Trade, social values, and the generalized trust.

Chan, Kenneth S.

The present paper investigates, empirically, whether global economic integration can lead to national social disintegration, as proclaimed by Rodrik. One can perhaps conjecture that a drop in the generalized trust in a country may signal a rise in social disintegration. Using the survey data on the generalized trust provided by the World Values Surveys, the present paper finds that, in general, there is a positive openness-trust relationship. However, when globalization undermines domestic income distribution, it will weaken this openness-trust relationship. There is also evidence of a threshold inequality above which the openness-trust relationship turns negative.

JEL Classification: F02, F10, F13, Z10

1. Introduction

The present paper investigates, empirically, whether global economic integration can lead to national social disintegration, as proclaimed by Rodrik (1997). By social disintegration, Rodrik refers to the uprooting of social values of a nation without a suitable replacement. "Nations do have legitimate reasons for worrying about what globalization does to their norms and social arrangement," says Rodrik (1997). Rodrik's pronouncement is closely akin to, if not part of, those contentious trade issues arising from differences in national labor and environmental standards, cultural practices, and child labor, for example.

To tackle this problem, the present paper chooses the concept of social capital (the stock of norms, rules, shared understandings, and expectations, etc.) in the literature as the basis of evaluation. The choice of this particular framework is discussed in Section 1. A useful function of social capital is to foster collective actions of the community. The latter can be measured by the trust among members of the community. One can perhaps conjecture that a drop in trust may signal a rise in social disintegration. Using the survey data on generalized trust provided by the World Values Surveys, various predictions on trade and generalized trust, reported in Section 2, can be tested. The present paper finds that the openness of a economy, ceteris paribus, does not always lead to a decrease in generalized trust. Income distribution is found to play an important role in this openness-trust relationship. Maldistribution of income reduces the positive impact of openness on generalized trust. Hence, when a more open economy raises the income inequality of a divided nation, the chance of social disintegration may increase. Section 3 concludes with a discussion of results.

2. Theoretical Analysis and Predictions

Rather than using the conventional utility or social welfare function to evaluate the aforementioned issues, the present paper uses output performance as a criterion to evaluate the impacts of different social values. In other words, social values are considered here as part of the so-called social capital, discussed further, that aids production. (1)

Literature Survey

Recently, it is increasingly common for social scientists to use the term "social capital"-the stock of norms, values, civic traditions, conventions, etc., that constrain a person's actions--to explain collective action in a community. Coleman (1988) defines social capital as "people's capacity to work voluntarily together." Similar to the concept of capital, the stock of social capital is productive because it reduces market transactions cost. (2) Most researchers (e.g., Stiglitz 2000) agree that this kind of civic cooperation is the foundation of the nation's informal institution that governs market exchanges and "glues" the community together. The previous discussion can be summarized by the following function:

V = [upsilon](K,[alpha]); [[upsilon].sub.1] > 0, (1)

where V is voluntary cooperation, K is social capital, and [alpha] is a country specific variable.

In the literature, Fukuyama (1995) and Dasgupta (2000) ascribe the ability to cooperate voluntarily to the trust among the people. Fukuyama and Dasgupta define trust as the common expectation of individuals that others' behavior in the community will be cooperative. Indeed, as argued by Paldam (2000), trust and the ease of voluntary cooperation should be two highly interlinked, if not almost identical, concepts. Without trust, cooperation will be limited only to activities that can be easily monitored or enforced. Paldam (2000) also shows that the two concepts imply each other. Elsewhere, Putnam (1993, p. 171) makes a similar argument: "Trust lubricates cooperation. The greater the level of trust within a community, the greater the likelihood of cooperation." In the same vein, Zak and Knack (2001) define trust as, "the resources that agents spend in production rather than in verifying or monitoring the actions of others, which is identical to the defined amount of cooperation in the community." The two concepts are, therefore, equivalent to the definition used by Zak and Knack. The argument in the literature can be summarized as:

T = T(V) = T([upsilon](K,[alpha])) T' > 0, (2)

where T is the generalized trust and is directly related to, if not identical to, voluntary cooperation, V. (3)

To support Rodrik's argument emphatically, one has to show that a drop in the generalized trust in a society leads to social disintegration. This theoretical link is tenuous at best. Although a low level of generalized trust can lead to a lack of social cohesion, it does not automatically imply social disintegration. Moreover, it is difficult to define the normative concept of social disintegration, let alone finding data that measures it. Because of this difficulty, this paper only examines the impact of trade on the generalized trust. This, by itself, is a positive and interesting research question, somewhat different from, but close enough to shed light on, the research question posted by Rodrik. There is no need to enlarge the research focus to include social disintegration or to generalize to such level of abstraction. There are two views on the impact of globalization on the social values of a nation:

The Pessimistic View

According to this view, globalization weakens the social cohesion within a country, which eventually leads to social disintegration (see Rodrik 1997). There are two channels through which this can happen:

(i) When conflicted with the social norms and traditional values of other nations, domestic norms and values (including workplace practices, rules, social insurance, etc.), which confine how domestic goods are produced, could be eradicated through the forces of trade. In terms of Equation 1, if some of the K is destroyed from trade, there will be less trust and cooperation. This will weaken the social cohesion of a nation.

(ii) Globalization exacerbates tension among groups and pushes the less fortunate into despair. With increasing resentment and insecurity from those who fall victim to globalization, social order and solidarity decline (Rodrik 1997). (4) Moreover, the tax base of welfare states is steadily eroding away as capital and skilled workers become internationally mobile. Funding for social insurance and redistributive programs are weakening considerably (see De Grauwe and Polan 2005 for an update). The retreat of welfare states everywhere aggravates social divisiveness. Hence, globalization weakens the social cohesion or trust within a country.

Rodrik's more elegant and updated argument echos Karl Polanyi's, that integration with the global market economy tended to destroy domestic social relationships that held human society together. "To separate labor from other activities of life and to subject it to the laws of the market was to annihilate all organic forms of existence and to replace them by a different type of organization, an atomistic and individualistic one," said Polanyi (1944, chapter 14, p. 163), and "such a scheme of destruction was best served by the application of the principle of freedom of contract. In practice this meant that the noncontractual organizations of kinship, neighborhood, profession, and creed were to be liquidated, since they claimed the allegiance of the individual and, thus, restrained his freedom." In the long run, by freeing social constraints, Polanyi argued, the market economy undermined social order and threatened to destroy the social institution upon which the market economy is based.

The Optimistic View

According to this view, globalization integrates national cultures, resulting in a better mix of pluralistic cultures. Although there is a clear loss of national cultural autonomy in a more open regime, it does not mean national cultures and values are dysfunctional. On the contrary, a better mix of pluralistic culture improves the functioning of domestic social capital, leading to a gain in social cohesion.

Sen argues (1999, p. 241), "When an economic adjustment takes place, few tears are shed for the superseded methods of production and for the overtaken technology. There may be some nostalgia for specialized and elegant objects (such as an ancient steam engine or an old-fashioned clock), but, in general, old and discarded machinery is not particularly wanted. In the case of culture, however, lost traditions may be greatly missed. The demise of old ways of living can cause anguish, and a deep sense of loss ... but it is up to the society to determine what, if anything, it wants to do to preserve old forms of living, perhaps even at significant economic cost. Ways of life can be preserved if the society decides to do just that, and it is a question of balancing the costs of such preservation with the value that the society attaches to the objects and the lifestyles preserved." Although there are clear circumstances in which Sen's argument is incorrect, (5) one can perhaps think of his argument as correct for most situations, rather than correct unequivocally.

