This chapter introduces six questionnaire items that I use throughout the book to measure the cultural norms that I call orientation toward authority (OTA) and definition of self-interest (DSI). The chapter presents statistical tests that seek to establish three points: (1) that these items are valid measures of the cultural norms they are designed to measure; (2) that the two norms are distinct from one another and not aspects of a single, more general norm; and (3) that culture – understood as people’s commitment to these and similar norms – changes more slowly than do social structure and political institutions. On this latter point hangs the argument that culture is an independent force in the regulation of behavior, not a mediating variable that simply transmits the effects of social and institutional change from the environment to the individual’s behavior. To begin, however, the chapter justifies the choice of mainland China and Taiwan as settings for the analysis.

Why Mainland China and Taiwan?

A major challenge facing cultural studies is that culture is intertwined with social structure and political institutions. We can show that culture (defined as norms) influences attitudes and behavior, but how do we know that culture does not itself simply reflect the impact of society’s structural resources and political institutions, instead of exerting an independent causal effect? The ideal way to separate the effects of culture from those of social structure and institutions is to locate research sites with the same cultural traditions but different levels of economic development (and consequently different social structures) and different kinds of political institutions. Mainland China and Taiwan, both under the influence of Confucian culture but separated for more than a century, present an excellent pair of cases through which to examine the relationships among social structure, political institutions, and culture.

Socioeconomic development varies widely between and within the two societies. At the time of the first survey in 1993, economic development in Taiwan was at least thirty years ahead of that in mainland China. At the time of the second survey in 2002, despite the changes that economic reform had brought to China, per capita GDP adjusted for purchasing power was only $2,881; by the time of the third survey in 2008 it was $5,302. The figures in Taiwan, by contrast, had reached $21,263 by 2002 and $31,891 by 2008.¹ Although economic development had minimized the urban – rural economic disparity in Taiwan, it was still a major problem on the mainland.

There were wide economic differences among regions in the mainland as well, which provided another variable for testing the cultural theory. The average level of education in Taiwan was also much higher than that of mainland China. If a person’s normative commitments are shaped by the social and economic environment of his or her adolescence, as some cultural theories posit, the normative orientations of our mainland and Taiwan survey respondents would have diverged from one another well before our first survey (Eckstein Reference Eckstein, Greenstein and Polsby1975).

Political institutions in the two societies were likewise fundamentally different. Mainland China at the time of our surveys was under the rule of an authoritarian government. People enjoyed few civil liberties or political rights, and the media were under tight government control. The only elections were those of rural village committee members, and these elections were only semi-competitive. People in Taiwan, by contrast, enjoyed a broad range of civil liberties and political rights. In 1994, Taiwan held its first provincial election for governor; in 1996, the first general election was held for the presidency of the Republic of China (the name of the government in Taiwan); and finally in 2000 the opposition party won a presidential election, leading to a peaceful change of government. The fundamental differences in the political institutions in these two societies – and especially the institutional transformation that occurred in Taiwan between the first and the second surveys – provide ideal variation for examining the impacts of institutions on culture.

Moreover, mainland China’s political history allows us to test the impact of institutional change on culture by comparing cultural orientations across age groups. In order to remake the Chinese people into “new socialist men,” Mao Zedong worked for nearly the first thirty years of the PRC to eliminate the influence of Confucian culture. The Thought Reform Campaign (1953–56) marked the beginning of government efforts to transform the minds of the public systematically. This was followed by significant institutional and structural transformations carried out through such programs as the People’s Communes and the Great Leap Forward of the late 1950s. In 1966, Mao called for Red Guards to clean up the so-called “Four Olds” (old customs, old culture, old habits, old ideas) and encouraged people to rebel against the ruling Communist Party establishment. The campaign to criticize Lin Biao and Confucius in the early 1970s exemplified Mao’s radical efforts to move Chinese society away from traditional culture. If culture is shaped by political institutions, a dramatic shift in cultural orientation should be found among those who came of age between 1966 and 1976 in China, the Cultural Revolution generation. If we find no significant cultural differences associated with the political experiences of various age groups, we may conclude that regime mobilization does not alter a society’s political culture.²

The Data

The data used for this analysis of political culture come from the five surveys that were described in the Introduction. The first two (one each from mainland China and Taiwan) were part of the 1993 project “Political Culture and Political Participation in Mainland China, Taiwan, and Hong Kong.”³ The second two (again, one each from mainland China and Taiwan) were part of the 2002 first-wave Asian Barometer Survey (ABS I). Because the ABS grew out of the 1993 study, the two sets of surveys share many questions in common.

