China s Ideological Spectrum

Transcription

1 China s Ideological Spectrum Jennifer Pan Yiqing Xu Harvard University M. I. T. Preliminary Draft Comments and Suggestions are Welcome! First Version: 10th April 2015 This Version: 12th April 2015 Abstract We offer the first large scale empirical analysis of ideology in contemporary China to determine whether individuals fall along a discernible and coherent ideological spectrum, and whether there are regional and inter-group variations in ideological orientation. Using principal component analysis (PCA) on a survey of 171,830 individuals, we identify one dominant ideological dimension in China. Individuals who are politically conservative, who emphasize the supremacy of the state and nationalism, are also likely to be economically conservative, supporting a return to socialism and state-control of the economy, and culturally conservative, supporting traditional, Confucian values. In contrast, political liberals, supportive of constitutional democracy and individual liberty, are also likely to be economic liberals who support market-oriented reform and social liberals who support modern science and values such as sexual freedom. This unidimensionality of ideology is robust to a wide variety of diagnostics and checks. Using post-stratification based on census data, we find a strong relationship between liberal orientation and modernization provinces with higher levels of economic development, trade openness, urbanization are more liberal than their poor, rural counterparts, and individuals with higher levels of education and income and more liberal than their less educated and lower-income peers. Keywords: ideology, public opinion, principal component analysis (PCA), survey, poststratification, authoritarianism, liberalism, conservatism, nationalism, China We are tremendously appreciative of the efforts made by David Mao and his team, who collected and publicly released the Zuobiao data of 2014 (zuobiao.me). We thank Justin Grimmer, Simon Jackman, and Dingding Wang for their extremely helpful comments and suggestions. Department of Government, Harvard University. jjpan@fas.harvard.edu. Department of Political Science, Massachusetts Institute of Technology. xyq@mit.edu.

2 1 Introduction A monolithic ideology is often described as a key characteristic of authoritarian and totalitarian regimes (Friedrich and Brzezinski 1965; Inkeles 1954; Linz 1975, 2000; Neumann 1957), while an ideological spectrum ranging from conservative to liberal is typically associated with regimes that have electoral competition. For some scholars, China under Mao approached the ideal type of totalitarianism where social control was diffused and reinforced by a dominant ideology motivating voluntary involvement of individuals (Linz 2000). In Maoist China, commitment of the ruling elites to the ideology of Marxism-Leninism imposed constraints on the range of heterodox policies and formed the basis for policies falling within those constraints (Ahmad and Hussain 1991; Barnett, Hofheinz and Solomon 1972; Whyte and Parish 1984). However, China s reform and opening up in 1978 ushered in policies that bled past the boundaries set by these prior ideological constraints. China s current mode of economic production one in which private ownership and market forces play an increasing role and one that has changed the county s social-economic structures produces contradictions for the Chinese Communist Party s ideology rooted in Marxism. The Chinese Communist Party (CCP) has put much effort in crafting a coherent ideological message to take into account the country s economic and social changes (Brown 2012). However, these economic and social policy changes, and indeed the CCP s own efforts to elaborate its ruling ideology, mean that ideology in China is no longer monolithic. 1 is a highly contested term, containing a range of meanings (Eagleton 1991). The Oxford English Dictionary defines ideology as a system of ideas and ideals, especially one that forms the basis of economic or political theory and policy. In the social science literature, ideology has been defined as a configuration of ideas and attitudes in which the elements are bound together by some form of constraint of functional interdependence (Converse 1964), or a cultural system that place particular symbols and particular strains (or interests) side by side (Geertz 1964), or simply a meaningful and organized representation of public policy preferences or opinion of elites or the mass (e.g., Poole and Rosenthal 1991, 2000; Heckman and Snyder 1997; Ansolabehere, Snyder and Stewart 2001; Stimson 2004; Ansolabehere, Rodden and Snyder 2008; Treier and Hillygus 2009; Stimson 2012). 2 We draw on the third definition in order to operationalize the quantitative analysis presented in this paper. 3 1 There is debate on whether ideology was ever monolithic, even at the height of Maoism. A monolithic ideology is an ideal type, and Maoism formed a dominant ideology that approached this ideal type. 2 See also Geiger (1932) for a Marxist interpretation of ideology and Eagleton (1991) for the history of the concept of ideology. 3 Cultural-psychological structure of the Chinese people ( 中国人的文化 - 心理结构 ), a concept proposed 1

3 Ideas associated with conservatism on the left ( 左 ) and liberalism on the right ( 右 ) are increasingly common subjects of discourse in China. These concepts of left and right are not yet fully elaborated or organized, but broadly speaking, Chinese intellectuals have defined the core ideological divides in China as those between conservatives on the left who support a socialist (authoritarian) state, who emphasize national unity and security, who think highly of the old communist/socialist economic system, and who value traditional culture, versus liberals on the right who advocate for a constitutional democracy, who embrace individual liberty, who support market-oriented reform, and who are enthusiastic about modern science and technology. Conceptions of left and right in China differ from their usage in the North American and Western European context. We argue this is because of China s legacies of imperial autocracy, Confucianism, Communism, and a tradition of using ideology to legitimize the rule of the dominant political power. The placement of liberals and conservatives on a left to right scale also differs between China and consolidated democracies. Instead of liberals on the left and conservatives on the right, contemporary Chinese public discourse places conservatives on the left and liberals on the right. For example, a person who supports the rights of homosexual communities in the U.S. is considered a liberal on the left while a person who supports gay rights in China is considered a liberal on the right. However, a person who supports laissez-faire economic policies in the U.S. is considered a conservative on the right while a person who supports market-based economic policies in China is considered a liberal on the right. We elaborated what left and right connote in political, economic, and social/cultural arenas in Section 2, where we also explain why conservatism and liberalism are respectively connected with the left and right ends of the ideological spectrum. 4 Although Chinese intellectuals and public opinion leaders use the terms left and right in writings and conversations and many have discussed the idea of China s ideological spectrum, our knowledge of the ideological orientation of ordinary Chinese people is extremely limited. Answers to many basic and important questions remain unknown. First, do ordinary citizens in China exhibit divergent ideological views? People may have diverse opinions on different issues, but their policy preferences may not be characterized by ideological divides. Second, if citizen preferences can somehow be represented by a set of stable and organized beliefs, is there a single ideological dimension that can capture different modes of thought among ordinary Chinese? For example, are those who are politically conservative also economically and socially conservative? Or, are there multiple ideological dimensions, as research suggest by the prominent Chinese philosopher Li Zehou, is also an appropriate characterization of the ideological spectrum we attempt to measure. See Li (2008). 4 The ideological spectrum can certainly be multi-dimensional. See detailed discussions in Section 5. 2

4 is likely the case in the United States and Western Europe (e.g., Kitschelt 1994; Marks et al. 2006; Treier and Hillygus 2009; Stimson 2012)? Third, if a dominant ideological spectrum exists and we label its two ends as liberalism and conservative, what do these terms actually mean in the context of contemporary China? Last but not least, if individuals fall along one or several ideological spectra, what types of individuals are conservatives and liberals? For example, are those who are better educated more liberal or more conservative? What regions are more conservative or liberal? Is economic development accompanied by more liberal or more conservative orientations? Our goal in this paper is to explore the empirical realities of ideology in China and determine whether individuals fall along a discernible and coherent ideological spectrum, how ideological orientations vary, and to provide an approach for measuring ideology through survey responses that do not constitute a random or representative sample. The data source we use comes from the Chinese Political Compass (CPoC) website (zuobiao.me), a survey of 50 items that reflect intellectuals conceptualization of China s ideological split to evaluate individual attitudes toward political, economic, and social/cultural issues. The publicly available sample includes responses of 171,830 individuals who completed the CPoC survey in It is not a random sample or representative of a certain population because anybody in the world can visit the website and answer these questions, and the answers will be recorded as valid observations. However, as we demonstrate in detail in Section 5, the non-representativeness of the sample does not prevent us from achieving the main objective of this paper to identify the ideological spectrum of ordinary Chinese. Using principal component analysis (PCA), a simple, efficient, and widely used statistical tool for analyzing policy preference, 5 we are able to identify one dominant ideological dimension in China. Monte Carlo methods show that this dominant uni-dimensional factor space (a line) estimated from vastly different samples remains extremely stable. Our diagnostics and robustness checks suggest that such a dimension, as defined by the 50 concrete questions in the CPoC, would likely remain the same if we had a scientifically designed representative sample based on the same set of questions. We find the latent space separately estimated from respondents answers to political, economic, and social/cultural items to be highly colinear in other words, in contrast to studies of political ideology in advanced democracies where preferences on economic and social issues often constitute two ideological dimensions (though they are correlated), respondents in China who are politically conservative are much more likely to be economically and socially conservative while respondents who are politically liberal are also much more likely to be economically and socially liberal. 5 See, for examples, Heckman and Snyder (1997); Ansolabehere, Snyder and Stewart (2001); Ansolabehere, Rodden and Snyder (2008); and Stimson (2012) provides a variation of it. 3