In a similar vein, Bhagwati (2004, chapter 9) argues that trade may spread useful values that improve domestic social conditions. "Trade is the friend, not the foe, of social agendas .... " he says (2001), "Trade means cultural as well as economic interchange ... It is the traditional elites who are most affected by globalization. And it is they who are most likely to react against social change."

According to the optimistic view, although globalization may eliminate some form of domestic social capital that is no longer useful, it may add other forms of foreign social capital that are more appropriate. There should be more of a choice of global norms and values in a more open regime than in a less open one.

An Analytical Framework

There is a general concern that the term "social capital" is just another "buzz word" for something we do not know rather than a precise and operational concept. In response to this concern, I adopt in this paper a definition of social capital by Lin (2001): "Social capital can be defined as the stock of rules, norms, values, traditions, etc., embedded in a social network, which can be mobilized by agents in the social network for collective action." This definition will be useful for the framework described further on.

A change in domestic prices from liberal trade may provide an impetus for change in social values. Learning useful foreign values is similar to acquiring useful foreign technologies. As contracts are renegotiated, the old norms or informal rules in which contracts are nested may need improvements in light of the new foreign values. Restructuring the higher set of dated rules and norms may improve renegotiations that raise productivity and competitiveness. When the old values are to be restructured, facilitated through the social network and the stock of social capital, groups that stand to benefit will devote resources to restructure the rules and norms at a higher level and may even compensate the groups that stand to lose from restructuring. But, similar to the trade in commodities, interest groups in a fractional society could block trade liberalization and could distort the domestic economy to the extent that the benefits from trade could be lowered. In a fragile social network plagued by opportunism, the exchange of national values could also be blocked by interests who are threatened by the slow erosion of old social patterns and traditional ways of life. There may be powerful domestic elites who depend upon traditional values for their legitimacy. These interest groups usually wrap themselves in nationalistic cultural identity. Therefore, in a fractional society, in which the gainers could not compensate the losers under the improved hybridized values, those who are threatened by the disappearance of the old and less productive social values would either block the adoption of the more productive foreign value or break away from the existing social network. In either case, the fragile social network would be weakened further, and subsequently, the generalized trust would be lowered. Therefore, a fractional society must affect the marginal impact of openness on the improvement of domestic social capital (i.e., affecting the partial derivative [partial derivative]K/[partial derivative] O, where O is openness; see Equation 4 later).

The analytical impact of income inequality on trust is in the literature. It is much easier for a homogenous group, in which members have similar interests, to achieve collective actions than a heterogenous group with diverse interests, argued Olson (1982). For one thing, it is easier for a homogenous group to achieve consensus. And for another, it is less costly to develop social incentive mechanisms to deter opportunism in a homogeneous group that is socially more interactive. Social capital and social networks are the basis upon which these informal incentive mechanisms are built. Olson's analytical framework should be applicable to communities with low income inequality as well.

Zak and Knack (2001) have also modeled the impact of income inequality on trust. They argue that production and supervision of transactions are alternative uses of time. When trust in the society is low, more time is spent in supervising transactions than in production. High income individuals, therefore, spend less time supervising in contrast to the choice of low income individuals who spend more time supervising. Zak and Knack (2001) assume that adjustments in supervision for low wage individuals are more sensitive to changes in wage than those of high wage individuals. Therefore, high wage inequality, hypothetically generated from a mean preserving spread on a distribution of wages, reduces trust, where Zak and Knack define trust as the sum of supervision time of the high and low wage individuals.

Following Zak and Knack's work on a closed economy, one can further argue that beneficial foreign values improve the technology of supervision. Hence, less time is spent on supervision and more on production. This improves the generalized trust. But, if globalization raises the income inequality of a country, it can lower the generalized trust and offset some of its initial improvements.

Elsewhere in the literature, Jackson and Wolinsky (1996) examine network stability. Their framework can be extended to the current problem. The Jackson and Wolinsky social network model introduces a very small amount of cooperation among agents into an otherwise noncooperative environment and finds that small amounts of cooperation can go a long way because agents are tied together into a social network. Although agents develop limited cooperative links among themselves, the transitive nature of links means that weak cooperation among agents can generate strong cooperation within the network. Jackson and Wolinsky also introduce the notion of stability of network, in which agents would neither sever an existing link nor create new links. The stability of a network, therefore, depends critically on other network or subnetwork options of the participants. The most efficient social networks may not be stable. One of their main findings is that an equal allocation of gains among participants within the network increases stability. (6) For a homogeneous group, there is no intuitive reason why an encompassing social network is not beneficial in terms of cost and benefit of participation [(see Burt (2001) for a similar argument on "structural holes" of networks]. But for a heterogenous group, a weak link may exist at which participating in the network no longer yields benefit to those with a weak link. Forming an encompassing social network is, therefore, difficult. The society may break into small subnetworks that can better serve the more homogeneous subgroups. This multitude of subnetworks and interests can be detrimental to the society. As cautioned in the literature [see Putnam (1993), among others], social capital can also facilitate small collusions that are malicious to the society. Easy examples of this are cults, and criminal and racist organizations. When each sector in the economy has its own self-contained social network and social capital, each sector will exploit its monopoly position. (7) Jackson and Wolinsky's work clearly concludes that the stability of a social network can weaken with income inequality. Their work explains the negative impact of inequality on the generalized trust. If trade liberalization breaks up the domestic social network into fragmented selfcontained subnetworks, (8) as warned by Rodrik, these disjointed subnetworks could be harmful. Another theoretical link between inequality and generalized trust comes from the experimental literature. Laboratory results in the past have shown that the behavior of participants in experiments often deviated widely from the predictions of conventional game theory models. Inspired by these confounders, experimental researchers have been calibrating theoretical models that incorporate non-self-interested components, such as inequality aversion and fairness, into the agent's objective function. On the whole, this new objective function is successful in explaining many of the confounders [see Camerer (2003), chapter 2 and the references cited]. Camerer (2003), in his review of the experimental literature on trust and dictatorial games, concludes that a subject's trustworthiness is mainly motivated by fairness considerations and inequality aversion. In other words, when payoff inequality in the laboratory environment is viewed as unfair by subjects, it destroys trust among them. Based on experimental findings, one can conjecture that in the field when income inequality is viewed as unjust, it can destroy generalized trust in the community. The previous discussion can be summarized by the following equations:

T = T([upsilon](K,[alpha])) = t(K; [alpha]) [t.sub.1] = T' [[upsilon].sub.1] > 0 (3)

K = K(0, I) [K.sub.1] > 0 [K.sub.2] < 0 [K.sub.12] < 0. (4)

Equation 3 is the same as Equation 2, reproduced here for convenience. In Equation 4, I is the index of income inequality, a higher value of I means higher inequality, and O is the degree of openness. Equation 4 says that domestic social capital K can be improved by foreign values,9 transmitted from openness, the marginal impact of which could be blocked by domestic interest groups (approximated by I) who stand to lose from adopting new foreign values (hence, [K.sub.12] < 0). Substitute Equation 4 into Equation 3, we have:

T = t(K(O, I); [alpha]) = H(O, I; [alpha]) [H.sub.1] = [t.sub.1][k.sub.1] > 0 [H.sub.2] = [t.sub.1] [K.sub,2] < 0 [H.sub.12] = [t.sub1] [K.sub.12] < 0. (5)

Equation 5 suggests that if globalization ultimately worsen the nation's income distribution, it can lower the generalized trust via the inequality-trust channel.