The fifth source of data is a 2008 mainland China survey that formed part of the second-wave Asian Barometer Survey (ABS II). Including 2008 data in the analysis stretches the time interval of cultural change that we can study in mainland China from nine to fifteen years. However, some changes in the questionnaire make it impossible to use this data set for a full comparison with the first four data sets. Five of the six questions in the 2008 survey designed to measure culture are identical to those used in the previous surveys, but the wording of the sixth question – one of three used to measure orientation toward authority – is somewhat different, which makes it impossible to construct an OTA scale that is fully comparable to the one used in the first four datasets. Furthermore, some dependent variables examined in the second part of this book are missing from the 2008 survey. The 2008 data therefore are used here in limited ways: in this chapter to illustrate the extent of continuity and change in people’s answers to questions about normative beliefs, and in Chapter 4 to test for the impact of structural and institutional changes on DSI. It was not possible to use the 2008 Taiwan data because four of the six questions used to measure OTA and DSI were not asked in the 2008 Taiwan survey.

Reliability is a problem for survey research in all settings, but it may seem more serious in the authoritarian setting of mainland China. I used five methods to ensure that we got honest answers in China. First, I used retired high school teachers as interviewers. Teachers in China retire at age 55, and most of them are healthy and available to take part-time jobs. Their experience working with students makes them willing to listen to different opinions, and they tend to be more trusted than people from other professions. Their previous teaching experience equips them with good interpersonal communication skills. I found that respondents tended to trust schoolteachers and were more willing to talk frankly to them than to other fieldworkers.⁴ I encouraged our interviewers to tell respondents that they were retired schoolteachers. For further details about how the interviewers were trained, see Appendix A. Second, I placed questions on socioeconomic status at the end of the questionnaire so as to increase respondents’ sense of anonymity during the bulk of the questioning. Third, the questionnaire refrained from asking highly sensitive questions, especially questions on people’s perceptions of national leaders, so as to avoid creating unnecessary anxiety for respondents as well as to avoid interference in the conduct of the survey from the government. Fourth, we asked interviewers to assure respondents of confidentiality and anonymity before the interviews and to insist on conducting interviews one-on-one at the respondents’ homes. Fifth, certain internal checks were written into the questionnaires so as to identify people who answered inconsistently. Appendix A gives further details about the survey design and implementation.

There are various indications that the answers we received were reliable. First, the questionnaire included an item for the interviewer to fill out, reporting whether he or she believed that the respondent had answered honestly, and most interviewers said yes. Second, the questionnaire included several items asking people to evaluate the political atmosphere in China. An overwhelming majority of respondents said that China was much freer than it was previously in terms of civil liberty and political rights, a perception that would encourage honesty in the respondents (Shi and Lou Reference Shi and Lou2010b). Although respondents’ subjective evaluation does not necessarily reflect the reality of Chinese society, people would more likely be deterred from telling the truth by their own evaluation of political reality than by objective assessments from experts. To the extent that respondents think the political atmosphere in China is relatively free, they are likely to answer survey questions frankly.

Third, we have evidence that political fear did not prevent our respondents from giving candid answers. To monitor our interviewers’ behavior, before dispatching them to the field we sent a letter with a self-addressed envelope to prospective respondents, notifying them of the survey and asking them to evaluate the performance of our interviewers afterward. In 1993, we received more than 1,500 letters back from our respondents. Among them, more than 1,000 included an additional letter reporting political, social, or economic problems in their localities, either addressed to us or asking us to transfer the letter to a higher authority. If political fear had been a major issue in China, few respondents would have sent such letters to our office, as they would have feared that the authorities would be able to trace the letter back to them.

Finally, two questions were built into the surveys to ask respondents whether they believed political fear was a problem in China. Although some said that it was, the analyses in Appendix B indicate that political fear did not influence the way respondents answered the questions used to measure cultural orientations or those measuring political attitudes and behaviors. Other survey researchers based in the West have reached similar conclusions on the reliability of answers from respondents in China (Chen Reference Chen2005; Manion Reference Manion1994; Tang Reference Tang2005). In light of all these positive indications, I believe that the survey answers from China can be trusted.

Measuring Culture: Orientation toward Authority and Definition of Self-Interest

To find out what norms people subscribe to, it is less effective to ask them to agree or disagree with abstract normative statements (since most respondents will agree with most socially appropriate abstractions) and more effective to pose a series of concrete decision situations. This allows researchers to identify the general principles underlying the choices respondents say they would approve. From a methodological point of view, we treat the norm as a “latent construct” that is measured through “observable variables.” But which observable variables should be used, and how do we define the relationship between them and the latent constructs?

