and in Kenya, even the situation can be a little better in Zimbabwe than in Kenya (in 2000, age standardized education per capita of Female/Male in Kenya is 3.9/6 while Zimbabwe 5/6.8, the population standardized figures are similar, data changes inconsiderably over time, IHME (2015)). The two countries also are at the same development level, with similar GDP per capita and human development index both at low levels, even Zimbabwe shows a little better indexes. Namely, in years around 2000, GDP per capita (constant 2005 $US) in Zimbabwe is around 700 while that of Kenya is about 500 compared to 777 weighted mean of the Sub-Saharan Africa region (48 countries) (World Bank (2016)). Human development index of Kenya in 2000 is 0.45 similar with 0.42 level of Zimbabwe, (the weighted mean of the region is 0.39) (UNDP (2015)). Human capital index, (based on years of schooling and returns to education) of Kenya in 2000 is around 1.93 while that of Zimbabwe is 2.06 (Feenstra et al. (2015)). Similarly, gross enrolment ration for primary school of Zimbabwe is 101.16% while hat of Kenya is 95.6%, both are higher than average regional level of more than 86%. In secondary school, the gross enrolment ratio in Kenya is 39.3% compared to 42.7% in Zimbabwe and 28.6% in the region during the similar period. Age standardized education per capita of Zimbabwe is about 5.95 higher than 4.95 level in Kenya in 2000 (IHME (2015)). These economic and development indexes that might drive political development in developing countries show the similar departure points for Zimbabwe and Kenya, even Zimbabwe is at better position and under more favourable conditions, albeit marginal, to democratize its political system. Not less important, there are several historical and social indicators that are considered as important determinants of Democracy are also in high similarity between two countries. Regarding history, both countries were former British colonies, which is believed to exert positive impacts on democratization process later on. Both countries use English as official language, which is a big advantage for democratization (CIA Factbook on Kenya and Zimbabwe). Alesina et al. (2003) produces Fractionalization index to measure the probability that two randomly selected people from a given 20

country will not share a certain characteristic, the higher the number the less probability of the two sharing that characteristic in ethnicity, language and religion (cited from Teorell et al. (2016)). A higher fractionalization means higher possibility of social conflicts, which deteriorate elements of young democracies. Kenya has higher ethnic fractionalization at 0.86 while Zimbabwe 0.39 and the weighted average of region is 0.77. Similar numbers are seen in language fractionalization, with 0.89 for Kenya, 0.45 for Zimbabwe and 0.77 for the whole region; ethnolinguistic fractionalization by Roeders (2001) with 0.88; 0.47 and 0.80, respectively. Religious fractionalization is similar in both countries, around 0.78 in Kenya, 0.74 in Zimbabwe and 0.64 in the whole region. From cultural and religious perspectives, Huntington (1991) shows that Islam and Confucianism might constrain democratization while Catholicism can facilitate democratization, at least during the third wave of democratization. This argument is proven by empirical evidence with different measurements of Democracy by Rowley & Smith (2009) and Potrafke (2012). The proportion of Muslim population in Kenya is 11.2%, more than 10 times higher than Zimbabwe (1.2%); while the proportion of Christianity in Kenya is 83%, lower than in Zimbabwe with 93 % (CIA Factbook, data in 2009 for Kenya and 2011 for Zimbabwe). Given the low proportion of Muslim population and very high proportion of Christian population, even if religion is an important determinant of democratization, it could not change dramatically Democracy landscape of the two countries. To conclude, social and historical features seem to put Zimbabwe in better position to democratize its political system, albeit not considerably higher than Kenya. Besides domestic factors, external factors could have significant impacts on Democracy of countries, especially developing and not large countries like Zimbabwe and Kenya. Most important external determinants are Democracy of bordering countries, which is referred to democratic diffusion concept (Brinks & Coppedge (2006)), Democracy aid; and as focus of our paper, Globalization or Westernization. Firstly, Democracy in neighbour countries show more favourable conditions for Zimbabwe. For Kenya, Freedom House index in 2000 indicates that out of five neighbouring countries of Kenya, there are no Free countries, three Partly Free (Tanzania, Uganda, Ethiopia) and two Not Free (South Sudan and Somalia), an obvious disadvantage compared to Zimbabwe, out of four Zimbabwe s neighbours, there are two Free countries (South Africa and Botswana) and two Partly Free countries (Zambia and Mozambique) (Freedom House). In better situation, Zimbabwe has rather democratic neighbours. Democracy promotion can be seen at aid, or even more details, Democracy aid from DAC countries. According to aid statistics from OECD, the total aid Kenya committed to receive from DAC donors from 1978 to 2001 is around 20 209 $US millions, double that of Zimbabwe with 10 762 $US millions (2014 price). Considering that population of Kenya is about 2.5 times of Zimbabwe, total aid committed for Zimbabwe is higher than that of Kenya. A better indicator of Democracy promotion is Democracy aid that is aid for government and civil society. Even population of Kenya is 2.4 times higher than that of Zimbabwe, the total committed aid for Zimbabwe in government and civil societies activities are higher than that of Kenya, 246.8 $US millions compared to 204.9 $US millions since 1995 to 2001 (data before 1995 is not available). 21