From mainstream trade theory, the impact of openness on domestic inequality is generally indeterminate: Although trade redistributes factor incomes, it can either improve or worsen the income inequality in a country. Moreover, the ability and the willingness of domestic institution to redistribute the gains from trade can play a role. One can therefore specify:

I = i'[O] [i.sup.>.sub.<]' (6)

However, if one accepts the joint null hypotheses that openness destroys social values and that openness raises income inequality (that is, I = i(O) and i' > 0), then the total impact of openness on trust in Equation 5 must be negative (that is, the derivative dT/dO = [H.sub.1] + [H.sub.2] i' < 0).

The following predictions summarize the aforementioned discussions:

Prediction I: In a fragmentary society, integration with the global economy can break up social networks and social capital. Consequently, an open economy lowers the generalized trust of the nation.

Prediction H: In a homogenous society, an open regime allows societies to integrate and reshape components of global cultures. Some new, more useful values and practices are adopted and some old less useful ones are relinquished. In this respect, an open regime fortifies, instead of diminishes, the ability of social capital to bind its citizens together for collective actions. Generalized trust in the more open economies should therefore be higher.

With help from Equation 5, the predictions can be tested in the following functional form:

Trust = [a.sub.0] + [a.sub.1] (Openness) + [a.sub.2](Inequality) + [a.sub.3](Openness) x (Inequality) + [b.sub.4](Country Specific Variable). (7)

An interactive term is added to Equation 7 to test for the interactions between openness and income inequality as proposed in (4) ([H.sub.12] = [t.sub.1][K.sub.12] = [a.sub.3]). Prediction I (II) suggests that the value of [a.sub.1] + [a.sub.3] X (Inequality), which is the partial derivative of Trust with respect to Openness in Equation 7, is negative (positive). However, on closer examination, the two predictions need not be mutually exclusive because Prediction I (II) refers to the situation where income inequality is high (low). The two aforementioned views, Rodrik and Polanyi on the one side and Bhagwati and Sen on the other, are perhaps not as mutually exclusive as they first appear. The country-specific variable in Equation 7 is the [alpha] term in Equation 5. I choose either a country's income per capita or the GDP-per-capita growth rate for this variable; a detailed explanation follows.

Multicollinearity in the estimations is a concern here. For one thing, the variables Openness and Inequality are correlated from Equation 6. Because the ordinary least squares (OLS) estimations are unbiased under multicollinearity, as long as the t-statistics of the estimated term, [a.sub.1] + [a.sub.3] X (Mean Inequality), is significant, multicollinearity should not pose a problem.

3. Empirical Analysis

As for the Trust variable, I employ the survey data from the World Values Surveys (WVS) compiled under the direction of Inglehart (1994). This is the most systematic global values survey currently available. In the WVS, generalized trust for each country is computed as the percentage of respondents who agree that "most people can be trusted" rather than the alternative that "you can't be too careful in dealing with people." The WVS data has three waves. The first one is in 1980; the second and the third are in 1990 and 1995, respectively. I ignore the first wave and average the country data from the second and the third wave surveys; the second and the third waves have more countries than the first wave and the two surveys are five years apart, reasonably close to each other. (10) A total of 39 countries, excluding only the transition and nonmarket economies, are chosen for the present study. Inglehart (1994) believes that the urban areas and better-educated people are overrepresented in the sample. Accordingly, a weight was constructed to reflect this bias. This weight is used in the present paper to adjust the "raw" trust values. The data on generalized trust from the WVS have robust predictive power in growth accounting literature (see Knack and Keefer 1997, Zak and Knack 2001, La Porta et al. 1997). These findings give confidence to the present choice of data.

As for the openness variable, I employ the openness index constructed by Sachs and Warner (1995), OpenSW, which measures the past years of trade policy openness for a large sample of countries. The OpenSW index describes past trade policy, an exogenous variable that gives a sense of causation to the empirical analysis. I also employ a contemporaneous trade shares index (Open), a five-year average of export and import share of GDP in the same sampling period as the WVS, as an alternative test. To capture the historical impact of openness, I construct another index, [Open.sub.-1], using the average historical export and import share of GDP. I sample from the years 1970 to 1989, the same sample period as the Sachs-Warner openness index. Pushing the starting date of the sample back to earlier than 1970 will result in an incomplete data series for some countries. The end sample date, 1989, is the starting date of the second wave of the WVS.

The Gini coefficient, taken from World Development Indicators published by the World Bank, is employed to capture inequality. An alternative measure of inequality is the ratio of income shares of the richest 20% of the population divided by the poorest 20% (income-share ratio). Another measurement of inequality is the ethnic and linguistic fragmentation profile in a country. The ethnolinguistic fractionalization index is taken from Alesina et al. (2002). Although the ethnolinguistic fractionalization index generates only a subset of inequality, it remains to be an interesting variable to investigate. The summary statistics of the aforementioned variables are reported in Table 1.

In the field data, generalized trust and GDP per capita are highly correlated. Many researchers (see Putnam 1993; Knack and Keefer 1997; La Porta et al. 1997; Zak and Knack 2001; Beugelsdijk, de Groot, and van Schaik 2004; among others) have argued that the causality goes from generalized trust to GDP per capita (or GDP-per-capita growth rate), although the causality can easily be argued to go both ways. The present paper investigates how much trade can add beyond this observed trust and GDP-per-capita relationship. For this reason, I add one more regressor, either the income per capita or the GDP-per-capita growth rate, into the regressions.

Adding the per-capita income, or per-capita GDP growth in the regressions can also facilitate the comparison of generalized trust across countries. Trust saves resources. Since low income economies may not need to save the same amount of resources as high income economies, the amount of trust should be adjusted for the size of the economy. Moreover, at different stages of social and economic development, the opportunity cost of governance, which is also part of the marginal benefit of generalized trust, could be different (see Zak and Knack 2001) and should be accounted for in the regressions. The per-capita income or GDP growth is a well-established control variable for the stages of social and economic development in a country. (11)

The problem of using the per-capita income or per-capita GDP growth as a regressor is that they are endogenous and could correlate with other regressors. For this reason, I replace the income per capita variable by a rank variable, the Income-per-capita Category, which uses the classification of countries by the World Bank into four categories, low, lower middle, higher middle, and high income per capita. I convert these categories into ranks, from 0 to 3, with 3 as the high income category. The Income-per-capita Category variable records this rank value of countries. This variable cuts down the multicollinearity with other variables in the regressions. As for the GDP-per-capita Growth variable, I continue to use this variable in the regressions because the correlation coefficients of this variable with other regressors are low. (12)

I first explore how much openness can explain generalized trust after the Income-per-capita Category or the GDP-per-capita Growth has been accounted for. Table 2 shows exploratory regressions under various combination of openness, per-capita income, and per-capita GDP growth. Regressions (1) and (2) in Table 2 focus on the explanatory power of per-capita income and per-capita GDP growth on generalized trust. The value and significance of the coefficients of the per-capita income and per-capita GDP growth drop substantially when the openness variable OpenSW is added. The value of the coefficients and the level of significance of the OpenSW variable, on the other hand, has not changed much with the inclusion of the per-capita income or the per-capita GDP growth [regressions (3), (4), and (5)]. That means, a lot of generalized trust, which had hitherto been explained by per-capita income or per-capita GDP growth, could be explained by OpenSW. This shows the high explanatory power of the OpenSW variable. A scatter plot of Trust and OpenSW is provided in Figure 1, after controlling for the influence of income per capita [regression (4) in Table 2]. From regressions (4) and (5) in Table 2, because the total impact of openness on trust is positive, the hypothesis that openness destroys social values and raises income inequality can be rejected.