In the 1993 and 2002 surveys, three questions served as the observable variables to gauge people’s orientations toward authority. Respondents were asked whether they strongly agreed, agreed, disagreed, or strongly disagreed with the following statements:

1. If a conflict should occur, we should ask senior people to uphold justice. (OTA 1)

2. Even if parents’ demands are unreasonable, children should still do what they ask. (OTA 2)

3. When a mother-in-law and a daughter-in-law come into conflict, even if the mother-in-law is in the wrong, the husband should still persuade his wife to obey his mother. (OTA 3)

These items were designed with several considerations in mind. First, political referents were avoided, so that later in the analysis we could assess how general attitudes toward authority influence attitudes toward politics in particular. A finding that normative orientations toward authority in general have influence on people’s interactions with government would support the hypothesis that culture affects people’s political attitudes and behavior.

Second, the questions were designed to steer respondents to answer clearly in terms of norms rather than other motives that might influence action. For example, an individual may obey his parents because he loves them, wants their money, or fears the consequences of disobedience. We were interested in the logic of appropriateness in the minds of the respondents. To make sure they understood this, we used “should” in the wording of the questions and instructed our interviewers to emphasize it when reading the survey questions in the interviews.

Third, the questions were framed to require respondents to choose only one of two possible norms governing the given social relationship. This is because an individual cannot simultaneously subscribe to two norms that govern the same social relationship, as that would provide him or her with contradictory guidance.

Three questions were similarly used to measure definition of self-interest. Respondents were asked whether they strongly agreed, agreed, disagreed, or strongly disagreed with the following statements:

1. A person should not insist on his/her own opinion if people around him/her disagree. (DSI 1)

2. Various interest groups competing in a locale will damage everyone’s interests. (DSI 2)

3. The state is like a big machine and the individual is only a small cog and thus should have no independent status. (DSI 3)

These questions refer to sequentially wider collective groups to assess the size of the circle of others with which the individual identifies. As defined earlier, whereas the allocentric definition of self-interest requires actors to ground their identity in certain groups, the idiocentric definition empowers them to ground their identities in the self. An allocentric definition does not prevent people from pursuing their own interests, but it guides them to see their interests as overlapping rather than conflicting with the collective interest. With this distinction in mind, we designed the questions to ask respondents whether they prioritized the isolated self or the socially embedded self in their interest calculations. Rather than asking respondents directly whether they embedded their private interests within those of a group, we asked whether they were prepared to sacrifice their personal interests to those of the group. I designed the questions in this way in order to avoid socially acquiescent easy answers. In societies where the norm of allocentric DSI dominates, I expected respondents to give the easy, socially compliant answer if the question allowed them to do so; by forcing a harder choice I wanted to push them to reveal their real beliefs.

Table 3.1displays the frequency of respondents’ answers to these questions in different surveys. Because we are interested in cultural change – throughout the book, we want to explore how many people abandon the traditional norms of hierarchical OTA and allocentric DSI and why they do so – our entries here, as in other tables, display the percentages of people who hold attitudes consistent with reciprocal OTA andidiocentric DSI. These people disagreed with the conventional Confucian norms presented in the questionnaire.⁵ The numbers in the parentheses are the Ns. For example, when answering OTA1 in 1993, 547 respondents, or 16.6 percent of the population, said that they disagreed with the statement.

Table 3.1. Reciprocal Orientation toward Authority and Idiocentric Definition of Self-Interest in Mainland China and Taiwan

^a Change in percentage from 1993 to the year given.

Note: Entries are the percentages of respondents giving negative answers to the question.

Source: 1993 Survey of Political Culture and Political Participation in Mainland China, Taiwan, and Hong Kong; 2002 China and Taiwan surveys, Asian Barometer Survey I; 2008 China Survey, Asian Barometer Survey II (China 1993 N = 3,287; 2002 N = 3,183; 2008 N = 5,098/Taiwan 1993 N = 1,402; 2002 N = 1,415).

The table shows that the distributions of normative beliefs in mainland China and Taiwan are similar. Despite fundamental differences in social structures and political institutions, minorities in both societies had abandoned Confucian cultural norms. Although higher percentages of respondents in Taiwan than in China subscribed to reciprocal OTA and idiocentric DSI, the differences between the two societies were small.

