Confucianism and Preferences Evidence From Lab Experiments in Taiwan and China Confucian Ch2

To measure individuals’ interpersonal trust and trustworthiness, we modeled our game after Berg, Dickhaut, and McCabe (1995). Each subject was randomly assigned the role of being an investor or a trustee. The investors were given 150 tokens and were paired with a subject who was a trustee, but they were not told who that person was. Investors had seven choices (between 0 and 150 in increments of 25 tokens) of how much to send to their trustee. We tripled the amount the investors decided to send to the trustee in each of six possible scenarios. The share of tokens that the investor gave out served as a proxy for interpersonal trust. Those who were more trusting would invest a higher share of the 150 tokens. The share of tokens returned by a trustee to the investor was a proxy for the trustworthiness of the trustee, where a higher share of tokens being returned indicated the trustee was more trustworthy. 3.6 Survey The survey collected basic information on the participants, including age, gender, major, parental education, and graduate student status; a set of questions asking about their upbringing, which sought to determine how much emphasis their parents had placed on observing Chinese traditions; and a set of questions about their subjective views (e.g., whether they agreed that it is unfortunate to have a daughter and no son). Having these answers to the subjective questions allowed us to construct a proxy for their conservatism that we used in the later analysis. 3.7 Data and Sample Selection Table 4 presents the summary statistics of subject characteristics by university. The last column in the table shows p-values on a test of equality of means between PKU and NTU. Among the recruited students, PKU students were more likely to be in graduate studies than the students at NTU. Their parents were also slightly less educated than the parents of NTU students. There are no statistical differences in gender, or in the students’ subjective evaluation of their parents’ emphasis on a traditional upbringing. Table 5 performs a randomization check within each school. In neither school are there statistically significant differences between treatment and control groups in all reported background 11 characteristics. Given that the randomization between control and treatment groups was successful, in later analysis we can conclude that any detected difference between treatment and control groups is attributed to the reactions to the Confucius prime. 4. Empirical Analysis 4.1 Descriptive Findings We first examined the differences between the treatment and control group across each of the games. Figure 1 presents the cumulative distribution over the number of safe choices by treatment status for the risk aversion game. It shows that the distribution of lottery choices for the treatment group is to the left relative to the distribution of the control group, that is, the treatment group chose fewer safe lotteries compared to the control group. Therefore, the treatment group was more risk loving than the control group. Figure 2 presents the distribution over the number of choice A (low-loss options) by the treatment and control groups in the loss aversion games. Similar to Figure 1, the distribution of the treatment group is also to the left of the distribution of the control group: these groups tended to choose fewer low-loss option, indicating that it was less loss averse than the control group. In the investment game, investors were asked to invest 0, 25, 50, 75, 100, 125, or 150 tokens with an anonymous trustee. Figure 3 presents the distribution over investment choices by treatment and control group. Here, an interesting pattern emerges: the treatment group was more likely to have extreme choices than the control group. A significant higher proportion of the control group invested half of the 150 tokens with the trustee than did those in the treatment group. The treatment group was more likely than the control group likely to invest either 0 or 150 tokens with the trustee. This suggests that after receiving the Confucius prime, subjects become more extreme in their interpersonal trust than subjects who were not treated. 12 In this same investment game, trustees would receive 0, 75, 150, 225, 300, 375, or 450 tokens. We asked trustees how many of the tokens they would return to the investors for each of the six scenarios (there is no token to be returned if they receive zero). Figure 4 presents the mean share of tokens that trustees returned for each of the scenarios by treatment status. The first noteworthy point is that the more tokens the trustees received, the more they would reciprocate and return a higher share of tokens. The treatment group returned a higher proportion of the tokens to the investors than the control group did regardless how much they received. In other words, the treatment group was more trustworthy than the control group.19 Figure 5 shows the choices in the CTB task by treatment status for each of the games. The left panel