[FIGURE 1 OMITTED]

The Open variable does not fare as well as OpenSW. Regression (6) in Table 2 tests the impact of the Open variable on generalized trust, the R-square of which is low at 0.11. The value of the coefficients of Open, Income-per-capita Category, and GDP-per-capita Growth are reasonably consistent with each other in regressions (6), (7), and (8). The high R-square in regressions (7) and (8) comes from the influence of the Income-per-capita Category and GDPper-capita Growth. The explanatory power of the Open variable on Trust is much weaker than the OpenS W variable. One plausible explanation is that social capital takes time to accumulate. Its history is, therefore, important. The OpenSW variable captures past liberal trade performance of the economy, whereas the Open variable captures current liberal trade performance. (13) The same pattern between OpenSW and Open are also observed in subsequent regressions. As for the [Open.sub.-1] index, which captures past trade share of GDP, regressions (9), (10), and (11) show that this index explains generalized trust well.

As to the income inequality variable, Table 3 tests the impact on generalized trust from various combination of inequality measures, Income-per-capita Category and GDP-per-capita Growth. Judging from the consistency of the coefficients and the corresponding t-statistics, the Gini Coefficient explains Trust very well. Comparing the regressions (12), (13), and (14) with (1) and (2) in Table 2, one can conclude that a lot of generalized trust, which was explained by Income-per-capita Category or GDP-per-capita Growth, can in fact be explained by the Gini Coefficient. As for the Income-share Ratio, from the consistency of the coefficients, the t-statistics, and the R-square, comparing regressions (15), (16), and (17) with (1) and (2), the Income-share Ratio also appears to explain generalized trust reasonably well. As for the Ethnolinguistic Fractionalization index, the statistics in regressions (18), (19), and (20) suggest that the Ethnolinguistic Fractionalization index does not explain generalized trust as well as the other inequality indices. This is perhaps not surprising because the ethnolinguistic fractionalization index represents only a subset of a country's inequality. (14)

With the exploratory regressions discussed in the previous paragraph, I can now estimate Equation 7. In Table 4, the impact on trust from OpenSWand three different inequality indices are presented. The estimated coefficients of the explanatory variables for a specific inequality index are by and large consistent with one another. With the OpenSW and inequality variables in the regression, the Income-per-capita Category or GDP-per-capita Growth does not seem to play a dominant role in explaining the generalized trust. With the presence of the interactive term in Equation 7, the marginal impact of the openness variable on Trust is: [a.sub.1] + [a.sub.3 x (Inequality). Since the estimated value of [a.sub.3] is negative and al is positive in all regressions, evaluating the partial derivative at different values of the Inequality Index is necessary to gauge its overall impact of openness. At one standard deviation above (below), the mean is defined here as a high (low) value of Inequality. This is recorded in the bottom rows of Table 4 as the marginal impact of Openness at high, mean, or low (H, M, or L) inequality. Similarly, the marginal impact of Inequality is: [a.sub.2] + [a.sub.3] x (Openness). Since the estimated values of [a.sub.2] and [a.sub.3] are negative, evaluating the marginal impact of Inequality at the Mean of the Openness variable is sufficient. This is recorded in the bottom row of Table 4.

Several interesting results emerge in Table 4. First and the foremost, the marginal impact of OpenSW increases with decreasing inequality, both in terms of magnitude and significance. There is a threshold level of inequality above (below) which the impact of OpenSW on trust is significantly positive (can be negative). The threshold level of inequality is [-a.sub.1]/[a.sub.3]. Second, the marginal impact of inequality on generalized trust is negative and significant except when inequality is measured in terms of the Ethnolinguistie Fractionalization index, where the impact is insignificant but negative.

The next task is to examine regressions based on the Open variable. The OLS regressions using the Open variable are presented in the Appendix. From earlier tests presented in Table 2, the trade share (the Open variable) is only a weak predictor of Trust, because it captures current liberal trade performance, while generalized trust is built over time through the accumulation of social capital. Moreover, the Open variable is an endogenous variable in the system. To overcome the endogeneity of the Open variable, I use the computed trade shares from the gravity model, constructed from geographic parameters by Frankel and Romer (1999), as instrument for the Open variable in two-stage least squares (2SLS) regressions. The OpenGM variable in Table 5 is computed from the predicted value of the Open variable regression in the first stage. The 2SLS regressions are presented in Table 5. From regressions (J) to (M) in Table 5, the OpenGM variable has no impact on Trust, while the Inequality variable has maintained a negative significant impact. As for the regressions (N) and (O), using the Ethnolinguistic Fractionalization index for Inequality, the impact of Inequality on Trust is insignificant, but the impact of the OpenGM variable resurfaces: The marginal impact of OpenGM at low Ethnolinguistic Fractionalization is positive and significant. (15) Table 6 reports regressions using the historical export and import share of GDP (the [Open.sub.-l] variable). Regression results using the [Open.sub.-1] index do not support the expected relationship with generalized trust as well as those using the OpenSW index. But compared with those using the Open (see the Appendix) and the OpenGM (see Table 5) indices, they are much more supportive.

Tests for Omitted Variables

As in all empirical works, the possibility of omitted variables could create biased estimates and spurious relationships. The possible omitted variables for the present model are those structural variables that capture cultural, historical, and economic characteristics of different countries. Since they are difficult to identify, I choose the following broad set of structural variables, hoping to register some of these effects: The size of population can capture the size of the domestic market and polity. The regional dummies (African, East Asian, and Latin American countries) depict cultural and historical characteristics. An oil-exporting country dummy catches the most apparent difference in economic endowment. The democracy variable (taken from Freedom House) captures a facet of the social and institutional feature of a country. And finally, religion fractionalization, taken from Alesina et al. (2002), reflects religious values and beliefs. To test how these structural variables may bias the estimations, they are used as regressors, one at a time, to check the consistency of the estimated coefficients. This is reported in Table 7. From the regressions in Table 7, the original estimated coefficients in Table 5 look fairly robust. Therefore, the chance of getting a biased result from omitted variables seems low.

4. Conclusion

Global integration requires nations to make adjustments in some of their social values and practices or be left behind economically. This process is deemed to improve or degrade cultural practices, depending on interests that stand to gain or lose from these practices. This is similar to free trade where there are winners and losers in spite of an overall welfare gain. The present paper uses the generalized trust as criterion to reckon the all-inclusive impact of globalization on national values. From the present empirical study, openness generally improves the generalized trust and hence strengthens the informal institution, the "civic glue" that holds together and governs the society. Although globalization may eliminate certain dated social values and may transmit some useful "foreign" values to a community, its total impact on the community's trust can be positive. However, there are qualifications. Economic inequality weakens this positive effect. When globalization worsens domestic income distribution, it can have a negative impact on a nation's generalized trust through this channel. There is evidence of a threshold inequality above which an increase in globalization lowers generalized trust instead of raising it.

Finally, the two seemingly opposite views, Rodrik and Polanyi on the one side and Bhagwati and Sen on the other, may not be as antithetical as they first appear. Drawing from the present findings, perhaps Rodrik's view is a stern warning to fractional nations that globalization could lead to social disintegration. Based on a civil and homogenous society, Bhagwati and Sen, on the other hand, are arguing that the transmission of foreign values from globalization benefits national values.