5 An alternative and increasingly popular method for estimating latent policy preferences is the Bayesian Item Response Theory (IRT). 6 Compared with PCA, the IRT approach has several advantages. First, it allows nonlinear functional forms and does not requires a multivariate normal distribution for the survey questions. In other words, observed variable can be dichotomous, ordinal, or continuous. Second, it is easier to make inference of the latent measures using an IRT model. Third, an IRT model can deal with the missing data problem and the situation in which respondents answer Don t know. Fourth, it can bridge the information from multiple survey and provide more precise estimates of the latent measures. In Section 3 where we describe the CPoC data, we explain why none of these advantages is particularly relevant for the goals of this paper and for the particular dataset we use. In contrast, PCA has the advantage of being extremely fast and transparent, allowing us to conduct multiple diagnostics and robustness checks in a short period of time. 7 However, neither PCA nor IRT models can solve the problem caused by a non-random and non-representative sample, which is a major concern if we want to make regional, temporal, or inter-group comparisons of ideology. We address this concern by reweighting the sample according to two census population targets (based on gender and age) after running PCA on the dataset. After adopting this post-stratification strategy, we examine the relationship between ideology and individual-level socioeconomic characteristics and across spatial dimensions. At the provincial level, we find that liberal ideology is highly correlated with modernization, measured by the level of economic development, trade openness, and urbanization. Coastal provinces (and municipalities) that are more economically developed, such as Shanghai, Guangdong, and Zhejiang, are much more liberal than poorer interior provinces, such as Guizhou, Guangxi, and Henan. At the individual level, we find a strong positive relationship between liberal ideology and both education and income. These results strongly confirm our priors and seem highly intuitive; however, we must bear in mind the possibility that sampling problems could mean there is significant bias in the estimates of the ideological orientation of a particular province or in the correlation coefficient between ideology and a particular individual-level variable. 8 The contribution of this paper is several-fold. First, it is the first project, to our knowledge, that defines and locates the ideological spectrum of Chinese citizens based on largescale empirical analysis. 9 Previously, the terminology of Chinese liberalism versus conservat- 6 See, for examples, Clinton, Jackman and Rivers (2004); Treier and Hillygus (2009); Tausanovitch and Warshaw (2013); Caughey and Warshaw (2015). 7 In the near future, we plan to use a flexible IRT model with the same data to cross-validate our ideology measure. We expect results from the two methods to be extremely similar. 8 To render the empirical pattern we identify void, one needs to imagine a very strange pattern of people in different places and of different social background answering questions on the CPoC website. 9 Using a sample of CPoC data in earlier years, a previous study shows that there are two dominant idealo- 4

6 ism and their actual meanings were not areas of great familiarity for most observers outside of China. We remedy this gap in knowledge through this work. Second, this project provides evidence that ideology in China today is strongly uni-dimensional, a fact with important political and policy implications. For example, if economic conservatism (e.g., opposition to the free market) is closely aligned with political and cultural conservatism (e.g., nationalism and anti-western sentiment), it may be an effective strategy for autocratic leaders to pander to a nationalistic base when facing economic challenges. Or another example, officials or politicians seeking public support from a conservative base may create coalitions promoting both conservative political and economic policies. Third, this project suggests that modernization, characterized by economic development, urbanization, increasing incomes, higher levels of education, as well as exposure to ideas from the Western democratic context, is tied to changes in ideological orientation. This finding also has practical implications, for example, it suggests that a liberal policy agenda is more likely welcomed by the rich and well-educated, but less by those who are poor, less educated, and culturally conservative. The arrangement of this paper is as follows. Section 2 discusses various meanings of conservatism and liberalism in contemporary China and give tentative explanations why they are different from the counterparts in the western world. Section 3 describes the data we use and Section 4 introduces our methods. The principal component analysis and resulting ideological dimensions are presented in Section 5. We then present the relationship between ideology and other variables, such as demographics, geography, and time, in Section 6. We discuss directions for future research in the concluding Section 7. 2 Ideological Spectrum in Contemporary China China s ideological spectrum, if it exists, derives from an overlapping mixture of diverse ideas from multiple intellectual sources, representing competing intellectual traditions ranging from Confucianism and Chinese Legalism, to anarchism, Marxist-Leninism, and classical liberalism. While in its manifestation, ideology must be simple enough to resonate with any ordinary person, the components of China s contemporary ideology is rooted in the country s history. In order to understand why the belief system of the Chinese population is what it appears today, we need to understand how ideas emerged, came into conflict, and receded into the background over the past one hundred and fifty years. Intellectuals have divided China s ideological development in recent history into five gical groups among the respondents (Wu 2013). We emphasize the ideological spectrum a low dimensional space that can meaningfully represent opinions of most respondents. 5

7 phases. 10 The first phase, prior to the mid 19th century, is characterized by traditional society with sacred values playing a central role in the system of ideas. The second phase, from the late Qing and Republican era in the mid 19th century to 1949, was a period of secularization, focused on increasing national strength and wealth as well as the liberation of the individual. 11 During this period, China interacted frequently with foreign powers and Chinese intellectuals start to introduce western liberal ideas, especially the work of enlightenment writers, into China. However, these ideas were mostly confined to small intellectual circles. In the Republican era of s, fast-paced development of commence and industry facilitated secularization of the Chinese society. However, even after the imperial system was abolished with the fall of the Qing, traditional cultural resources were regularly used by revolutionists in the pursuit of their goals (Perry 2012). The third phase during Mao s reign from 1949 to 1978 was a period of radical idealism, where boundaries between the individual and the state dissipated and where individualistic identity was subsumed by the will of the state. Countless political campaigns and social movements greatly altered the belief system of the Chinese people. The political compass of left and right was constantly redefined by Mao. This era of the Cultural Revolution was later described as the extreme left, a term officially recognized by the CCP. 12 Since the Maoist era of revolution, the old political discourse of left and right was almost entirely reserved. In minds of the Chinese populous, no period could be more radical than radicalism of the Cultural Revolution. In some sense, the entire spectrum of ideology dominant in the west from socialism to conservatism are all too conservative (and thus on the right) compared with Mao s radicalism. The fourth phase, from 1978 to 1990 was again marked by secularization and modernization as well as a rejection of the Cultural Revolution and a reestablishment of boundaries between the individual and the state. During this period, there was broad intellectual and social consensus on the need for market incentives in other words,the rejection of socialist economic institutions as well as on a rejection of absolutism or a freeing of the constraints on the mind (Wang 1998). Economic reform and liberalization in both social and political spheres are all aspects of the rejection of the Mao era, and thus are all regarded as moving towards a liberal right). 10 See, for examples, Jilin Xu, The Spiritual Lives of Chinese in a Secular Society [ 许纪霖 : 世俗社会的中国人精神生活 ] and Zehou Li, History of Modern Chinese Thought [ 李泽厚 : 中国现代思想史论 ]. 11 Li (2008) famously argued that emphasis on individual freedom gradually gives way to pursuit of a strong nation because of the hostile international environment National salvation overwhelms enlightenment ( 救亡压倒启蒙 ). 12 Resolutions On Certain Questions in the History of the Party Since the Founding of the Nation. June 27, [ 关于建国以来党的若干历史问题的决议 ], available at /71387/71588/ html. 6

8 Since 1990s, motivated in part by social changes increasing inequality, materialism, lack of interests in civic participation there is increasing debate in China about whether the market is the solution to problems China faces or the cause of these problems. This debate about the market is accompanied by debates about the solution whether the solution necessitates a return to the previous socialism system, whether the solution is deepening market reform, or whether the solution is some third alternative. 13 Nostalgia for the Maoist era is common among people who are unsatisfied with the current situation. Although many of those who are nostalgic are too young to have experienced the Cultural Revolution, they have begun to revitalize the symbols, ideas, and even conduct of the Mao era (Barmé 1996; Friedman 1994; Misra 2001; Swislocki 2009). At the same time, China s economic reform and opening-up has brought in an influx of new ideas as well as scientific and technological advances, and has dramatically lifted the standard of living for the Chinese population. Economic openness has also brought in waves of new information and norms that are changing public opinion The beneficiaries of market reform have become natural supporters of further economic liberalization and are usually against state intervention in the economy and infringement of individual property rights. These beneficiaries have incentives to embrace some liberal institutions, such as electoral competition and civil liberties, and have the tendency to believe that these institutions may be effective, since the market reform has borne fruits. These debates are not occurring in a vacuum, but in a context where the CCP has been and continues to be acutely invested in using ideology, with a special emphasis on national unity, independence, and strength, to legitimize its rule and to maintain its dominant political power. For example, the CCP s concept of the Three Confidences to achieve the China Dream. 14 relates to the question of whether western constitutional democracy should be introduced to China. Pieces of concepts and ideas from all five periods contribute to the evolving system of ideas that constitute ideology in China today. Table 1 lists the ideas typically associated with the conservative left and the liberal right. They include ideas mentioned by numerous intellectuals and public opinion leaders in their discussions of Chinese ideology (e.g., Chen 2007, 2012; Li 2008; Pan 2003; Qin 2006; Ren 2013; Wang 1998; Wang 2006; Wang 2011; Xu 2007a; Xu 2007b, 2009; Yao 2006). As Table 1 shows, ideas of the conservative left include nationalism, a strong and powerful state, big government, Chinese identity, and cultural conservatism. 15 On the liberal right, ideas include 13 For examples, see Yang Yao, The Complexity of Problems of Contemporary China 姚洋 : 当代中国问题的复杂性. Available at 14 See 15 In recent years, the CCP has started to put effort in promoting traditional Confucian thought and values. 7