Comparing the data over time, the distribution of responses in each society changed little from 1993 to 2002. Contrary to convergence theory, which argues that socioeconomic development and institutional change make people adopt norms that are more modern, our data indicate a mixed direction of change in both norms, with responses to different questionnaire items increasing or decreasing by small amounts from one survey to the next. The data from Taiwan were similarly mixed. The largest change in China was a drop of 17 percent in the number of people who disagreed that children should do what unreasonable parents ask. In Taiwan, the largest change was a decline of 12.7 percent in the number of respondents who disagreed with the norm that a person should yield to the opinion of his or her neighbors.

Comparison with mainland China data from 2008 provides further evidence that structural and institutional changes in China had only limited impacts on people’s normative orientations. Although people’s answers changed from survey to survey, the pattern of responses swings around a center. This suggests that random rather than nonrandom errors were at least partially responsible for the fluctuations we observe, whereas underlying attitudes remained essentially stable.

Creating Indexes of Orientation toward Authority and Definition of Self-Interest

To reduce the impact of such random fluctuations on our ability to measure real changes in norms, I use the six questionnaire items to create two multi-item indexes, one for each of the two norms. To do so, I use confirmatory factor analysis based on item response theory (IRT), also known as latent trait theory.⁶ I show, first, that OTA and DSI are indeed separate norms – that is, in factor analysis terms, that they constitute separate dimensions. This is important because I will show later that each of these norms has distinct implications for political behavior and that each changes differently in response to changes in the social and political environment. Second, I show that each of the three questionnaire items designed to measure each norm does in fact measure it – that is, that the standardized coefficient indicating the strength of each variable’s association with the latent construct is statistically significant. Third, I show that the factor structure is valid across my four datasets (China 1993 and 2002 and Taiwan 1993 and 2002). Finally, in Chapter 4, I combine the three questionnaire items that belong to each factor to build a score for each respondent on each of the two cultural variables.

For the first step, I conducted a confirmatory factor analysis using the 1993 mainland China data. I imposed a two-factor structure on the data and assigned the three questions about orientation toward authority (OTA1 through OTA3) to the first factor and the three questions about definition of self-interest (DSI1 through DSI3) to the second factor. The results are presented in Table 3.2.

Table 3.2. Confirmatory IRT Model for 1993 Mainland China Data

^a All standardized coefficients are significant at the .01 level.

^b Difficulty is the measure of how unlikely it is that respondents will disagree with a questionnaire item.

Note: Mplus is used for the estimation. For binary variables, the result from Mplus is identical to the results from other IRT estimates.

Source: 1993 Survey of Political Culture and Political Participation in Mainland China, Taiwan, and Hong Kong (N = 3,287).

Three statistics displayed at the bottom of the table confirm that the two-factor model provides a good fit with the data.⁷ The first two fit statistics compare the given model with an alternative model. Confirmatory Fit Index (CFI) is the comparative fit index, which varies from 0 to 1.⁸ A CFI close to 1 indicates a very good fit. In our case, CFI = .977 which indicates that the model fits very well. The second fit statistic is the TLI (Tucker-Lewis Index). Similar to CFI, a TLI close to 1 indicates a good fit. In our case, TLI =.966.

Other things being equal, more complex models always generate a better fit than less complex models. Another goodness-of-fit test, Root Mean Square Error of Approximation (RMSEA), incorporates the discrepancy function criterion (comparing observed and predicted covariance matrices) and the parsimony criterion. If RMSEA is less than or equal to .05, the model fit is generally considered good. If RMSEA is less than or equal to .08, the model fit is considered adequate. In our case, RMSEA is .041, which tells us that the model fits extremely well. Altogether, the statistical tests of the model confirm my hypothesis on the structure of people’s normative orientation in mainland China: there are indeed two dimensions in people’s normative orientations, and OTA can be clearly distinguished from DSI.

Second, we see that each of the standardized coefficients in Table 3.2 is statistically significantly correlated with the underlying construct at the .01 level. In IRT analysis, in contrast to classical measurement theory (CMT), this measure of statistical significance, rather than the size of the standardized coefficient, assures us that the observable variable has been assigned to the correct latent variable. The size of the standardized coefficient, meanwhile, tells researchers how strongly a variable correlates with the latent constructs and how much it should contribute to the IRT score compared with other variables. (See Appendix C for a more detailed discussion of the benefits of IRT analysis versus CMT for studying culture.) In Table 3.2, all coefficients are statistically significant, which indicates that all of them contribute something to the score. The “difficulty” of each item is a measure of how unlikely it is that respondents will disagree with the question. In IRT, a higher difficulty score means that the item is weighted more heavily in the combined scale.