I am indebted to two anonymous referees, Will Coleman, Tom Crossley, Lonnie Magee, and seminar participants at the City University of Hong Kong and at the National University of Singapore, for comments on earlier drafts. The usual disclaimers apply.

Appendix

Dependent Variable: Generalized Trust--OLS Estimation

 (V1) (V2) (V3)

Open -0.35 0.00 0.04
 (0.81) (0.00) (0.18)
Gini Coefficient -1.13 *** -1.05 ***
 (4.25) (3.17)
Open x Gini Coefficient 0.01 0.00
 (1.21) (0.31)
Income-Share Ratio -1.60 ***
 (3.922)
Open x Income-Share Ratio 0.01
 (0.61)
Ethnolinguistic
 Fractionalization
Open x Ethnolinguistic
 Fractionalization
Income-per- Capita Category 4.67 ** 6.08 ***
 (2.42) (3.31)
GDP-per-Capita Growth 0.32
 (0.66)
R-square 0.57 0.50 0.52
Marginal impact of Open at 0.30 0.18 0.22
 H-inequality (2.67) (0.84) (1.52)
Marginal impact of Open at 0.15 0.14 0.15
 M-inequality (1.40) (0.08) (1.26)
Marginal impact of Open at 0.01 0.10 0.08
 L-inequality (0.00) (0.25) (0.19)
Marginal impact of inequality -0.75 *** -0.94 *** -1.22 ***
 at mean Open (17.68) (24.01) (10.10)

 (V4) (V5) (V6)

Open -0.31 0.27 0.37
 (1.14) (1.26) (1.47)
Gini Coefficient

Open x Gini Coefficient

Income-Share Ratio -1.00 *
 (1.97)
Open x Income-Share Ratio -0.03
 (0.99)
Ethnolinguistic -8.42 -14.57
 Fractionalization (0.49) (0.80)
Open x Ethnolinguistic -0.01 -0.27
 Fractionalization (0.19) (0.51)
Income-per- Capita Category 6.29 **
 (2.38)
GDP-per-Capita Growth 0.65 0.47
 (1.32) (0.88)
R-square 0.41 0.37 0.26
Marginal impact of Open at -0.04 0.21 0.21
 H-inequality (0.03) (1.01) (1.35)
Marginal impact of Open at 0.10 0.24 * 0.28 *
 M-inequality (0.04) (2.92) (3.37)
Marginal impact of Open at 0.24 0.27 0.30
 L-inequality (1.22) (2.01) (2.48)
Marginal impact of inequality -1.72 *** -11.40 -22.19 **
 at mean Open (13.98) (1.34) (5.22)

The values in parentheses for the marginal impacts are F-statistics;
H-, M-, or L- indicates that the marginal impact is evaluated at the
high (one standard deviation above the mean) level of inequality,
at the mean, or at the low (one standard deviation below) level of
inequality, respectively. See footnote to Table 2

Received November 2004; accepted April 2006.

References

Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat, and R. Wacziarg. 2002. Fractionalization. NBER Working Paper No. 9411.

Beugelsdijk, S., H. L. F. de Groot, and A. B. T. M. van Schaik. 2004. Trust and economic growth: A robust analysis. Oxford Economic Papers 56:118-34.

Bhagwati, J. 2001. IMF Seminar. IMF Survey, April 2, 2001.

Bhagwati, J. 2004. In defense of globalization. New York: Oxford University Press.

Burt, R. S. 2001. Structural holes versus network closure as social capital. In Social capital: Theory and research, edited by N. Lin, K. Cook, and R. S. Burt. New York: Aldine De Gruyter.

Camerer, Colin F. 2003. Behavioral game theory: Experiments in strategic interaction. New York: Russell Sage Foundation.

Chenery, H. B., and L. Taylor. 1968. Development patterns: Among countries and over time. Review of Economics and Statistics 41:391-416.

Coleman, J. S. 1988. Social capital in the creation of human capital. American Journal of Sociology (Supplement) 94:95-120.

Dasgupta, P. 2000. Introduction. In Social capital: A multifaceted perspective, edited by P. Dasgupta and I. Serageldin. Washington DC: The World Bank, pp. 3-5.

De Grauwe, P., and M. Polan. 2005. Globalization and social spending. Pacific Economic Review 10:105-23.

Frankel, J. A., and D. Romer. 1999. Does trade cause growth? American Economic Review 89:379-99.

Freedom House Publisher. Annual Survey of Freedom Country Scores; various issues.

Fukuyama, F. 1995. Trust: Social virtues and the creation of prosperity. New York: New York Free Press.

Inglehart, R. 1994. Codebook for world values surveys. Ann Arbor, MI: Institute for Social Research.

Jackson, M. O., and A. Wolinsky. 1996. A strategic model of social and economic networks. Journal of Economic Theory 71:44-74.

Knack, S., and P. Keefer. 1997. Does social capital have an economic payoff? A cross-country investigation. Quarterly Journal of Economics 112:1252-88.

La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. W. Vishny. 1997. Trust in large organization. American Economic Review, Papers and Proceedings 87:333-8.

Lawrence, R. Z. 1991. Efficient or exclusionist? The import behavior of Japanese corporate groups. Brookings Papers on Economic Activity 11:311-31.

Lin, Nan. 2001. Social capital: A theory of social structure and action. New York: Cambridge University Press.

Olson, M. 1982. The rise and fall of nations: Economic growth, stagflation and social rigidities. New Haven and London: Yale University Press.

Paldam, M. 2000. Social capital: One or many? Definition and measurement. Journal of Economic Surveys 14:629-53.

Polanyi, Karl. 1944. The great transformation: The political and economic origins of our time. Boston: Beacon Press.

Putnam, R. D. 1993. Making democracy work: Civic traditions in modern Italy. Princeton, NJ: Princeton University Press.

Rodrik, D. 1997. Has globalization gone too far? Washington DC: Institute for International Economics.

Sachs, J., and A. Warner. 1995. Economic reform and the process of global integration. Brookings Papers on Economic Activity 1:1-118.

Schiff, M. 1998. Ethnic diversity and economic reform in sub-Sahara Africa. Journal of African Economies 7:348-62.

Schiff, M. 2002. Love thy neighbor: Trade, migration and social capital. European Journal of Political Economy 18:87-107.

Sen, A. 1999. Development as freedom. New York: Anchor Books, Random House.

Solow, R. M. 2000. Notes on social capital and economic performance. In Social capital: A multifaceted perspective, edited by P. Dasgupta and I. Serageldin. Washington DC: The World Bank, pp. 6-10.

Stiglitz, J. E. 2000. Formal and informal institutions. In Social capital: A multifaceted perspective, edited by P. Dasgupta and I. Serageldin. Washington DC: The World Bank, pp. 59-68.

World Development Indicators, World Bank, various issues.

Zak, P. J., and S. Knack. 2001. Trust and growth. Economic Journal 111:295-321.

(1) There are numerous difficulties with the traditional utility analysis on national social values. General speaking, social values should be an integral part of the utility function, often assumed to be completely autonomous. There is no room for change in social values within this framework. It is, therefore, difficult to trace how social values can be transmitted to or uprooted in a country. Moreover, traditional utility analysis is based on individual rationality while social values are based on collective rationality.

Coleman (1988, p. 17), proponent of this social capital framework, argues, "In this paper, I have attempted to introduce into social theory a concept, 'social capital,' paralleling the concepts of financial capital, physical capital, and human capital--but embodied in relations among persons. This is part of a theoretical strategy that involves use of the paradigm of rational action but without the assumption of atomistic elements stripped of social relationships."