9 Table 1. Left and Right in Comtemporary China (left) (right) Political National unity Civil liberty Party-state Electoral competition Socialist state/statism Constitutional democracy National security Transparency Maoism Legalism Chinese characteristics Universality State sovereignty Human rights Patriotism Internationalism Anti-West Pro-West Economic Welfare state* Free market intervention Less intervention Public interest Individual rights Legitimacy of labor gains Legitimacy of property gains Social/Cultural Traditional values Western values Collectivism Individualism Confucianism Individual freedom Traditional wisdom Modern science Thriftiness Materialism * Note: See footnote 16. constitutional democracy, and its legal and societal foundations of legal order, protection of individual liberties, and government transparency. Promotion of a free market economy, which is considered conservative in the context of developed democracies, is on the right in China because it is the most significant rejection of the radicalism of the Maoist era. 16 and more, the most central ideological issue seems to be political: upholding a socialist (authoritarian) state on the left versus advocating for a liberal constitutional democracy on the right. Indeed, the appropriateness of democracy and liberalism is a core feature of the ideological debate in contemporary China while this is not the case in consolidated democracies. Political ideology in the U.S. and Western Europe tend to be variation on classic liberalism rooted in the ideas of political theorists such as John Locke, Montesquieu, Adam Smith, Immanuel Kant, John Stewart Mill. In the U.S. and Western Europe, the vast majority of individuals believe in representative democracy, civil liberties, and a market economy. Parties in these regimes differ on how best to achieve these goals, through more competition or more protection for instance, but the ideological debate does not question 16 The label welfare state is somewhat misleading in the Chinese context. Some prominent liberal intellectuals advocate strongly for citizen welfare (e.g. Qin 2006). Although the left is inclined to ask for more state protection, some argue that people should sacrifice their (short-term) welfare for building a strong nation in Mao s words tightening the belts ( 勒紧裤腰带 ). 8

10 the validity of values such as democracy, rule of law, and civil liberties. In contrast, the appropriateness of democracy and individual liberties defines the ideological divide between conservatives and liberal in China. While we use the term conservatism to characterize the left, the term is not as commonly used as liberalism in Chinese public discourse. However, we believe the use of the term is appropriate because the left in China today aims to conserve the communist/socialist system, and especially the dominance of the state as embodied by the CCP. To augment these socialist and statist values, traditional values and symbols such as Confucianism and collectivism have been co-opted by the CCP in recent years in an effort to re-establish a sense of national identity. Furthermore, because Chinese public discourse associates liberal ideas and institutions primarily with the U.S., which has a historical legacy of antagonism with communist regimes, a rejection of the U.S., which is believed to threaten China s rise, has become a key component of the leftist, conservative narrative (Ren 2013). It is worth noting that the potential tension between pro-market values and pro-democratic values is not a main divide among liberals in China today. While democracy is connected to conceptions of equality, economic liberalization can result in severe inequality, but in China today, liberals are mostly concerned with putting constraints on the power of the state, and less with inequality. 17 In the rest of this paper, we will test (1) whether these observations about China s contemporary ideological orientation are accurate and supported by empirical data; (2) whether ideology in the political, economic, and social/culture spheres is strongly correlated or orthogonal; and (3) whether there is any indication that the evolution of ideology in China is influenced by economic forces and social changes. 3 Data The Chinese Political Compass ( 中国政治坐标系 ) is a website modeled on the UK platform The Political Compass ( that is tailored to the Chinese political context. It was set up by individuals affiliated with Peking University interested in measuring various dimensions of ideology in China. The website was launched in August, 2007, and in March of 2015, the website became blocked by the Great Firewall of China. In response, CPoC organizers decided to publicly release the data they have collected on the 17 Some liberal intellectuals have been advocating for a more complex liberalism, in which individual freedom and social equality are both taken seriously and basic democratic values are expressed with, and enhanced by, Chinese cultural elements. See, for example, The Explanatory Power of A Complex ism: A Conservation with Zhou Lian. 汪丁丁 : 复杂自由主义的解释力 与周濂对话. Available at 9

11 CPoC platform over the years. Currently, the data for 2014, including 171,830 respondents who completed the survey, has been released. We utilize this 2014 dataset for our analysis in this paper. Visitors to the CPoC are presented with 50 statements and allowed to select from a fourpoint scale ranging from strongly disagree, disagree, agree, to strongly disagree. 18 Respondents can only move to the next question after completing the previous question, and after all 50 questions are complete, the website shows the respondent s responses along political, economic, and social/cultural dimensions relative to all other participants. The CPoC only records a respondent s answers if responses to all 50 statements are obtained. In other words, if a visitor to the website stops answering at statement number 20, we do not have a record of this individual s first twenty responses. Respondents can also go back and change the answer to a survey item. Influenced by the intellectual strands described in Section 2, the CPoC survey measures attitudes along three dimensions political, economic, and social/cultural. 19 The statements in the survey were designed by its creators to cover the spectrum of opinions along these three dimension. Bearing close resemblance to the items in Table 1, the 20 items in the survey that pertain to politics include statements such as Western multiparty systems are unsuitable for China in its current state, 20 When events that have major repercussions for the safety and security of people occur, the government should freely disseminate information even if information disclosure increases the risks of unrest, 21 Force should be used to reunify Taiwan with China if conditions permit, 22 and Even if procedural rules are violated in the process of investigation and evidence gathering, those who have actually committed crimes should be punished. 23 Economic items includes statements such as Attempting to control real estate prices will undermine economic development, 24 Individuals should be able to own, buy and sell land, 25 Sectors related to national security and important to the national economy and 18 In Chinese, 强烈反对 反对 同意 强烈同意. 19 Appendix A.1 contains the full set of items, and we have ordered the items so the first 20 relate to politics, the second 20 to economics, and the last 10 to society and culture. 20 西方的多党制不适合中国国情, where agreement connotes conservatism (on the left) and disagreement connotes liberalism (on the right). 21 发生重大社会安全事件时, 即使认为信息公开会导致骚乱的风险, 政府仍应该开放信息传播 where agreement connotes liberalism (on the right) and disagreement connotes conservatism (on the left). 22 条件允许的话应该武力统一台湾 where agreement represents conservatism (on the left) and disagreement represents liberalism (on the right). 23 哪怕经历了违反程序规定的审讯和取证过程, 确实有罪的罪犯也应被处刑 where agreement represents conservatism (on the left) and disagreement represents liberalism (on the right). 24 试图控制房地产价格的行为会破坏经济发展 ) where agreement denotes liberals (on the right) and disagreement conservatives (on the left). 25 私人应当可以拥有和买卖土地 ) with agreement denoting liberalism and disagreement conservatism. 10

12 people s livelihoods must be controlled by state-owned enterprises, 26 People who make money through capital gains contribute less to the society than people make money through labor. 27. Social/cultural items include statements such as Traditional Chinese classics should be the basic education material for children, 28 Two adults should be free to engage in voluntary sexual behavior regardless of their marital status, 29 and Even with population pressures, the state and the society have no right to interfere in the decision to have a child, or how many children to have. 30 In addition to the 50 statements used to measure ideology, the CPoC asks respondents for their gender, year of birth, their level of education, and their annual income. 31 For level of education, respondents can select middle school and below, high school, college, and advanced degree. 32 For annual income, respondents can select less than 25 thousand RMB, 25 to 50 thousand RMB, 50 to 75 thousand RMB, 75 to 100 thousand RMB, 100 to 150 thousand RMB, 150 to 300 thousand RMB, and more than 300 thousand RMB. 33 CPoC also record the date and time the survey was completed, as well as the IP address associated with the respondent to the provincial and sometimes city level. Because of the setup of the online questionnaire, this dataset has several features. First, there are no missing data in responses. 34 Second, Don t know is not an option. Although the variables are ordinal (with four options), a PCA provides the best linear approximation of the probability function and the gain from estimated a nonlinear model is limited (Heckman and Snyder 1997). Third, the massive data render inference a relatively small issue but impose harder constraints on computational speed. All these features set the stage for using PCA, as opposed to an IRT model, in an attempt of finding the dominant ideological space. Descriptive Statistics. We identify the locations of the respondent by looking up their 26 那些关系到国家安全 以及其他重要国计民生的领域, 必须全部由国有企业掌控 ) with agreement denoting conservatism (on the left) and agreement denoting liberalism (on the right). 27 靠运作资金赚钱的人对社会的贡献比不上靠劳动赚钱的人 ) with agreement denoting conservatism (on the left) and disagreement liberalism (on the right) 28 应当将中国传统文化的经典作品作为儿童基础教育读物 ) where agreement connotes conservatism to the left, and disagreement connotes liberalism to the right. 29 两个成年人之间自愿的性行为是其自由, 无论其婚姻关系为何 ) where agreement denotes liberalism to the right and disagreement conservative to the left. 30 即使有人口压力, 国家和社会也无权干涉个人要不要孩子, 要几个孩子 ) where agreement denotes liberalism on the right and disagreement denotes conservatism on the left. 31 In Chinese, 您的性别 您的出生年份 您的学历 您的年收入. 32 It is worth noting that the education question is somewhat ambiguous: it is unclear whether the level of education refers to the highest obtained degree, or the degree that is being pursued. 33 The exact meaning of annual income is unclear, for instance, whether it includes stipend. Many college students in the sample reported that they earn an annual income beyond the lowest interval. 34 A tiny proportion of missing values are observed for the respondents socio-economic status. 11