presents the decisions for the allocation of tokens between receiving it today versus in 4 weeks (games 1 to 5 in Table 3), and the right panel presents the choices made between receiving the tokens in 6 weeks versus 10 weeks (games 6 to 10 in Table 3). Each bar indicates the mean proportion of the tokens allocated to the earlier payment date. The lower the bar is, the more patient the subjects were. There are a few noteworthy points here. First, when interest rates increased, the share of tokens allocated to the earlier date decreased. Second, there seems to be some evidence of present bias to allocating more tokens to an earlier payment for any given interest rate in the left panel than the right panel. However, for the purpose of the comparison across treatment and control groups, the difference is negligible. 4.2 The Effect of Confucius Prime on Risk Preferences, Trust/Trustworthiness, and Time Preferences In this section, we pursued regression analysis in examining the differences in priming effect and report the result in Table 6. First, we compared the responses of the treatment and control groups for their choices in the incentivized lottery tasks. In these regressions, both the direct measure of risk or loss aversion from the experiments (the number of safe options and number of low-loss options) and the 19 Interestingly, the mean proportion of tokens returned (20%-30%) in both the treatment and control groups were lower than other trust experiments done with Chinese subjects (Buchan, Johnson, & Croson, 2006, reported 34%). 13 imputed measures of risk or loss aversion coefficients (σ, λ from equation 2) were the dependent variables. These regressions controlled for a subject’s gender, age, whether he or she was a graduate student, his or her father’s and mother's education levels, a dummy indicating whether the major was science- or engineering-related, and a dummy indicating that he or she was a PKU student.20 Standard errors are clustered at the session level. The regression results and the mean and standard deviation of each of the dependent variables are presented in Table 6. The coefficients of relative risk aversion have a mean of 0.45 and are similar to other experiments using different experimental protocols in China (Liu, 2013; Frijters, Kong, & Liu, 2013). We found that the treatment group was less risk averse than the control group at a 5% level in both the direct and imputed measures of risk aversion. The size of the treatment effect is considerable: one-fifth of a standard deviation.21 The treatment group also behaved in a less lossaverse way than the control group, although the coefficients are not statistically different from 0. Based on our hypothesis in section 2, it is possible that the salience of Confucianism caused subjects to react negatively to Confucian values. We later found that this result was driven by the reaction of PKU students who were behaving in opposition to Confucian values. Next, we examined interpersonal trust and trustworthiness. Given our finding that the treatment group had more extreme choices than the control group (from Figure 3), it is not surprising that the average treatment effect of the Confucius prime on trust was not statistically different from 0. In terms of trustworthiness, subjects returned almost 4.57 percentage points more tokens to the investor, a nearly 20% increase from the baseline. This suggests that the treatment group behaved in a more trustworthy way compared to the control group, in accordance with our prediction. To examine the impact of treatment on time preference, we had two sets of measures: the share of tokens allocated to the earlier payment (the direct measure) and structurally estimated parameters of discount factors. For the direct measure, each subject was asked to make decisions between receiving 20 Results are robust when these individual characteristics are excluded. This table is available in the Web Appendix on the authors’ website, www.uh.edu/~emliu. 21 To compare the magnitude to other covariates, the treatment effect is larger than coefficients on being female. 14 payment today or 4 weeks later with varying interest rates (games 1 to 5 in Table 3) and another set of decisions between receiving payments in 6 weeks or 10 weeks later (games 6 to 10 in Table 3). The following equation is estimated for games 1 to 5 and games 6 to 10 separately: (3) = ℎ + + where i is indexed for subject, g is an index for game, and r is the corresponding interest rate for game g. The dependent variable is the share of tokens allocated to an earlier payment. includes the same set of control variables as in other columns of Table 6. However, standard errors are clustered at the individual level. The regression results for the allocation choices for today versus 4 weeks is presented in column 7, and allocation choices for 6 weeks versus 10 weeks are presented in column 8 of Table 6. The intertemporal choices between 6 versus 10 weeks capture the discount factor, while the intertemporal choice between today versus 4 weeks encapsulate the present bias and discount factor. As