(2) Skeptics question what should be counted as social capital and how social capital should be measured, let alone the scantiness of available data (see Solow in Dasgupta and Serageldin 2000). Proponents of social capital rebut the argument by saying that because social capital makes the economy more productive, it can be measured from its output; by adding market value to existing physical assets, the contribution (value) of social capital can be reckoned (see Stiglitz 2000, p. 60).

(3) For convenience, I will use the terms, trust and generalized trust, interchangeably in the paper. Generalized trust is the conventional term and is used in the World Value Surveys.

(4) In an interesting case study by Schiff (1998) on the ethnically divided sub-Saharan African societies, strong supportive evidence is provided for Rodrik's argument. Economic and trade reforms in these countries increase ethnic inequality and tension among ethnic groups. Since they have a negative impact on interethnic cooperation and political stability, more resources must be spent on security. Consequently, trade reforms can be welfare reducing.

(5) For example, Sen's argument may need qualification in the presence of network externalities. Take language as one element of culture. Indigenous language, even if it was deemed more useful, may have to give way to new foreign languages that may be less useful, but has enjoyed a large global network of users. I am indebted to a referee for pointing this out.

(6) To assure the network is stable, Jackson and Wolinsky argue that "one is forced to allocate resources to nodes that are not responsible for any of the production. We characterize one such allocation rule: the equal split rule, and another rule that arises naturally from bargaining of the players." (p. 44, 1996).

(7) The keiretsu in Japan provide an interesting illustrative example. Keiretsu is a form of social capital embedded in economic networks that overlap and pretty much span the entire Japanese economy. Critics argue that the keiretsu may have used anticompetitive business practices in reducing foreign imports and direct investments. Supporters rebut that the keiretsu only lowers the cost of doing business (see Lawrence 1991).

That fragmented social capital can be socially detrimental means that the concept of social capital currently used is not just a circular definition for successful economies because it can predict different outcomes.

(8) Note that Jackson and Wolinsky's concept of social network breakup is closely akin to Rodrik's concept of social disintegration.

(9) Can the home country adopt "unproductive" foreign values? Under our framework that social capital aids productive activities, this cannot happen. Adopting unproductive foreign social capital is unlikely because it would not survive competition with the productive domestic values. Perhaps without the intense global competitive pressure, the chance of adopting "unproductive" foreign social values could be higher than otherwise.

(10) Some countries are in one but not both of the surveys. Also, in the third survey, some countries complete the survey in 1997-1998 instead of 1995. To adjust for this, I match the data of the other regressors to the Survey dates.

(11) In the literature, in Chenery and Taylor (1968), log-GDP-per-capita and log-population are often used to capture the stages of social and economic development of nations. Some authors (see Schiff 2002) use the growth rate of GDP per capita as a proxy for institutional effectiveness.

(12) I have constructed the rank version of GDP-per-capita growth, where the growth rates are classified by their positions within one standard deviation. Growth rate that is below (above) one standard deviation takes a rank of 0 (3), and that between the sample mean and one standard deviation below (above) the mean takes the rank of 1 (2). The differences in the regression results between using the rank version and the value version of GDP-per-capita growth are marginal. The former gives a slightly higher t-statistic than the latter in most regressions.

(13) Without doubt, the crude qualitative dummy variable used in the construction of the Sachs Warner index reduces the precision of measurement and ultimately weakens the inference using this index. But on the positive side, the imprecision in the Sachs-Warner index also decreases the correlation between this index and the other regressors. The present data set has multicollinearity. Hence, there is a trade-off here. Although better indices increase the statistical significance of the estimation and improve inference, they may also increase multicollinearity, which reduces it.

Perhaps the Sachs Warner index can be viewed as a bottom line case. But, even without good precision regression results using the Sachs Warner index are reasonable, as shown in Table 2, and later in Table 4. This only lends credibility to the present results.

(14) Note that the ethnolinguistic fractionalization index captures the racial, linguistic, and cultural profile in a country. This index remains unchanged even when a country's racial and cultural tension has gone up from globalization. For this reason, it might not have captured fully the rise and fall of racial conflicts in a country. But on the positive side, this index has little correlation with other regressors. The multicollinearity problem could be less severe [see footnote 15 and the two standard least squares (SLS) regressions in Table 5].

(15) A plausible explanation for the weak t-statistics in the 2SLS regressions in Table 5 is the presence of multicollinearity in the data. Since the racial, linguistic, and cultural profile in a country are exogenous, the Ethnolinguistic Fractionalization index is unaffected by trade. Therefore, the correlation between OpenGM and Ethnolinguistic Fractionalization index is lower than the correlations between OpenGM and other measures of inequality. The low correlation could explain why the estimated marginal impact of openness, in regressions (N) and (O) are statistically more significant than those from regressions (J) to (M) in Table 5.

Kenneth S. Chan, Department of Economics and Finance, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon, Hong Kong; and the Department of Economics, McMaster University, Hamilton, Ontario, Canada; E-mail kschan@cityu. edu.hk.

Table 1. Summary Statistics

 Standard
Variable Mean Deviation Minimum Maximum

Generalized Trust 31.7546 16.03 4.11 65.21
OpenSW 22.45 15.60 0 45
Open 28.71 13.79 8.02 67.24
[Open.sub.-1] 48.50 25.45 12.63 119.53
Gini Coefficient 37.58 10.72 23.1 59.3
Income-share ratio 8.12 5.43 3.2 24.23
Ethnolinguistic
 Fractionalization 0.32 0.25 0.002 0.85
GDP per-capita Growth 2.32 4.27 -14.8 9.06
Income-per-capita Category 2.15 1.07 0 3

The data on Trust are taken from WVS (1990-1995). OpenSW is
the years of liberal trade policy taken from Sachs
and Warner (1995). Open ([Open.sub.-1],) is the contemporaneous
(past average, 1970-1989) export and import share of GDP.
The Income-per-capita Category (0, 1, 2, 3) represents the low,
lower middle, upper middle, and high income per capita
category, respectively. Income-share ratio equals the income
share of the highest 20% of population over the income
share of the lowest 20%. Ethnolinguistic Fractionalization
(from 0 to 1, where 1 is the most fractional society) is taken
from Alesina et al. (2002). The rest of the data are taken from
World Development Indicator, World Bank (various
years, at PPP).

Table 2. Dependent Variable: Generalized Trust

 (1) (2) (3) (4)

OpenSW 0.69 *** 0.71 ***
 (5.77) (3.33)
Open

[Open.sub.-1]

GDP-per-capita 1.17 ***
 Growth (2.99)
Income-per-capita 8.38 *** -0.37
 Category (4.01) (0.10)
R-square 0.1 0.1 0.31 0.44 0.44

 (5) (6) (7) (8)

OpenSW 0.64 ***
 (2.04)
Open 0.38 ** 0.24 0.29
 (2.23) (1.56) (1.57)
[Open.sub.-1]

GDP-per-capita 0.703 0.83 *
 Growth (1.34) (1.75)
Income-per-capita 7.65 ***
 Category (3.67)
R-square 0.1 0.47 0.11 0.35 0.15

 (9) (10) (11)

OpenSW

Open

[Open.sub.-1] 0.31 *** 0.18 ** 0.27 ***
 (3.32) (2.09) (2.74)
GDP-per-capita 0.72
 Growth (1.49)
Income-per-capita 6.36 ***
 Category (3.00)
R-square 0.1 0.24 0.38 0.27

The t-statistics are in parentheses. Regressions are OLS estimation
and are adjusted for heteroscedasticity. Sample size is 38. Values
in parentheses are t-statistics. *, **. and *** are the 10, 5, and
1% level of significance. respectively. The subscript -1 indicates
that the data for the [Open.sub.-1] variable is the average of
export and import share of GDP from 1970 to 1989.