13 Internet Protocol (IP) addresses. 35 The 171,830 respondents who completed the survey in 2014 are nearly all from mainland China (89 percent), with a few 1,765 (1 percent) from Hong Kong, none from Taiwan, and the remaining 16,830 (10 percent) from countries outside of China. As shown in Figure 1, the largest proportion of respondents (23 percent) come Figure 1. Number of Respondents in Each Province and Abroad from Beijing, followed by Guangdong (9 percent), Shanghai (8 percent) and Jiangsu (7 percent). 36 Nine provinces in western China 37 and the island province of Hainan have the fewest respondents, together representing less than 4 percent of the survey sample. spacial distribution of the sample is summarized in Table A2 in Appendix. Our 2014 CPoC sample is dominated by male college students. 64 percent of the respondents are male, and 67 percent are college students or those who have completed college. The left panel of Figure 2 shows that it is due to the preponderance of college students that nearly half of respondents report income below 25 thousand RMB per year. The right panel of Figure 2 shows the large number of young respondents, 45 percent are between the ages of 19 and 22, is also related to the abundance of college students in the sample. Respondents age represent 97 percent of the observations and forms the focus of this paper. For the summary statistics of respondent self-reported socio-economic status, see Table A3 in Appendix. 35 CPoC provides the first three segments of the IPv4 address of each respondent. The level of IPv4 information allows us to identify the province and in some region the prefecture or city of the respondent. 36 All shape files of the maps we use in this paper come from The National Geomatics Center of China: 37 Including Guizhou, Shaanxi, Yunnan, Gansu, Xinjiang, Inner Mongolia, Ningxia, Qinghai, and Tibet. The 12

14 Figure 2. Self-reported Education, Income Level, and Age of Respondents >College College High school <High school >College College High school <High school 0 25K 50 75K K >300K Self reported Annual Income Age Completion of the survey followed a bursty pattern. Figure 3 shows the number of respondents by day in The largest bursts of activity on CPoC occur in January, Figure 3. Responses over Time in Space Count of Respondents All locations Beijing Guangdong Shanghai Jan Mar May Jul Sep Nov Jan February, March, May, and December of Bursts of activity occur across localities. For example, in Figure 3, the largest burst of survey activity occurring in December 2014 included bursts of respondents from Beijing (shared in dark gray), Guangdong (in medium 13

15 gray), and in Shanghai (in light gray). 4 Methods The main objective of this paper is to locate and definite ideological spectrum of ordinary Chinese. In this section, we briefly discuss Principal Component Analysis (PCA) as a method of estimating latent policy preference. The method has been widely used in the literature and has the advantages of being fast, transparent, and easily interpretable (Ansolabehere, Snyder and Stewart 2001; Ansolabehere, Rodden and Snyder 2008; Carsey and Layman 2006; Heckman and Snyder 1997). The main challenge to our analysis, however, relates to the non-random nature of the sample. It is unlikely that respondents who completed the 50 question survey at CPoC are representative of the general population given similar gender, age, education, and income. For one, these respondents have access to computers and the Internet. For another, they are specific types of individual who have the interest and patience in completing online polls. Furthermore, the sample is dominated by college students, especially those from Beijing. However, first, the sampling problem may not be fatal for our purpose of uncovering the latent ideological space, as we will show in the next section. Second, we adopt a poststratification strategy to alleviate the problem of sample selection. The re-weighting schemes we use probably cannot solve the sampling problem in fact, given the distributions of the data, it seems unlikely that any methods can but by adopting this strategy, we want to show that our main results are not extremely sensitive to sample compositions. We first brief explain the mechanism and intuition of PCA. Then we move onto the discussion of post-stratification. 4.1 Principal Component Analysis Taking advantage of the large size of the sample, we utilize PCA to identify whether the 50 CPoC items correspond with ideological dimensions. PCA converts responses to the 50 CPoC items, which may be correlated, into linearly uncorrelated composite variables called principal components (PC). In other words, PCA transforms our correlated observed responses to the 50 items to a smaller set of important composite variables plus errors. These composite variables are orthogonal linear combinations of the original variables that maximize the total variance among original variables. 38 As such, these composite variables can be thought of as a latent trait that explains value a set of variables take on. Using 38 In practice, the constraint that the square of the weights equals 1 is added. 14

16 notation: P = X R (1) where X is an (N k) matrix containing the original data with N number of observations and k variables; R is a (k k) rotation matrix; and P is an (N k) rotated matrix that consists of k eigenvectors (which are the principal components) of X. These orthogonal vectors are sorted based on their corresponding eigenvalues, e.g., the eigenvector that has the largest eigenvalue is placed in the first column. Since in our case we have 50 items, k = 50. Equation (1) can be rewritten as: X = P R 1 which can be written as, x ij = ( p k ) P C ir λ rj + P C is λ sj. r=1 s=p+1 in which x ij is respondent i s choice on item j; P C ir is the rth principal component (or measure of ideology) for unit i and λ rj is the loading of item j on the rth principal component. Once p is fixed, the term in the parenthesis is regarded as errors. They are orthogonal to the PCs outside the parenthesis by construction. Intuitively, what PCA does is to find a lower-dimensional subspace, such as a line or a plane, in high-dimensional space (in our case, 50-dimensional space) such that the lowerdimensional space can capture the most of the variations in the observables. In our case, if we are able to summarize respondents answers to the 50 items in only a few dimensions, we label these dimensions as ideological spectrum and each respondent s linear projection on the corresponding subspace as ideology. Each PC s corresponding columns in the rotation matrix R capture each item contributions to the PCs. They describe intuitively what a particular dimension of ideology means. 4.2 Survey Reweighing (Post-Stratification) After conducting PCA on the full dataset, we address the issue of the non-random sampling and attempt to recover a more representative sample by reweighing the observations (and their associated PCs). We use individual level data from the 2005 Census to define our two target populations. This is what we call post-stratification survey reweighing. The first reweighted dataset focuses on youth in mainland China, who are individuals age based 15

17 on the United Nations definition, 39 and the second re-weighted dataset focuses on young adults and adults aged 16 to 40 in mainland China. We focus on these two sub-populations because of the volume of CPoC data clustered within these age ranges. In the rest of the paper, we refer to the original dataset as the Unweighted sample, the reweighted data focused on the yrs age group as Youth the sample, and reweighted data focused on the 16 to 40 yrs age group as the Adult sample. To obtain the Youth sample, we first calculate the probability of each province-gendercohorts cell in the population and assign weights to each of these cells in the sample to match these probabilities. 40. To obtain the Adult sample, we conduct a similar procedure, but this time create age bins of five years each (from 26 to 40) to avoid empty cells in some provinces and match on the probability of each province-gender-age/age bin cell of the population Figure 4. Age and Gender Distribution of Unweighted and Weighted Data Unweighted Women Men Age Youth Women Men Age Adult Women Men Age Figure 4 shows the age and gender distribution of the three datasets. The left panel shows the age distribution by gender for the Unweighted dataset. As expected, the sample is dominated by male respondents. In the center and right panels of Figure 4, which show the Youth and Adult weighted samples, respectively, the gender breakdown is much more balanced, and the age distribution is less dominated by college-aged individuals in large cities. In the rest of the paper, we conduct our analyses with all three datasets. 39 See 40 For a very small proportion of respondents (<2 percent), this IP data is not sufficient to differentiate between provincial locations. For this very small proportion of individuals, we repeat their responses in the dataset and assign an IP weight to each repeated the response, which corresponds to 1/N, where N is the total number of possible locations identified by the IP address. We did not match on education and income because the 2005 Census lack these data and they do not seem to be precisely measured in the CPoC sample as well. 16

18 5 China s Ideological Spectrum In this section we describe the results of PCAs using the unweighted, original data. We conduct three sets of analysis. First, we run a PCA using the entire dataset and conduct diagnostics on the results. Second, we conduct three separate PCAs using (1) only the political items (20 questions); (2) only the economic items (20 questions); and (3) only the social/cultural items (10 question). We then plot the each of the first PC from the three PCAs and test their colinearity. Third, we perform a demanding robustness check, in which we drop observations non-randomly across space and over time, and re-conduct a PCA using the remaining dataset; we then compare the first PC from the PCA using the remaining dataset and the first PC from the PCA using the entire dataset. Results from all three analyses suggest that there is one dominant ideological dimension, that citizen beliefs around political, economic, and societal/cultural ideas align along one dimension. In other words, individuals who respond conservatively to political statements mostly likely also respond conservatively to economic and social/cultural statements. Likewise, individuals who are liberal on political issues are also liberal on economic and social/cultural issues. We also find the dominant ideological dimension we recover is robust to significant changes of the sample composition and/or reweighting of the dataset. 5.1 PCA Results First, we conduct a PCA using the entire dataset. It reveals that there exist one strong and dominant dimension. The left panel of Figure 5 shows the variances (on the y-axis) associated with the first 10 principle component (x-axis). Variance in PCA refers to the multivariate or overall variability of responses to the CPoC statements. 41 We clearly see from the left panel of Figure 5 that the first principal component explain the bulk of the overall variability of our data (around 17 percent of total variation in X). The variance of the second PC is much smaller, explaining 5.4 percent of the variation in the data. The right panel of Figure 5 shows the distribution of the 8 principal components (the 8 eigenvectors in the rotated matrix P ). Again, this plot clearly shows that the first principal component (PC1) captures a much greater amount of overall variability than the second (PC2) and subsequent PCs Overall variability refers to the sum of the diagonal of the variance-covariance matrix of the 50 CPoC survey items, which is equivalent to the sum of eigenvalues of the 50 eigenvectors (or principal components). 42 It is still possible that including the second dimension improve prediction of the survey outcomes. Stimson (2012) suggests that in the United States, two dimensions are needed to capture enough variability in ideology although two dimensions are highly correlated. We plan to test the additional prediction power brought by the second dimension using a cross-validation method. Although there is a statistical possibility, since the 17