shown here, the direct measures of the effect of the Confucius prime on time preference in a reduced form are not statistically different from 0. However, such a simple comparison would ignore the role of concavity in the utility function. For example, two individuals with the same allocation of tokens for an earlier payment in games 1 to 10 could still have different estimated time preference parameters if their risk aversion parameters differed. In particular, following Laibson (1997) and Andreoni and Sprenger (2012), suppose the utility function forms from different periods are time separable and can be expressed as (4) In this proposed utility function, ( , )= ( ) + ( , )= ( ) + ( ( ) ) =0 >0 is the exponential discount factor and there is no evidence of present bias; if < 1, present bias exists. A higher is the present bias.22 If = 1, would indicate that a subject is more patient and better able to delay gratification. 22 Anderson et al. (2008) suggest that ignoring the concavity would result in upward bias estimates of discount rates. 15 As we have shown in Table 6, columns 1 and 2, the concavity of utility function, , differs between treatment and control groups, so it is important to incorporate risk aversion coefficient in our estimation of the time preference parameters by treatment status. Using the estimated risk preference parameters based on subjects’ response from the Holt and Laury game and following Andreoni et al.’s (2013) interval censored regression estimation method, we impute the present bias (β) and exponential discounting factors (δ). These parameters are stratified by treatment status and shown in Table 7.23 β is statistically different from 1 for both treatment and control groups, so we cannot reject the existence of present bias. Interestingly, the treatment group had less present bias but a lower discount factor (the subjects were less patient), which is surprising. In the next section, we find that the reduction in present bias was being driven by the NTU subjects, and the surprising finding of subjects becoming more impatient was being driven by the PKU students, who were reacting negatively to Confucius teaching. 4.3 Experimental Results by School In this section, we examine whether the Confucius priming effect was different between the groups from the two institutions in this study, and we investigate whether the puzzling finding from the previous section is being driven by the negative reaction of PKU students. Our regression specification is (5) = + ( ∗ )+ + , where i indexed for subject i; NTU is a dummy indicating whether the subject was an NTU student; includes a subject’s gender, age, graduate student status, whether major is science and engineering related, and father’s as well as mother's education level; treat equals 1 if a subject was in the Confucius-prime group; captures the treatment effect on PKU students; and is the difference between the treatment 23 Andreoni et al. (2013) propose several different methods of estimating parameters in CTB, including nonlinear least squares and two-limit tobit (censored) maximum likelihood regressions. As they suggest, tobit is a better method when there are corner solutions. In our experiments, we have a considerable number of choices, and for the purpose of our analysis, we do not need individual estimates since we only need to compare the mean across the treatment and control groups. Thus we opted for the censored regression method. In our estimation, we also used the weekly expenditure of 1539 TWD (for PKU students) and 1746 TWD (for NTU students). These expenditures are based on the exit survey. 16 effect on NTU students and on PKU students. The treatment effect on NTU students would be equal to ( ( + , ). The dependent varible , is the same as those in Table 6. Table 8 shows the coefficients ) and the p-values from the F-test ( + =0). In section 4.2, we saw that being primed on Confucianism caused subjects to behave in a more risk-loving and less loss-averse manner. From columns 1 to 3 of Table 8, it is evident that the negative treatment effect on risk preference and loss aversion was mostly driven by the reactions of the PKU students. The F-test results reported in columns 1 to 3 suggest that the priming effects on NTU students were not statistically different from 0 at a 10% level. Moreover, there was no priming effect on trust for students from either school. Although there is also no significant treatment effect on trustworthiness in each school if separately estimated, the sign of the estimates suggests that the signficantly positive effect of Confucius primes on trustworthiness, reported in the pool regressions of Table 6, stemmed mainly from the positive reactions of NTU students. Columns 7 and 8, which show the share of tokens invested to the earlier payment, reveal little change in these direct measures of time preferences. 