Table 3. Dependent Variable: Generalized Trust

 (12) (13) (14) (15)

Gini Coefficient -1.03 *** -0.83 *** -0.97 ***
 (6.04) (4.96) (5.14)
Income-share
 Ratio -1.76 ***
 (5.50)
Ethnolinguistic
Fractionalization
GDP-per-capita
 Growth 0.50
 (1.21)
Income-per-capita
 Category 4.50 **
 (2.45)
R-square 0.47 0.54 0.49 0.35

 (16) (17) (18)

Gini Coefficient

Income-share
 Ratio -1.40 *** -1.62 ***
 (5.16) (4.77)
Ethnolinguistic -26.5 ***
Fractionalization (2.84)
GDP-per-capita
 Growth 0.75 *
 (1.70)
Income-per-capita
 Category 6.12 ***
 (3.45)
R-square 0.5 0.39 0.17

 (19) (20)

Gini Coefficient

Income-share
 Ratio

Ethnolinguistic -11.44 -22.78 **
Fractionalization (1.25) (2.24)
GDP-per-capita
 Growth 0.76 *
 (1.71)
Income-per-capita
 Category 7.02 ***
 (3.01)
R-square 0.33 0.21

OILS estimation. See Notes

Table 4. Dependent Variable: Generalized Trust, OLS Estimation

 (A) (B) (C)

OpenSW 1.26 ** 1.20 * 1.22 **
 (2.142) (1.839) (2.12)
Gini Coefficient -0.25 -0.28 -0.22
 (0.90) -0.92 (0.80)
OpenSW x Gini Coefficient -0.02 -0.02 -0.02
 (1.50) -1.44 (1.51)
Income-Share Ratio

OpenSW x Income-Share Ratio

Ethnolinguistic
 Fractionalization

OpenSW x Ethnolinguistic
 Fractionalization 0.42
Income-per-capita Category
 -0.134
GDP-per-capita Growth 0.43
 (1.01)
R-square 0.57 0.57 0.59
Marginal Impact of 0.10 0.09 0.11
 OpenSW at H-Inequality (0.19) -0.09 (0.23)
Marginal Impact of 0.36 ** 0.03 0.344 **
 OpenSW at M-Inequality (6.91) -1.96 (7.21)
Marginal Impact of 0.62 *** 0.58 * 0.60 ***
 OpenSW at L-Inequality (9.79) -3.83 (9.21)
Marginal Impact of -0.79 *** -0.79 *** -0.74 ***
 Inequality at Mean (10.12) -9.69 (9.04)
 OpenSW

 (D) (E) (F)

OpenSW 0.91 *** 0.81 ** 0.87 ***
 (3.24) (2.34) (3.15)
Gini Coefficient

OpenSW x Gini Coefficient

Income-Share Ratio -0.32 -0.41 -0.29
 (0.71) (0.87) (0.67)
OpenSW x Income-Share Ratio -0.07 * -0.06 * -0.06 *
 (1.73) (1.69) (1.82)
Ethnolinguistic
 Fractionalization

OpenSW x Ethnolinguistic
 Fractionalization 1.10
Income-per-capita Category (0.34)

GDP-per-capita Growth 0.51
 (1.09)
R-square 0.55 0.55 0.57
Marginal Impact of 0.01 -0.04 0.02
 OpenSW at H-Inequality (0.00) (0.01) (0.00)
Marginal Impact of 0.37 ** 0.31 0.36 ***
 OpenSW at M-Inequality (6.92) (1.46) (7.66)
Marginal Impact of 0.73 *** 0.64 ** 0.70 ***
 OpenSW at L-Inequality (14.11) (5.09) (12.54)
Marginal Impact of -1.81 *** -1.81 *** -1.69 ***
 Inequality at Mean (7.53) (7.40) (8.27)
 OpenSW

 (G) (H) (I)

OpenSW 0.84 *** 0.94 *** 0.80 ***
 (3.53) (3.51) (3.07)
Gini Coefficient

OpenSW x Gini Coefficient

Income-Share Ratio

OpenSW x Income-Share Ratio

Ethnolinguistic
 Fractionalization 2.42 1.27 3.71
 (0.27) (0.14) (0.44)
 -0.56 -0.57 (0.49)
OpenSW x Ethnolinguistic (1.14) (1.17) (0.98)
 Fractionalization -1.81
Income-per-capita Category (0.54)

GDP-per-capita Growth 0.55
 (1.00)
R-square 0.48 0.48 0.5
Marginal Impact of 0.52 *** 0.62 *** 0.52 ***
 OpenSW at H-Inequality (11.43) (6.96) (12.97)
Marginal Impact of 0.66 *** 0.76 *** 0.64 ***
 OpenSW at M-Inequality (22.99) (13.37) (19.53)
Marginal Impact of 0.80 *** 0.90 *** 0.77 ***
 OpenSW at L-Inequality (14.51) (13.22) (11.01)
Marginal Impact of -10.15 -11.52 -7.38
 Inequality at Mean (1.43) (1.70) (0.79)
 OpenSW

The value in parentheses for the marginal impacts are F-statistics;
H-, M-, or L- indicate that the marginal impact was evaluated at a
high level of inequality (one standard deviation above the mean
level of inequality), at the mean, or at a low (one standard
deviation below) level of inequality, respectively.
See footnote to Table 2.

Table 5. Dependent Variable: Generalized Trust--2SLS Estimation

 (J) (K) (L)

OpenGM -0.46 -0.02 0.07
 (0.64) (0.03) (0.16)
Gini Coefficient -1.18 -1.01
 (2.38) ** (1.82) *
OpenGM x Gini Coefficient 0.02 0.00
 (0.72) (0.16)
Income-Share Ratio -1.42
 (1.16)
OpenGM x Income-Share 0.00
 Ratio (0.07)
Ethnolinguistic
 Fractionalization
OpenGM X Ethnolinguistic
 Fractionalization
Income-Per-Capita Category 4.67 5.91
 (2.39) ** (3.11) ***
GDP per-capita Growth 0.47
 -1.12
R-Square 0.55 0.49 0.50
Marginal impact of 0.29 0.16 0.12
 OpenGM at H-inequality (0.53) (0.15) (0.06)
Marginal impact of 0.13 0.12 0.10
 OpenGM at M-inequality (0.32) (0.23) (0.19)
Marginal impact of -0.04 0.08 0.08
 OpenGM at L-inequality (0.04) (0.09) (0.08)
Marginal impact of -0.73 -0.91 -1.3
 inequality at mean (9.08) *** (13.64) *** (3.83) *
 OpenGM

 (M) (N) (O)