As a third step in building indexes of the two cultural norms, I show in Table 3.3 that the factor structure identified in the 1993 China data set also appears in the other three data sets. It is important in particular to establish whether people in Taiwan subscribe to the same cultural norms as people in China. Some people believe that the century-plus-long separation of Taiwan from mainland China fundamentally transformed political culture there – that although people in Taiwan were influenced by Confucianism in the past, economic development, and more importantly, democratic transition and consolidation brought about fundamental cultural changes. They argue that whereas political culture in mainland China remains authoritarian, political culture in Taiwan has become democratic.⁹ However, the claim that political culture in Taiwan has changed has either been deduced from the fact that it successfully accomplished democratic transition or imputed on the basis of changes in Taiwanese people’s political attitudes rather than from the direct study of their normative orientations.

Table 3.3. Confirmatory IRT Model for Combined Mainland China and Taiwan Data

^a All standardized coefficients are significant at the .01 level.

^b Difficulty is the measure of how unlikely it is that respondents will disagree with a questionnaire item.

Note: Mplus is used for the estimates. For binary variables, the results from Mplus are identical to the results from other IRT estimates.

Sources: 1993 Survey of Political Culture and Political Participation in Mainland China, Taiwan, and Hong Kong; 2002 China and Taiwan surveys, Asian Barometer Survey I (China 1993 N = 3,287; 2002 N = 3,183/ Taiwan 1993 N = 1,402; 2002 N = 1,415).

The impacts of structural and institutional changes on culture, however, are more complicated than current political culture theory suggests. In Chapter 4, I develop a new theory of cultural change. Here, I examine whether the variables designed to measure people’s normative orientations in mainland China perform in the same way and with equal efficiency in measuring cultural norms in Taiwan.

To carry out this analysis, I merged the four data sets and treated them statistically as subsamples representing the same population. On this dataset I conducted an IRT confirmatory factor analysis conditioned by three constraints. First, I imposed the same two-dimensional factor structure as in Table 3.2; second, I assigned the same three variables to each of the two latent constructs as in Table 3.2; and third, I imposed the constraint that the difficulty level for each indicator should be the same across the four subgroups. If the structures of cultural norms differed across the four subgroups, or if the observable variables were not related to the same latent constructs in every subgroup, the standardized coefficients that express the relationships between latent constructs and observable variables for different subsamples would be significantly different or the fit statistics for the overall model would be poor. If, on the other hand, the structure of norms in Taiwan is similar to that in mainland China, we should see similar standardized coefficients for each variable in each subsample and strong overall fit statistics.

Table 3.3confirms that the six variables performed equally well in measuring people’s normative orientations in mainland China and in Taiwan at different times. The structure of cultural orientations matched across the two societies and remained stable over time. The fit statistics are also excellent: the model CFI is .965 and TLI is .944. The RMSEA equals .048.

Data from mainland China in 2008 were also examined to determine whether cultural shifts occurred after 2002. The same three constraints as in Table 3.3 were imposed on the model using the three China data sets. Again, as shown in Table 3.4, the factor structure proves robust across the three data sets. The observable variables are associated with the latent constructs in the same way as in the previous tables, and the model generates strong goodness-of-fit statistics. If structural and institutional changes had altered people’s cultural orientations between 2002 and 2008, the relationship between latent constructs and observable variables for different subsamples would have been significantly different, and the fit statistics would have been weak.

Table 3.4. Confirmatory IRT Model for Combined Mainland China Data

^a All standardized coefficients are significant at the .01 level.

^b Difficulty is the measure of how unlikely it is that respondents will disagree with a questionnaire item.

Notes: Mplus is used for the estimates. For binary variables, the results from Mplus are identical to the results from other IRT estimates.

Source: 1993 Survey of Political Culture and Political Participation in Mainland China, Taiwan, and Hong Kong; 2002 China and Taiwan surveys, Asian Barometer Survey I; 2008 China Survey, Asian Barometer Survey II (1993 N = 3,287; 2002 N = 3,183; 2008 N = 5,098).

Conclusion

Despite the long separation between mainland China and Taiwan, the dramatic political changes in Taiwan, the rapid pace of social and political change in both societies, and the differences in the developmental levels they have attained, the structure of normative orientations remained remarkably similar in both and stable over time. Two separate norms could be empirically identified in both societies, and questions designed to capture people’s orientations toward the two norms did a good job of capturing them. These findings enable us to go on in Chapter 4 to construct scores that reflect where people stand and to begin to analyze how and why people’s orientations toward norms change.