19 Figure 5. Number of Principal Components Variances PC1 PC (a) Variances of PCs (b) Distributions of PCs In Figure 6, we visualize the two-dimensional space spanned by the two PCs. The blue dots represent the values of the first and second PCs plotted against each other, and the gray arrows, each corresponding to a statement in the CPoC survey, are the first two columns of the rotation matrix R. The position of each blue dots shows the latent values of the respondent in both dimensions. If we define the first dimension as ideology, the x-coordinate of each dot will be the measure of ideology of that person. The gray arrow and the position of survey questions signify the contribution of each question to both dimensions. The position of the question label literally means the ideological position of a person (in the two-dimensional space) if s/he agrees with an average person in the population on all but one questions and is one unit more positive towards that particular statement than the average person. 43 Figure 6 not only shows that the first dimension has a much bigger variance than the second one; 44, but also gives meanings to the first PC, which is labelled as ideology. Intuitively, agreement with the subset of items on the right will pull a respondent to the right of ideological spectrum while agreeing items on the left will pull him/her to the left. Combining answers to all 50 questions, a person s ideological position (on the x-axis) is thus determined. We now examine the statements falling to the right and left of Figure 6 to substantively second dimension we uncover is not easily interpretable (Table A5 in Appendix), our main argument that the political, economic, and social/cultural dimensions of China s ideological stratum remains unchanged. 43 The length of the arrows is rescaled to facilitate visualization. 44 This point is further and more clearly illustrate in Figure A2 in Appendix. In the scatterplot matrix, the variances of PC2-PC5 are much smaller than that of PC1. 18

20 Figure 6. Survey Questions and the First Two Dimensions interpret the meaning of what we are calling ideology. The following are the five statements that have the largest positive values for first PC (they are the question furthest to the right of Figure 6). Numbers in the parentheses are the corresponding number in the rotation matrix. Q12 (0.21) It is acceptable besmirch the images of national leaders and founding leaders in literary and artistic works. Q3 (0.21) When events that have major repercussions for the safety and security of people occur, the government should freely disseminate information even if information disclosure increases the risks of unrest. Q6 (0.20) It is preferable to let universities recruit students by themselves than to have a unified national college entrance examination system. Q2 (0.20) Human rights take precedence over sovereignty. Q17 (0.19) Lawyers should do their utmost to defend clients even if the client has committed a crime. 19

21 As previously mentioned, the statements are ordered so that the first 20 pertain to politics, the second 20 to economics, and the last 10 to society and culture. What we see from the statements listed above is they are statements for which affirmative answers would clearly signal a liberal orientation, demonstrating support for liberal institutions and individual freedoms over the traditional and political hierarchies. Individuals who answer these questions affirmatively would be political liberals who do not hold up China s top leaders and founders as sacred cows (Q12), who show unconditional support for the freedom of information (Q3), who oppose the nationalization of educational standards (Q6), who hold human rights to be above sovereignty (Q2), and who support legal due process (Q17). The following are the five statements have the largest negative values for the first principal component (they are the question furthest to the left of Figure 6) and agreeing to them will pull a respondent to the left end of the spectrum. Q9 (22) National unity and territorial integrity are the highest interest of society. Q46 (21) The Eight Diagrams (Bagua) in The Book of Changes (Zhouyi) can explain many things well. Q20 (20) The state should take measures to train and support athletes so they can win glory for the country in various international competitions. Q33 (19) People who make money through capital gains contribute less to the society than people make money through labor Q48 (19) It is unnecessary to push forward the simplification of Chinese characters. What we see from this second set of statements is that they are statements for which affirmative answers would clearly signal a conservative orientation. Individuals who would answer affirmatively these questions falling furthest to the left of Figure 6 reveal conservatism to political, economic, and social values. These types of respondents are nationalistic (Q9 and Q20), they support traditional Confucian values or at least pay respect to a traditional wisdom or superstition (Q46), they are against a transformation of the Chinese language toward convenience (Q48), and believe that the value of labor trumps that of capital (Q33). These questions and many more of them shown in Appendix A.3 Table A2 give concrete meanings to the conservatism-liberalism ideology spectrum as we measure it. They are consistent with our prior knowledge that liberalism in the Chinese context are closely related to constitutional democracy, individual freedom, and human rights while conservatism is tied with nationalism, statism, and traditional Chinese culture as ordinary people understand it. 20

22 5.2 Dimensionality Next, we show the alignment of ideological spectrum along political, economical and social/cultural dimensions. We conduct three separate PCAs using questions covering each of the three dimensions (20, 20, and 10 questions each). We then plot the resulting first PCs there are three of them in a three-dimensional (3D) space. Figure 7 shows two views of the 3D plot. Each dot is colored from red (conservative) to blue (liberal) based on the main ideology measure we obtain from the PCA using a complete set of questions. It represents a respondent s ideological position in the 3D political-economic-cultural space. 45 Figure 7. Three Faces of Chinese In this plot, we can see a strong correlation between high positive values for cultural questions with high positive values for political questions and with high positive values for economic questions. This means that respondents who are politically liberal are likely to be economically and socially liberal; and vise versa. This result strongly confirms our previous finding that there is a dominant and consistent ideological spectrum among contemporary Chinese, and such spectrum is not particular to political, economic, or cultural issues. It is likely to represent belief systems that shapes a person s on a wide range of public policy issues. Going forward, we will refer to the first PC as ideology, where positive values refer to more liberal political, economic, and social/cultural views, and where negative values refer to more conservative political, economic, and social/cultural views. 45 Three two-dimensional plots are provided in Figure A1 in Appendix. 21

23 5.3 Robustness check Because the sample is not representative of a target population, we are concerned about the the stability of the ideological dimensions recovered. To assess its stability, we conduct a robustness check by dropping observations from the data, and assessing the correlation in the PCs obtained from a PCA using the full dataset and the PCS obtained from datasets with dropped observations for the remaining respondents in the datasets. 46 To push the experiment to the extreme, we first order the observation according to location, starting first with respondents outside of China, then respondents in HK, then respondents in each province ordered based on the number of CPoC respondents in that province. By doing so, the distribution of the discarded observations will be quite different from those remaining in the sample. Figure 8. Robustness of the Measure Correlation of PCs Abroad / HK PC1 PC2 Beijing Guangdong Shanghai Jiangsu Zhejiang Sichuan Hubei Shandong Tianjin 99.95% 98.41% Fujian Hunan Liaoning Hebei Shanxi Henan Anhui Correlation of PCs Jan PC1 PC2 Feb Mar Apr May Jun Jul Aug Sep Oct 99.96% 88.09% Nov Dev Proportion of Data Discarded (a) Across space Proportion of Data Discarded (b) Over time The left panel in Figure 8 is a plot of the correlation for the first and second PCs between the original PCA and the PCA of 171 datasets each with 1000 fewer observations. For example, at approximately 0.1 on the x-axis (where the plot is labeled Abroad/HK), the value of the black line represents the correlation in the first principal component between the original PCA and the PCA of a dataset missing 10 percent of the observations, namely missing all observations from outside of China and from Hong Kong. Similarly, at approximately 46 For example, we estimate PCs using our full sample of 171,829 observations. Then we drop 170,000 of the observations and use the remaining 1,829 observations to conduct another PCA. We can compare the correlation of PCs from the two PCAs for the 1,829 observations. 22

24 0.4 on the x-axis (where the plot is labelled Guangdong), the value of the black line is the correlation in the first principal component between the original PCA and the PCA of a dataset missing 40 percent of observation, specifically all observations from outside China, Hong Kong, Beijing, and Guangdong. The red line shows the correlation for the second PC. In the right panel, we conduct a similar experiment by dropping observations along the 365-day time line. For instance, we first drop observations from January, then February, then March, etc. What Figure 8 shows is that there is high agreement for the first PC even when very few observations remain and even when observations are no way to be representative of the full sample. For example, in the first experiment, the last 1 percent of the data (less than 2,000 observations), representing respondents from provinces with the fewest CPoC participants (e.g. Hainan, Ningxia, Qinghai, and Tibet). Agreement (in terms of correlation) for the first PC from is percent. In the second experiment, the last 1 percent of data contain observations in the last 7 days of The agreement for the first PC is percent while that for the second PC is below 90 percent. In other words, even when the sample is severely reduced in size, we still recovered almost precisely the same uni-dimensional ideological space. 5.4 Reweighting We reweight the data after conducting PCA in order to examine how ideology varies across space and among individuals with different characteristics. Figure 9 shows that reweighting Figure 9. Reweighted Distributions of Ideoglogy Unweighted Youth Adult Abroad Unweighted Youth Adult HK (a) Individual level (b) Provincial level 23