Figure 1: Globalization Index of Kenya and Zimbabwe 1970-2014 In terms of Globalization, both figures of Kenya and Zimbabwe are in common trend and in general are higher than the average whole region. The figure below shows this similar trend that is persistent overtime in both countries, although during 2002-2010, Kenya has exceeded marginally Zimbabwe in several continuous years, indicating the relative decrease of Westernization process in Zimbabwe and contradict pattern in Kenya. In general, Zimbabwe has a little better condition to build up its Democracy from not only long-term causes like religion or societal fractionalization, but also medium factors like education and income level. In reality, looking at Democracy level in the past, the two countries have almost the same level, even Zimbabwe performed rather better since 1980s (Figure 3). The following part presents how differently China involves in two countries. From the logic of most similar cases design, if the Chinese engagement is considerably varied, while Zimbabwe and Kenya have similar inputs of Democracy, we can come to conclusions that China is one important cause explaining the Democracy divergence in these two countries. 4c: China s involvement in Kenya and Zimbabwe Based on theoretical framework developed in Burnell (2007, 2010) and Vanderhill (2013), the mechanism that China negates Democratization in developing countries can be seen in different channels: the natural diffusion of China economic miracle, hard or soft pressure that China imposes in target countries, political supports in international arena and just keeping their business as usual. These channels operate mostly through the elite class. The following parts show the different fields (cultural, economic, political and military relations) that China involves and affects the economic, political and societal situations in its partners. Firstly, in the field of cultural relation, Confucius Institutes, the symbol of Chinese culture has been established firstly in 2005 in Kenya and only one year later in Zimbabwe. Currently, four Confucius Institutes have been in operation in Kenya and 1 in Zimbabwe. Confucius Institutes began in 2004, working 22

as a tool of soft power of China, aiming to, in rhetoric statement, enhancing understanding of the Chinese language and culture by these peoples, to strengthening educational and cultural exchange and cooperation between China and other countries, to deepening friendly relationships with other nations, to promote the development of multi-culturalism, and to construct a harmonious world (Article 1, Constitution and By-Laws of the Confucius Institutes, Hanban). As an information channel, Confucius Institute works firstly to bring China closer to the world and show up China imagines. Until now, it has 110 member institutes in Asia, 46 in Africa, 157 in America169 in Europe and 18 in Oceania. However, over a very short period of operation, it has been highly criticized in numerous universities in the USA and Canada because of concerns on academic freedom (Lahtine (2015)). Despite of this, Chinese becomes increasing popular in Zimbabwe. With Confucius Institutes and other Chinese language classes, Zimbabwe leads the rest of the continent in the training of local teachers of Chinese and is poised to export surplus teachers of Chinese to other countries as well (Mukeredzi (2013)). Confucius Institute in Zimbabwe is ranked as the best institute in the whole continent, moreover, Zimbabwe is only one in the world that have local lecturers in teaching Chinese as a foreign language and also the country with largest number of indigenous people with a Master degree in Chinese (Mashininga (2013); Herald (2014)). In short, the cultural relationship between China and Zimbabwe is rather strong. At least in raw number of teachers and learners of Chinese language, Zimbabwe outnumbers Kenya. It is difficult to compare cultural ties between China and Kenya and Zimbabwe, but economic linkages are much easier to evaluate. In terms of trade, investment, aid and other economic relations, China s engagement in Zimbabwe is much stronger than in Kenya. Firstly, in agriculture, according to Land Matrix project, China is the biggest investor country with 4 out of 8 recorded large scale land acquisition deals in Zimbabwe while there is no Chinese investors who buy large scale land in Kenya since 2000 (Land Matrix 2016). More obviously, the total Chinese foreign direct investment per capita since 2003-2012 in Zimbabwe is 7 times higher than that of Kenya. Namely, the total foreign direct investment from China is 2.5 times higher than that in Kenya since 2003-2012(UNCTAD (2016)). Total export to China in Zimbabwe is 8. 87 times higher than that in Kenya from 2002-2014, if taking per capita, this difference is around 25 times (IMF). If taking the proportion of export to China out of total export, export to China accounts for more than 13.7% in Zimbabwe while only 0.7% in Kenya on average 2002-2014 (IMF). Recently, alongside the United States dollar, Zimbabwe decided to make Chinese yuan legal currency to increase their trade with China (The Guardian (2015a)). Regarding financial aid from China during 2002-2013, on average one Zimbabwe person receives 2.7 times higher than one Kenya person (total Chinese aid for each country is almost similar, more than 7.3 $US billions, updated from Tierney et al. 2011). However, financial aid from China to Zimbabwe is proven as a tool to patron the elite of Zimbabwe: Chinese money is the political preservation of the Mugabe reign and personal aggrandisement through corruption and kickbacks by his ZANU-PF cronies flowing from Chinese investments (Karumbidza (2007), cited by Hodzi et al. (2012)). This is in stark contrast with financial supports from the West, which always requires conditions on political reforms and human rights protection. 23