24 Table 9 presents the structural estimates of present bias and the discount factor from equation 3 stratified by school and treatment status. The treatment groups at NTU (column 4) exhibited less present bias, but those at PKU (column 3) show no statistically significant differences. However, the surprising results from Table 7 that those receiving the Confucius prime became less patient (lower discounting factor) is driven mainly by the changes in PKU students (see Column 1). In this section, we have presented some evidence that PKU students reacted against the commonly understood Confucian values when they were primed with Confucian teachings, but NTU students reacted in a way that is either consistent with or not affected by the Confucian values. That the 24 As a robustness check, we also split the sample and estimate the treatment effect for the two schools separately. One reason to do this is that the control variables may have different effects on the dependent variables for the Chinese and the Taiwanese students. The results are consistent with the findings in Table 8. However, due to the small sample and a lack of degree of freedom, the standard errors are larger than those in Table 8. This table is available in the Web appendix on the corresponding author’s website, www.uh.edu/~emliu. 17 same primes evoked different feelings in students from the two schools is an interesting finding.25 An explanation for these differences may be that the priming instrument we developed was not priming Confucianism but instead evoked other feelings of the PKU students, the reason for this opposite reaction. We investigate this possibility in the next section. 4.4 Validation Check We conducted a few extra surveys at both universities to validate the efficacy of our priming instrument. The surveys, distributed in several economic undergraduate classes to 389 students (none of whom participated in the experiment) in both schools, took about 5 to 10 minutes and the students were not paid. As with our full experiments, subjects were randomly assigned to receive either the Confucius prime or a neutral prime. Immediately after the prime, they were asked to complete a short questionnaire asking about their evaluation of Confucianism and some of their individual characteristics. Two of the questions allowed us to conduct a validation check: “Among the following belief systems (Rationalism, Confucianism, Eastern Religion (Buddhism, Taoism), Western Religion (Catholicism, Christianity, etc), please rank them based on their values from most agree (1) to least agree (4),” and, “On a scale of 1 to10, how much do you agree with each of the following belief systems (1 being least agree and 10 being most agree)?” Because these questions were answered immediately after the priming questions, the priming effect, if any, should be clear. The regression results using these answers are reported in Table 10. In column 1, the dependent variable is the ranking of Confucianism (1 being the best, 4 being the worst), and ordered probit estimation method is used. Column 2, the dependent variable, shows how much the students agreed with Confucianism.26 In column 3, we created a dummy indicating whether Confucianism was ranked as the most important. The results suggest that the PKU students ranked Confucianism lower and disagreed 25 Psychologists Wheeler and Berger (2007) suggest that the same priming instrument can have different effects on the subsequent choices of different groups of people. 26 Each person reported the values for all four belief systems. Some people tended to agree and put down a higher value for all, while others tended to disagree with everything and had lower values for all systems. Thus, the dependent variable is their rating on Confucian value minus their individual mean of evaluation of all four systems. 18 more with Confucian values after they received Confucian prime. For NTU students, the Confucius prime improved the ranking of Confucianism and increased how much NTU students agreed with Confucian values than the control group did. Combining the results here with those in the previous section reinforces the idea that PKU students reacted to the Confucius prime differently from the NTU students. 4.5 Discussion While we cannot clearly identify the reasons behind the observed differences in reaction to the Confucianism priming in the two schools, there are a few possible explanations. First, the differences in the sample characteristics may play a role. For instance, Table 4 suggests that the PKU students are slightly older, more likely to be in science- and engineering-related majors, and more likely to be graduate students than NTU students. The parents of PKU students are also less likely to hold a bachelor’s degree. However, unreported regressions show that the estimated priming effects in almost all dimensions are not affected by these variables (i.e., students’ majors, graduate degree, etc.).27 Therefore the school differences in the priming effect are not caused by the differences in these characteristics. Second, Table 10 shows that Taiwanese students exposed to the neutral prime hold significantly less favorable attitudes toward Confucianism than their Chinese peers. It could be that those who started from a lower identification with Confucianism