25 slightly alters the ideological distribution of the raw data. The left panel of Figure 9 shows the ideological distribution at the individual level, showing that the weighted Adult sample in green has a very similar ideology to the Unweighted sample, and showing that the Youth sample also followed a similar pattern but with slightly more ideological centrists. The right panel of Figure 9 shows the ideological distribution at the provincial level. Here, we can see that the Unweighted sample has a right tail of very liberal respondents from overseas and from Hong Kong. Since the weighted samples only contain respondents from mainland China, we do not see this right tail in the Youth and Adult samples. At the provincial level, we can see that the Youth sample is more conservative, and the Adult sample is slightly more liberal. The reweighted samples also have large variances because (1) some observations are discarded; and (2) the gender-cohort cells that have dense data in a province are now down-weighted. 6 Characteristics of China s Ideological Spectrum In this section, we move to our secondary goal of examining whether and how ideology varies across space and by individual characteristics. 47 We find that there is variation between provinces in terms of ideological orientation, and we identify a strong relationship at the provincial level between liberalism and common features associated with modernization the level of per capita income, the degree of trade openness, and the extent of urbanization. These provincial level results dovetail with ideological variation in individual level characteristics where we find striking results for education and income. Overall, we find a positive correlation between liberalism and the level of education, as well as a positive correlation between liberalism and self-reported income. Across all respondents, we identify an inverse U shape between liberalism and age; however, this relationship between ideology and age is conditioned by education and income such as those with lower education and lower income exhibit much more conservatism as they age relative to better educated and higher income individuals in their age group. These results remain qualitatively the same regardless of whether Unweighted, Youth, or Adult samples are used. 47 Here and in subsequent analyses, when we use the term ideology, we mean the first PC, and liberal refers to larger (positive) numeric values and conservative refers to smaller (negative) numeric values. 24

26 6.1 Regional Variation in To examine how ideological orientation varies by region, we rank 22 provinces, 4 autonomous regions, and 4 centrally administered municipalities 48 by the mean from most liberal to most conservative for the Unweighted, Youth, and Adult samples. We then average the ranks from the three datasets for each region, and re-rank the regions from most liberal to most conservative. Based on this ranking procedure, the 10 most liberal provinces are in blue in Figure 10, while the 10 most conservative are shaded in red and the 9 provinces in the middle Figure 10. Provincial Ideological Rank are in purple. In the ranking of average rank between the three datasets, we find that the 10 most liberal provinces are the same the 10 most liberal provinces from the Unweighted data, and we find that the 10 most conservative provinces include eight of the most conservative provinces from the Unweighted data (see Table A6 in Appendix A.5 for overall ranking and ranking by sample). Figure 10 shows that eastern coastal provinces, such as Shanghai, Jiangsu, Zhejiang, and Guangdong, as well as Beijing, Shanxi, and Hubei emerge as the most liberal. Hinterland provinces such as Xinjiang and Inner Mongolia as well as central provinces such as Henan and Hunan, and Jiangxi emerge as the most conservative. Provinces in the Northeast and in populous Western provinces such as Sichuan and Chongqing fall into the middle of the ideological spectrum. We find the distribution of ideology to be broadly consistent across the three samples, and consistent based both on province ranking and (weighted) mean value 48 We refer to all these regions as provinces for ease of communication. Tibet and Qinghai are not included in the ranking due to small numbers of respondents in those two provinces, although if we include them, they both are placed close to the bottom of the rank (being very conservative). 25

27 of the ideological distribution by province. In Appendix A.5 we present maps of based on ideological rank (Figure A6) and intensity (Figure A5) for each dataset. Economic and Demographic Variation. We explore the relationship between provinciallevel ideology (unweighted mean) and economic and demographic variables. 49 The left panel of Figure 11 plots the mean ideology for each province against the log income per capita for Figure 11. Correlates of Provincial Shanghai Guangdong Beijing Log Income per capita Trade openness Urbanization the province in This figure reveals a high level of correlation between ideology and average disposable income of urban residents. Provinces with higher levels of income per capita are more liberal while provinces with lower per capita incomes are more conservative. The center panel of Figure 11 shows ideology and trade openness, specifically, imports and exports as a proportion of GDP for each province. 50 While Shanghai, Guangdong, and Beijing with high levels of trade openness also have the most liberal ideological orientations, the positive relationship between openness and liberalism can also be seen among provinces with lower levels of trade openness. 51 The right panel of Figure 11 shows the percent of urban population on the x-axis, calculated as the percentage of permanent urban residents in total population for each province based on 2013 data, against the mean ideology of each province. Again, there is a strong positive correlation between the level of urbanization and liberalism. 49 Using weighted means yields almost the same results. 50 The calculation for the x-axis is (imports + exports) / GDP 100 by province based on the 2013 data. All provincial economic and demographic indicators are from China Statistical Yearbook (2014). 51 This result is consistent with Lan and Li (2015), who show that nationalism is negatively correlated with trade openness at the city level based on 2009 CPoC data. They measure nationalism using an additive scale approach by averaging responses to four questions they believe closely related to nationalism (Q8, Q9 Q15, and Q19). 26

28 6.2 and Individual-Level Characteristics We examine the relationship between ideology and individual-level characteristics gathered through the CPoC website, specifically gender, age, education, and self-reported annual income. Figure 12 shows how these four demographic variables correspond to ideology in Figure 12. and Demographics Unweighted Youth Adult Unweighted Youth Adult Female Gender Male Age in Unweighted Youth Adult Unweighted Youth Adult Less than High school High school College than College 0 25K 25 50K 50 75K K K K > 300K Highest Level of Education Completed Self reported Income the Unweighted, Youth, and Adult samples. The point estimate are the mean ideology, and the confidence intervals are 95 percent bootstrapped confidence intervals. In the top left panel, we see that the ideological orientation of men and women are very similar in all three samples. In the top right panel, we see an inverse U relationship between age and ideology in the Unweighted sample. Specifically, age cohorts between 14 to 40 show increasing levels of liberalism, while after 40 years, age cohorts show increasing levels of conservatism. We do 27

29 not see this pattern in the Youth and Adult samples since those sample represent truncated age cohorts; however, in the Youth sample, we see a small inverse U where individuals in college seem to exhibit a peak in liberalism. 52 In the bottom left panel of Figure 12, we see that individuals with higher levels of education are more liberal than their less well educated counterparts. This pattern persists across datasets. We see a similar positive correlation between liberalism and income as seen in the bottom right panel of Figure 12 across all three data sets. Unsurprisingly, we have fewer observations in the Youth and Adult for the highest levels of income, but despite the uncertainty of the estimate, the upward trend of income and liberalism is hard to deny. When we examine the interactions between these demographic variables, some interesting pattern emerge, especially with respect to age, income, and education. Figure 13 shows how Figure 13. Correlates of Education, Income and by Age than College College High School Unweighted sample Self reported Annual Income: 25 50k k >300k Unweighted sample Age Age the ideological orientations of respondents vary in different ways by age cohorts depending on the level of education and income. The left panel of Figure 13 plots the interaction of education and age for the Unweighted sample. The plot shows that there is an inverse U relationship between age and ideology for individuals at all levels of education; however, the increase in conservatism that occurs at 40 years and beyond is most dramatic for individuals with the least amount of education. 53 The right panel of Figure 13 shows the interaction of income and age. Here, for individuals with the lowest levels of self-reported income, those who are older are more conservative 52 Cantoni et al. (2015) study the causal effect of school curricula on youths ideology in China by exploiting a textbook reform that was rolled out between 2004 and This pattern can be seen more clearly in the Adult sample. See Figure A4 in Appendix A.4. 28

30 than those who are younger. For individuals with middling levels of income, older age cohorts exhibit similar ideological orientation to younger age cohorts. This could be due to a generational difference. And finally, for individuals at the highest level of income, those who are older on average are more liberal than those who are younger even though there is more variation in the ideological orientation of those who are older (mainly due to the small number of older respondents). In short, after the respondents pass the age of college education, a divergence in ideology appears: those who are better off and/or better educated tend to be more liberal; and those who are worse off and/or less educated are likely to become more conservative. We also examine the interactions between gender and age, gender and education, and gender and income; however, we do not observe interaction effects with gender. In other words, men and women are very similar in their ideological outlook across age, education, and income cohorts (see Table A3 in Appendix A.4). 7 Conclusion and Future Directions Our analysis provides evidence that ideology in China today is strongly uni-dimensional. Political conservatism goes hand in hand with economic and social conservatism, while political liberalism is linked to economic and social liberalism. While more research is needed to explore causes of this alignment, it may be the result of efforts by the CCP to craft a coherent ideological narrative to legitimize its rule. In recent years, the CCP has emphasized nationalism and national unity as well as Confucian values (as ordinary Chinese understand them). These political and social values are now associated with the statism and socialism of the single-party regime. The uni-dimensionality of our result is robust to a wide variety of diagnostics and checks. Thus, the ideological specturm we observe, as defined by the 50 items of the CPoC, would likely remain the same for a random, representative sample of individual respondents. Our analysis also reveals a strong, positive relationship between ideology and modernization. Provinces and centrally administered municipalities with higher levels of economic development, trade openness, and urbanization are more liberal. Of course, we do not know if this correlation is due to greater exposure to and interaction with liberal ideas, or because these provinces contain more individuals who have benefited from market liberalization, or some combination of factors. Our analysis of ideological differences between groups reveal that those with higher levels of self-report income and higher levels of education are more liberal across age cohorts than individuals who lower levels of income and lower levels of 29

31 education. One important caveat to this analysis on ideological variation is that the nonrepresentative nature of the CPoC sample could be driving (at least part of) these results. There are three main directions for follow-up research. The first pertains to methods. We plan to use a flexible IRT model with the same data to cross-validate our measure of ideology. We expect results from the IRT model to be extremely similar to that uncovered by the PCA presented here. The second direction relates to measuring ideology in a representative sample in order to validate the relationship between ideology and both provincial and individuallevel characteristics. Finally, as additional data becomes available, we plan to analyze over time trends in ideology, and examine the relationship between ideological orientation and real-world events. 30