Hodzi et al. (2012) shows that generous aid from China works as alternative for ODA from the West. However, Chinese assistance without any conditions of good governance increased corruption amidst diminished political accountability, and undermined the role of civil society within the country (Hodzi et al. (2012: 96)). To compare military relation between China and Zimbabwe and China and Kenya, I look at Chinese military equipment imported by two countries and high-ranking military official visits. Total import of military equipment from China in Kenya since 2002 is 40 million, in Zimbabwe is 42 $US million (at constant 1990 prices) while population of Kenya is almost 2.5 times higher than Zimbabwe. China becomes the biggest military exporter of Zimbabwe in the recent decades (SIPRI). On other perspective, since 2003, China has three high-ranking visits to Zimbabwe to promote military cooperation (in years 2004, 2009 and 2014) while the number is two for Kenya (in years 2006, 2010). Moreover, while Zimbabwe are imposed arms embargoes by the West, importing military equipment from China strengthens Mugabe s regime, especially when there were reports of murders and torture being perpetrated by government sponsored militia and China s assistance to Zimbabwe will be shown to have served to fuel internal conflict as it compensates combatants (Hodzi et al. 2012: 91). With regard to political relations, the recognition of China towards the repressive government of Zimbabwe headed by President Mugabe could be a better proxy. Despite of electoral fraud in Zimbabwe, China still recognizes the leadership of Mugabe. Recently, Zimbabwe's Robert Mugabe was awarded Confucius peace prize, which is considered as 'China's Nobel peace prize' (The Guardian (2015b)). In 2008, China, together Russia, vetoed a resolution of UN Security Council that attempted to put sanctions against Zimbabwe due to the violation of human rights and free election that President Mugabe must be responsible for (Lynch 2008). In addition to this, the intervention of China on Zimbabwe s presidential election has been detected by a journalists collective (Moore (2014: 103)). One of the released information states that: documents from Zimbabwe s Central Intelligence Organization obtained by 100Reporters suggest that the success of Mugabe and his ZANUPF party reflected direct intervention by the Chinese Communist Party, financial support topping $1 billion in diamonds and revenues from three companies and two African presidents, armed intimidation by security forces and vote rigging en masse (Sharife (2013)). In conclusion, the political, economic and diplomatic supports of China for the elite in Zimbabwe by giving them financial promises and political commitments through different channels can contribute to the stability of autocracies and definitely help the autocratic leaders stay longer in power. This becomes more dangerous in the political context of Zimbabwe when the West abandons their traditional supports, making China the only player in the battle of Zimbabwe for Democracy. It cannot be denied that China also has more and more effects in Kenya, but obviously, in Kenya, China faces at the same time efforts to promote Democracy of the West and international organizations, which lack in Zimbabwe currently. This can be seen in Figure 24

2, where Globalization in Zimbabwe tends to slow down while that of Kenya outperforms, albeit unstably, in several points of time after 2002. 4d: China, Globalization and Democracy in Zimbabwe and Kenya It is concluded that Zimbabwe and Kenya started from the similar departure, even Zimbabwe has some advantages. One important difference is the involvement of China in both countries. The more China involves in its partners, the more autocratic their partners would be. However, it should be noted that China is not the only determinant of the divergence of Zimbabwe and Kenya. There are several factors other than China might explain the divergence of Zimbabwe and Kenya analysed in literature. The first and probably most important determinants is the domestic factor, namely the leadership of Mugabe and his fraud government in Zimbabwe, which is driving force of backsliding Democracy situation in this country recently. This determinant of course is essential in explaining political landscape in Zimbabwe, however, it cannot be sufficient. It is reasonable that China is not the culprit for the start of political and social crisis in Zimbabwe, which began in 2000 when the European Union (EU) and then USA (in 2001) imposed sanctions after extensive fraud Zimbabwean parliamentary election (while China s go-out policy started in 2002, see Figure 3 for changes over time). However, the previous part shows that China, which stands behinds Mugabe and his government, is an important force and even a root for the stability of Mugabe s regime after this country is isolated by the West. Saati (2014) believes that political elites play central role in explaining why Kenya has better Democracy than Zimbabwe does. Saati (2014) traces back to history of cooperation and violence of the two countries to illustrate her arguments. It should be acknowledged that her arguments based on careful history investigation. However, the article ignores the very modern issue: the divergence of Democracy in two countries is not long, it has already occurred in the recent two decades. Secondly, the article does not show sufficiently how external factors dominate such small and poor countries like Zimbabwe and Kenya through elite class. It cannot be denied that China has strong impacts on Zimbabwe s elites and this is of utmost importance to analyse the different level of elite cohesion in the two countries. Figure 3: Democracy Index of Kenya and Zimbabwe 1978-2014 25