OpenGM 0.52 0.74 0.99
 (1.09) (2.33) ** (2.63) **
Gini Coefficient

OpenGM x Gini Coefficient

Income-Share Ratio -0.2
 (0.15)
OpenGM x Income-Share -0.06
 Ratio (0.92)
Ethnolinguistic 21.50 24.36
 Fractionalization (0.97) (0.91)
OpenGM X Ethnolinguistic -1.16 -1.15
 Fractionalization (1.76) * (1.86) *
Income-Per-Capita Category 5.67
 (2.25) **
GDP per-capita Growth 0.70 0.80
 -1.64 (1.96) *
R-Square 0.42 0.41 0.35
Marginal impact of -0.29 0.08 0.133
 OpenGM at H-inequality (0.30) (0.16) (0.41)
Marginal impact of 0.04 0.37 0.51
 OpenGM at M-inequality (0.02) (4.41) ** (7.52) ***
Marginal impact of 0.36 0.66 0.88
 OpenGM at L-inequality (1.15) (5.59)** (7.33) **
Marginal impact of -1.91 -11.83 -18.65
 inequality at mean (7.06) ** -1.72 (4.29) **
 OpenGM

OpenGM is the estimated (predicted) value of Open from regressing
the actual trade share on the constructed trade share from geographic
parameters, the gravity model (from Frankel and Romer). The values in
parentheses for the marginal impacts are F-statistics; H-M-, or
L-indicates that the marginal impact is evaluated at the high (one
standard deviation above the mean) level of inequality, at the mean,
or at the low (one standard deviation below) level of inequality
respectively. See footnote to Table 2.

Table 6. Dependent Variable: Generalized Trust--OLS Estimation

 (P) (Q) (R)

[Open.sub.-1] -0.11 -0.3 0.07
 (0.36) (0.10) (0.56)
Gini Coefficient -1.02 *** -1.06 ***
 (2.90) (2.69)
[Open.sub.-1] x Gini 0.01 0.01
 Coefficient (0.73) (0.53)
Income-Share Ratio -1.50 ***
 (3.17)
[Open.sub.-1] x Income-Share 0.01
 Ratio (0.46)
Ethnolinguistic
 Fractionalization
[Open.sub.-1] x Ethnolinguistic
 Fractionalization
Income-per- Capita Category 3.85 * 5.19 ***
 (1.98) (2.80)
GDP-per-Capita Growth 0.32
 (0.64)
R-square 0.56 0.52 0.53
Marginal impact of 0.18 0.20 0.16
 [Open.sub.-1] at H-inequality (1.27) (1.81) (2.10)
Marginal impact of 1.11 0.15 0.12
 [Open.sub.-1] at M-inequality (1.84) (2.58) (2.45)
Marginal impact of 0.05 0.10 0.09
 [Open.sub.-1] at L-inequality (0.26) (0.72) (0.73)
Marginal impact of inequality -0.74 *** -0.84 *** -1.21 ***
 at mean [Open.sub.-1] (17.03) (19.72) (12.56)

 (S) (T) (U)

[Open.sub.-1] 0.19 0.32 *** 0.37 ***
 (1.20) (2.91) (3.03)
Gini Coefficient

[Open.sub.-1] x Gini
 Coefficient
Income-Share Ratio -1.30 **
 (2.33)
[Open.sub.-1] x Income-Share -0.002
 Ratio (0.15)
Ethnolinguistic 11.52 0.80
 Fractionalization (0.64) (0.05)
[Open.sub.-1] x Ethnolinguistic -0.47 * -0.4
 Fractionalization (1.65) (1.35)
Income-per- Capita Category 5.40 *
 (1.97)
GDP-per-Capita Growth 0.512 0.48
 (1.00) (1.00)
R-square 0.45 0.42 0.36
Marginal impact of 0.15 0.06 0.14
 [Open.sub.-1] at H-inequality (1.29) (0.31) (1.91)
Marginal impact of 0.17* 0.17** 0.24 ***
 [Open.sub.-1] at M-inequality (3.31) (6.26) (10.98)
Marginal impact of 0.18 0.29 *** 0.34 ***
 [Open.sub.-1] at L-inequality (2.17) (9.12) (10.34)
Marginal impact of inequality -1.41 *** -11.14 -18.48 **
 at mean [Open.sub.-1] (11.29) (1.22) -4.17

[Open.sub.-1] is the average exports and imports share of GDP from
1970 to 1989. The values in parentheses for the marginal impacts are
F-statistics; H-, M-, or L-indicates that the marginal impact is
evaluated at the high (one standard deviation above the mean) level
of inequality, at the mean, or at the low (one standard deviation
below) level of inequality, respectively. See footnote to Table 2.

Table 7. Testing for Omitted Variables. Dependent Variable:
Generalized Trust-OLS Estimation

OpenSW 1.25 ** 1.23 ** 1.14 ** 1.08 *
 (2.06) (2.07) (2.02) (1.81)
Gini Coefficient -0.29 -0.31 -0.29 -0.24
 (0.84) (1.09) (1.07) (0.85)
OpenSW x Gini -0.02 -0.02 -0.02 -0.02
 Coefficient (1.38) (1.42) (0.60) (1.47)
Income-Share Ratio

OpenSW x Income-
 Share Ratio
Ethnolinguistic
 Fractionalization
OpenSW x
 Ethnolinguistic
 Fractionalization
Regional dummies YES
Oil exporting 4.93
 country dummy (0.89)
Log-population -1.02
 (0.60)
Democracy -2.56
 (1.24)
Religious
 fractionalization
R-square 0.60 0.58 0.58 0.59

OpenSW 1.20 ** 0.90 *** 0.93 *** 0.84 ***
 (2.04) (2.94) (3.16) (3.31)
Gini Coefficient -0.31
 (1.09)
OpenSW x Gini -0.24
 Coefficient (1.50)
Income-Share Ratio -0.35 -0.34 -0.36
 (0.72) (0.76) (0.83)
OpenSW x Income- -0.07 -0.07 -0.06
 Share Ratio (1.63) (1.69) (1.64)
Ethnolinguistic
 Fractionalization
OpenSW x
 Ethnolinguistic
 Fractionalization
Regional dummies YES
Oil exporting 4.14
 country dummy (0.80)
Log-population -1.16
 (0.68)
Democracy

Religious 7.17
 fractionalization (1.02)
R-square 0.58 0.57 0.55 0.56

OpenSW 0.71 ** 0.97 *** 0.64 ** 0.84 ***
 (2.44) (2.98) (2.34) (3.48)
Gini Coefficient

OpenSW x Gini
 Coefficient
Income-Share Ratio -0.31 -0.41
 (0.68) (0.96)
OpenSW x Income- -0.07 -0.07
 Share Ratio (1.76) (1.71)
Ethnolinguistic
 Fractionalization -3.67 1.28
OpenSW x (0.44) (0.13)
 Ethnolinguistic -0.35 -0.53
 Fractionalization (0.72) (1.06)
Regional dummies YES
Oil exporting 1.65
 country dummy (0.25)
Log-population

Democracy -2.96
 (1.35)
Religious 7.03
 fractionalization (0.94)
R-square 0.57 0.56 0.52 0.48

OpenSW 0.81 *** 0.74 *** 0.80 ***
 (3.60) (2.74) (2.79)
Gini Coefficient

OpenSW x Gini
 Coefficient
Income-Share Ratio

OpenSW x Income-
 Share Ratio
Ethnolinguistic
 Fractionalization 3.37 18.15 -0.33
OpenSW x (0.33) (1.18) (0.03)
 Ethnolinguistic -0.55 -0.9 -0.51
 Fractionalization (1.10) (1.54) (0.95)
Regional dummies
Oil exporting
 country dummy
Log-population -1.38
 (0.71)
Democracy -4.55
 (1.47)
Religious 4.44
 fractionalization (0.51)
R-square 0.49 0.50 0.48

See footnote to Table 2.