32 References Ahmad, Ehtisham and Athar Hussain Social Security in China: A Historical Perspective. In Social Security in Developing Countries, ed. Ehtisham Ahmad, Jean Dreze, John Hills and Amartya Sen. London: Clarendon Press. Ansolabehere, Stephen, James M. Snyder and Charles Stewart Candidate Positioning in U.S. House Elections. American Journal of Political Science 45(1):136. Ansolabehere, Stephen, Jonathan Rodden and James M. Snyder The Strength of Issues: Using Multiple Measures to Gauge Preference Stability, Ideological Constraint, and Issue Voting. American Political Science Review 102(02). Barmé, Geremie Shades of Mao: The posthumous cult of the great leader. Vol. 28 ME Sharpe. Barnett, A Doak, Roy Hofheinz and Richard H Solomon Chinese communist politics in action. University of Washington Press. Brown, Kerry The Communist Party of China and. China: An International Journal 10(2): Cantoni, Davide, Yuyu Chen, David Y. Yang, Noam Yuchtman and Y. Jane Zhang Curriculum and. NBER Working Paper Carsey, Thomas M. and Geoffrey C. Layman Changing Sides or Changing Minds? Party Identification and Policy Preferences in the American Electorate. American Journal of Political Science 50(2): Caughey, D. and C. Warshaw Dynamic Estimation of Latent Opinion Using a Hierarchical Group-Level IRT Model. Political Analysis. Chen, Ziming The United Front of the Left and Right Under Constitutionalism [ 陈子明 : 宪政旗帜下的左右翼联合阵线 ]. html. Chen, Ziming Seven Agreements the Left and Right Can Reach [ 陈子明 : 左右两派可以达成的七点共识 ] html. Clinton, Joshua, Simon Jackman and Douglas Rivers The Statistical Analysis of Roll Call Data. American Political Science Review 98(02): Converse, Philip E The Nature of Belief Systems in Mass Publics. In and Discontent, ed. David E. Apter. New York: The Free Press of Glencoe. Eagleton, Terry : an introduction. Vol. 9 Cambridge Univ Press. 31

33 Friedman, Edward China 3(6): Democracy and Mao fever. The Journal of Contemporary Friedrich, Carl and ZK Brzezinski Totalitarian Dictatorship. Cambridge, MA: Harvard UP. Geertz, Clifford as a Cultural System. In and Discontent, ed. David E. Apter. New York: The Free Press of Glencoe. Geiger, Theodor Julius Die soziale Schichtung des deutschen Volkes: soziographischer Versuch auf statistischer Grundlage. Number 1 F. Enke. Heckman, James J. and James M. Jr. Snyder Linear Probability Models of the Demand for Attributes with an Empirical Application to Estimating the Preferences of Legislators. The RAND Journal of Economics 28:S142. Inkeles, Alex The totalitarian mystique: Some impressions of the dynamics of totalitarian society. Harvard University Press. Kitschelt, Herbert The transformation of European social democracy. Cambridge University Press. Lan, Xiaohuan and Ben Li The Economics of Nationalism. American Economic Journal: Economic Policy (forthcoming). Li, Zehou History of Modern Chinese Thought 李泽厚 : 中国现代思想史论. Shanghai: SDX Joint Publishing Company. Linz, Juan J Totalitarian and authoritarian regimes. Handbook of political science 3: Linz, Juan José Totalitarian and authoritarian regimes. Lynne Rienner Publishers. Marks, Gary, Liesbet Hooghe, Moira Nelson and Erica Edwards Party Competition and European Integration in the East and West Different Structure, Same Causality. Comparative Political Studies 39(2): Misra, Kalpana Curing the Sickness and Saving the Party: New-Maoism and New- Conservatism in the 1990s. Chinese political culture pp Neumann, Franz Notes on the Theory of Dictatorship. The Democratic and the Authoritarian State: Essays in Political and Legal Theory pp Pan, Wei Rule of Law and A Blind Faith in Democracy: China s Modernization and The World Order From a Legalist Perspective 潘维 : 法治与 民主迷信 : 一个法治主义者眼中的中国现代化和世界秩序. Hong Kong: Hong Kong Social Science Press. Perry, Elizabeth J Anyuan: mining China s revolutionary tradition. Vol. 24 Univ of California Press. 32

34 Poole, Keith T. and Howard Rosenthal Patterns of Congressional Voting. American Journal of Political Science 35(1):228. Poole, Keith T. and Howard Rosenthal Congress: A Political-economic History of Roll Call Voting. Oxford University Press. Qin, Hui Left and Right in Contemporary China [ 秦晖 : 当代中国的 左与右 ]. Ren, Jiantao To the Left or to the Right? A Value of Chinese Politics. The 21st Century (In Chinese) Dec Stimson, James A Tides of Consent: How Public Opinion Shapes American Politics. Cambridge University Press. Stimson, James A On the Meaning & Measurement of Mood. Daedalus 141(4): Swislocki, Mark Culinary nostalgia: regional food culture and the urban experience in Shanghai. Stanford University Press. Tausanovitch, Chris and Christopher Warshaw Measuring Constituent Policy Preferences in Congress, State Legislatures, and Cities. The Journal of Politics 75(02): Treier, S. and D. S. Hillygus The Nature of Political in the Contemporary Electorate. Public Opinion Quarterly 73(4): Wang, Dingding A Brief Remark on The Complexity of the Thoughts of Mr. Wang Yuan-hua [ 汪丁丁 : 简论王元化先生自由主义思想的复杂性 ]. http: // Wang, Hui Contemporary Chinese Thought and the Question of Modernity. Social Text (55):9 44. translated by Rebecca E. Karl. Wang, Sirui New Tendencies of the Left and the Right [ 王思睿 : 左右两派的新动向 ]. Whyte, Martin and William Parish Urban Life in Contemporary China. University of Chicago Press. Wu, Angela Xiao Ideological Polarization over a China-as-superpower Mindset: An Exploratory Charting of Belief Systems among Chinese Internet users, International Journal of Communication 8: Xu, Jilin. 2007a. The Spiritual Lives of Chinese in a Secular Society [ 许纪霖 : 世俗社会的中国人精神生活 ]. Xu, Youyu. 2007b. My Personal Experience of the Battle between ism and the New Left[ 徐友渔 : 我亲历的 自由主义一新左派 之争 ]. articles/sxwh/shsc/article_ html. 33

35 Xu, Youyu The Left and Right Battle in the Past Thirty Years in China [ 徐友渔 : 中国三十年左右之争 ]. boardid=2&read=1. Yao, Yang The Complexity of Problems of Contemporary China [ 姚洋 : 当代中国问题的复杂性 ]. 34

36 A Appendix A.1 CPoC Survey Questions The CPoC contains 50 questions. The first 20 pertain to politics. The second 20 pertain to economics, and the last 10 questions relate to social/cultural values. A-1

37 如果人民没有受过民主教育, 他们是不应该拥有普选权的 人权高于主权 发生重大社会安全事件时, 即使认为信息公开会导致骚乱的风险, 政府仍应该开放信息传播 西方的多党制不适合中国国情 在中国照搬西方式的言论自由会导致社会失序 由高校自主考试招生比全国统一考试招生更好 应该容许宗教人士在非宗教场所公开传教 无论中小学生或大学生, 都应参加由国家统一安排的军训 国家的统一和领土完整是社会的最高利益 哪怕经历了违反程序规定的审讯和取证过程, 确实有罪的罪犯也应被处刑 国家有义务进行对外援助 国家领导人及开国领袖的形象可以作为文艺作品的丑化对象 当法律未能充分制止罪恶行为时, 人民群众有权自发对罪恶行为进行制裁 应当允许媒体代表某一特定阶层或利益集团发言 如果国家综合实力许可, 那么中国有权为了维护自己的利益而采取任何行动 条件允许的话应该武力统一台湾 律师即使明知被辩护人的犯罪事实也应当尽力为其进行辩护 应该允许中国公民同时具有外国国籍 以美国为首的西方国家不可能真正容许中国崛起成为一流强国 国家应当采取措施培养和支持体育健儿在各种国际比赛场合为国争光

38 最低工资应由国家规定 中国改革开放以来的经济发展的成果都被一小群人占有了, 大多数人没得到什么好处 在重大工程项目的决策中, 个人利益应该为社会利益让路 浪费粮食也是个人的自由 如果猪肉价格过高, 政府应当干预 应当对国外同类产品征收高额关税来保护国内民族工业 教育应当尽可能公立 国有企业的利益属于国家利益 试图控制房地产价格的行为会破坏经济发展 改善低收入者生活的首要手段是国家给予财政补贴和扶持 有钱人理应获得更好的医疗服务 高收入者应该公开自己的经济来源 靠运作资金赚钱的人对社会的贡献比不上靠劳动赚钱的人 与其让国有企业亏损破产, 不如转卖给资本家 那些关系到国家安全 以及其他重要国计民生的领域, 必须全部由国有企业掌控 资本积累的过程总是伴随着对普通劳动人民利益的伤害 私人应当可以拥有和买卖土地 政府应当采用较高的粮食收购价格以增加农民收入 在华外国资本应享受和民族资本同样的待遇 市场竞争中自然形成的垄断地位是无害的

39 两个成年人之间自愿的性行为是其自由, 无论其婚姻关系为何 不应公开谈论自己长辈的缺点 现代中国社会需要儒家思想 判断艺术作品的价值的根本标准是看是不是受到人民大众喜爱 即使有人口压力, 国家和社会也无权干涉个人要不要孩子, 要几个孩子 周易八卦能够有效的解释很多事情 中国传统医学对人体健康的观念比现代主流医学更高明 汉字无需人为推行简化 应当将中国传统文化的经典作品作为儿童基础教育读物 如果是出于自愿, 我会认可我的孩子和同性结成伴侣关系