can demonstrate a larger positive reaction to the priming. However, this line of reasoning cannot explain why PKU students’ high level of identification with Confucianism in the neutral treatment is associated with a negative reaction to the Confucianism priming. In a regression not reported here, we investigated whether the treatment effects can differ based on students’ subjective view of traditionalism (based on their answers in the exit survey), but found no statistically significant evidence. Third, there were historical differences in the experiences with Confucianism in these two regions. As we have suggested earlier, the Cultural Revolution in China has devastated many Confucian values for 27 The only significant effect is that if the parents' education levels are higher, then subjects tend to demonstrate more positive treatment effects on the degree of trustworthiness. 19 a decade, which may have long-lasting effects even today, while in Taiwan Confucianism was relatively well preserved. As with most laboratory experiments, our studies have certain limitations. First, our subjects are students in top universities in China and Taiwan. Having university students as subjects often raises the question of the external validity of experiments. However, this problem may be less severe given that the key motivation of our study is to understand how Confucian philosophy shapes Chinese society. For more than a thousand years, Confucianism had remained influential in Chinese society because of its key role in imperial civil service examination system (Mayer and Schmidt, 2004). Memorization of Confucian texts had been a necessary preparation for examinations (Elman, 1991), and Confucian morals have been inculcated in intellectuals (Brokaw, 1991). These intellectuals possessed political power and shaped the norms and the thinking of the society. The rest of society was influenced by the Confucian teachings indirectly. Similarly, today, Confucian texts are taught in middle schools and high schools in Taiwan and China, and students are often tested on Confucian texts during college entrance examinations. Hence, university students who are well-versed in Confucian sayings are in fact the subjects we were most interested in. They are influenced by Confucianism directly through active learning and reading texts, unlike the rest who are only indirectly influenced through societal norms. We suspect that those who are uneducated (in Confucian texts) may not have as much reaction to our priming task at all. Second, our validation tests were carried out on a more special group of students who had economics major, and so the validation results may not directly apply to students in other majors. In unreported regressions, we find that the priming effects are not affected by students’ majors, which to some extent alleviates the problem. 5. Conclusion We investigated how Chinese and Taiwanese subjects reacted to Confucius primes and whether there were differences of reaction between the two groups. Our findings suggest that subjects became more risk loving, more impatient, less present-biased and more trustworthy when they were exposed to a 20 Confucius prime. However, this simple characterization of the impact of Confucianism is misleading since we found differences in priming effects depending on whether they are from China or from Taiwan. Chinese subjects who were primed for Confucianism became more risk loving, less loss averse and more impatient than the control group, behaviors not in accord with Confucian values. Taiwanese subjects who were primed for Confucianism demonstrated less present bias than the control group, and was also the main driver force for the more trustworthy treatment effect, in accordance with Confucian values. We performed validation checks of our priming instruments by asking subjects to rank four different belief systems (Confucianism, rationalism, Eastern religion, and Western religion) and reported on a scale of 1 to 10 on how much they agreed with each of these systems. Chinese subjects who were primed for Confucianism were more likely to rank Confucianism lower and also tended to agree less with Confucian values than the control group did. We did not find this effect on the Taiwan subjects. Our study therefore suggests that Chinese and Taiwanese subjects have different reactions to Confucianism. Different historical events and experiences in the two places are a very likely explanation for these differences. There has been a revival of interest in Confucianism in China. Unlike the anti-Confucius rhetoric from the 1960s and 1970s, government officials have once again acknowledged Confucius as a great contributor to the glorious Chinese tradition. Professor Yu Dan at Beijing Normal University has a popular TV show on the Analects of Confucius and her book, Reflection on the Analects, became a top seller in 2009. Our study suggests that social norms and social identities may be sticky, but they can also be altered. It is possible that we will see the currently diverging paths between China and Taiwan converging again. 21