40 A.2 Summary Statistics Table A2. Spacial Distribution of the Sample Cumulative Location Counts Percent Percent Beijing 40, Abroad 16, Guangdong 15, Shanghai 14, Jiangsu 11, Zhejiang 8, Sichuan 6, Hubei 6, Shandong 5, Tianjin 4, Fujian 4, Hunan 3, Liaoning 3, Hebei 3, Shanxi 3, Henan 3, Anhui 3, Guangxi 2, Chongqing 2, Heilongjiang 2, Jilin 2, Jiangxi 1, Hongkong 1, Guizhou 1, Shaanxi 1, Yunnan 1, Gansu Xinjiang Inner Mongolia Hainan Ningxia Qinghai Tibet Note: Locations are based on IP addresses. A-5

41 Table A3. Descriptive Statistics of Respondent Socio-Economic Status Variable Obs. Mean SD Min Max Female 171, Age 171, Age: , Age: , Age: , Age: , Age: >50 171, Education: less than high school 170, Education: high school 170, Education: college 170, Education: more than college 170, Annual income: , Annual income: , Annual income: , Annual income: , Annual income: , Annual income: , Annual income: > , Notes: Numbers of annual income are in 1,000RMB. Age is capped at 80. Reported age beyond 80 is recoded as 80. A-6

42 A.3 Principal Components Table A2 shows the importance of CPoC questions to the first principal component, which we refer to as ideology. The question are sorted in order of important to the principal component, from most important to least. If the sign of the question is positive, that means respondents who answered this question in the affirmative are more liberal (to the right). If the sign of the question is negative, that means respondents who answered this question in the affirmative are more conservative (to the left). Figure A1. Correlates for the First Principal Component () from Three Seperate PCAs A-7

43 国家的统一和领土完整是社会的最高利益 - 国家领导人及开国领袖的形象可以作为文艺作品的丑化对象 发生重大社会安全事件时, 即使认为信息公开会导致骚乱的风险, 政府仍应该开放信息传播 周易八卦能够有效的解释很多事情 - 由高校自主考试招生比全国统一考试招生更好 人权高于主权 国家应当采取措施培养和支持体育健儿在各种国际比赛场合为国争光 - 靠运作资金赚钱的人对社会的贡献比不上靠劳动赚钱的人 - 汉字无需人为推行简化 - 无论中小学生或大学生, 都应参加由国家统一安排的军训 - 律师即使明知被辩护人的犯罪事实也应当尽力为其进行辩护 条件允许的话应该武力统一台湾 - 判断艺术作品的价值的根本标准是看是不是受到人民大众喜爱 那些关系到国家安全 以及其他重要国计民生的领域, 必须全部由国有企业掌控 - 应该允许中国公民同时具有外国国籍 资本积累的过程总是伴随着对普通劳动人民利益的伤害 - 如果猪肉价格过高, 政府应当干预 - 与其让国有企业亏损破产, 不如转卖给资本家 应该容许宗教人士在非宗教场所公开传教 私人应当可以拥有和买卖土地 国家有义务进行对外援助 - 在华外国资本应享受和民族资本同样的待遇 教育应当尽可能公立 - 如果国家综合实力许可, 那么中国有权为了维护自己的利益而采取任何行动 - 如果是出于自愿, 我会认可我的孩子和同性结成伴侣关系 试图控制房地产价格的行为会破坏经济发展

44 最低工资应由国家规定 - 中国传统医学对人体健康的观念比现代主流医学更高明 - 中国改革开放以来的经济发展的成果都被一小群人占有了, 大多数人没得到什么好处 - 政府应当采用较高的粮食收购价格以增加农民收入 - 西方的多党制不适合中国国情 - 现代中国社会需要儒家思想 - 以美国为首的西方国家不可能真正容许中国崛起成为一流强国 - 改善低收入者生活的首要手段是国家给予财政补贴和扶持 - 有钱人理应获得更好的医疗服务 即使有人口压力, 国家和社会也无权干涉个人要不要孩子, 要几个孩子 两个成年人之间自愿的性行为是其自由, 无论其婚姻关系为何 浪费粮食也是个人的自由 - 在重大工程项目的决策中, 个人利益应该为社会利益让路 - 在中国照搬西方式的言论自由会导致社会失序 - 国有企业的利益属于国家利益 应当将中国传统文化的经典作品作为儿童基础教育读物 - 不应公开谈论自己长辈的缺点 - 如果人民没有受过民主教育, 他们是不应该拥有普选权的 - 应当允许媒体代表某一特定阶层或利益集团发言 应当对国外同类产品征收高额关税来保护国内民族工业 - 当法律未能充分制止罪恶行为时, 人民群众有权自发对罪恶行为进行制裁 - 市场竞争中自然形成的垄断地位是无害的 哪怕经历了违反程序规定的审讯和取证过程, 确实有罪的罪犯也应被处刑 - 高收入者应该公开自己的经济来源

45 高收入者应该公开自己的经济来源 + 不应公开谈论自己长辈的缺点 + 市场竞争中自然形成的垄断地位是无害的 - 当法律未能充分制止罪恶行为时, 人民群众有权自发对罪恶行为进行制裁 + 西方的多党制不适合中国国情 - 应当将中国传统文化的经典作品作为儿童基础教育读物 + 两个成年人之间自愿的性行为是其自由, 无论其婚姻关系为何 - 教育应当尽可能公立 + 国有企业的利益属于国家利益 + 如果猪肉价格过高, 政府应当干预 - 现代中国社会需要儒家思想 + + 发生重大社会安全事件时, 即使认为信息公开会导致骚乱的风险, 政府仍应该开放信息传播 人权高于主权 + 应当对国外同类产品征收高额关税来保护国内民族工业 + 在华外国资本应享受和民族资本同样的待遇 - 中国传统医学对人体健康的观念比现代主流医学更高明 + 浪费粮食也是个人的自由 + 如果人民没有受过民主教育, 他们是不应该拥有普选权的 - 由高校自主考试招生比全国统一考试招生更好 + 汉字无需人为推行简化 - 应该容许宗教人士在非宗教场所公开传教 + 以美国为首的西方国家不可能真正容许中国崛起成为一流强国 + 那些关系到国家安全 以及其他重要国计民生的领域, 必须全部由国有企业掌控 + 在中国照搬西方式的言论自由会导致社会失序 - 试图控制房地产价格的行为会破坏经济发展 - 国家有义务进行对外援助 +

46 如果国家综合实力许可, 那么中国有权为了维护自己的利益而采取任何行动 + 与其让国有企业亏损破产, 不如转卖给资本家 - 条件允许的话应该武力统一台湾 - 国家领导人及开国领袖的形象可以作为文艺作品的丑化对象 + 判断艺术作品的价值的根本标准是看是不是受到人民大众喜爱 + 资本积累的过程总是伴随着对普通劳动人民利益的伤害 + 如果是出于自愿, 我会认可我的孩子和同性结成伴侣关系 + 周易八卦能够有效的解释很多事情 - 改善低收入者生活的首要手段是国家给予财政补贴和扶持 + 在重大工程项目的决策中, 个人利益应该为社会利益让路 + 靠运作资金赚钱的人对社会的贡献比不上靠劳动赚钱的人 - 最低工资应由国家规定 - 律师即使明知被辩护人的犯罪事实也应当尽力为其进行辩护 - 有钱人理应获得更好的医疗服务 + 国家的统一和领土完整是社会的最高利益 + 中国改革开放以来的经济发展的成果都被一小群人占有了, 大多数人没得到什么好处 + 哪怕经历了违反程序规定的审讯和取证过程, 确实有罪的罪犯也应被处刑 - 无论中小学生或大学生, 都应参加由国家统一安排的军训 + 应该允许中国公民同时具有外国国籍 - 政府应当采用较高的粮食收购价格以增加农民收入 - 国家应当采取措施培养和支持体育健儿在各种国际比赛场合为国争光 - 私人应当可以拥有和买卖土地 + 应当允许媒体代表某一特定阶层或利益集团发言 - 即使有人口压力, 国家和社会也无权干涉个人要不要孩子, 要几个孩子 +

47 Figure A2. Scatterplot Matrix of Five PCs A-12

48 A.4 Demographics Figure A3. Demographics and Ideoglogy Women Men Unweighted sample Women Men Unweighted sample Women Men Age Education Income Figure A4. Interactions between Income and Age for Youth and Adult Samples Youth sample Self reported Annual Income: 25 50k k >300k Self reported Annual Income: 25 50k k >300k Adult sample Age Age A-13

49 A.5 Provincial Analysis Table A6. Provincial Rank in Rank Province Unweighted Youth Adult Overall (1) (2) (3) (4) Shanghai Guangdong Zhejiang Beijing Jiangsu Fujian Hainan Shanxi Hubei Liaoning Tianjin Chongqing Sichuan Jilin Gansu Yunnan Shandong Shaanxi Heilongjiang Hebei Inner Mongolia Anhui Hunan Jiangxi Henan Ningxia Guangxi Guizhou Xinjiang Note: This Table shows the ranks of liberal ideology in 29 provinces. Qinghai and Tibet are not included because of too few observations. The overall rank shown in column (4) is the rank of the sum of ranks in the previous three columns. A-14

50 Figure A5. Provincial-level Ideological Intensity (a) Unweighted (b) Youth: (c) Adults: A-15