Is International Trade Relevant to Social Trust Formation? Evidence from Cross-country Analysis

ABSTRACT

In this paper, we investigate whether international trade itself can contribute to the level of generalized trust. We extend the existing empirical research in several ways. First, we use a larger sample size, we test and reject the treatment of international trade as an exogenous variable, and we address trade endogeneity using instrumental variables estimators. Second, we use geographical variables and international trade prices to instrument for international trade. Third, we perform instrumental variables diagnostics tests to determine the suitability and relevance of our instruments; we also perform tests of the statistical significance of our parameter of interest that are robust to the presence of weak instruments. Our empirical analysis suggests that international trade does have a significant and relatively large positive effect on social trust and reconfirms the role played by other variables like income inequality in the formation of trust found in the literature.

KEYWORDS:

• Social trust
• international trade
• empirical

JEL CLASSIFICATIONS:

• Z10
• Z13
• F10

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 An interesting paper on whether culture affects economic outcomes is Guiso, Sapienza, and Zingales (2006).

2 The relationship between trade openness and growth is a highly debated topic in the growth literature. Yanikkaya (2003) is a good survey of this literature.

3 See also Antoci, Sabatini, and Sodini (2009), who develop a theoretical model to show how social participation through economic exchange helps generate social capital.

4 While trade today is based on global value (or production) chains it also could lead to value chains within countries or can lead to agglomeration as in the coastal regions of China, which in turn can increase or decrease trust. For example, increase in trust can occur because agglomeration of production and assembly might lead people to learn to work with each other through greater interactions with each other. Alternatively, they might feel they are not getting their fair share in the gains from trade, and trust may decline as a result. We owe this comment to an anonymous referee.

5 Hirschman (1982) provides a historical and comparative analysis of the different views on the social consequences of a market and global economy. Chan (2007) provides a nice recount of the ‘optimistic’ (Sen, 1999; Bhagwati, 2004) and ‘pessimistic’ (Rodrik, 1997) views on the effect of globalization on social capital. On globalization and income inequality, see Feenstra and Hanson (1996) and the references therein.

6 Guiso et al. (2009) and Yu et al. (2011) acknowledge reverse causality. They indicate that if trust affects trade, it is also equally possible that trade breeds trust.

7 They use two measures of informal trust based on data from the Eurobarometer 1996 survey. See their paper for details on these indicators.

8 The specific question is ‘Generally speaking, would you say that most people can be trusted or that you cannot be too careful in dealing with people?’ This metric is referred to as generalized trust.

9 Research on the sources of social trust has quickened over the past two decades. The earliest research in this area identified several key variables (such as income inequality) that affect the level of generalized trust at the national level. Research that is more recent has extended this analysis to test the importance of other factors that could also play a role in its generation and sustainability. For example, Wang and Gordon (2011) find evidence that social institutions (such as a well-developed legal system) help promote generalized trust. Other references on the sources of generalized trust are Arrow (2000), Glaeser et al. (2000), Putnam (1995), Alesina and La Ferrara (2002), Uslaner (2003), Bjornskov (2007), Delhey and Newton (2005), Brandt (2008), Fischer (2008), Guiso, Sapienza, and Zingales (2008a). See also Nannestad (2008) who provides a nice survey of prior work in this area.

10 Bjornskov (2007) finds this variable to be statistically insignificant.

11 See Delhey and Newton (2005), who conclude that the exclusion of the Nordic countries in an empirical equation of trust does not drive the statistical significance of other factors that have been found to affect trust (among them income inequality, wealth, protestant religious traditions and ethnic fractionalization).

12 The literature on economic outcomes and social capital argues that the connection between these two variables is from trust to GDP. Other authors argue for mutual determination. To avoid issues of endogeneity, we use GDP data from 1995.

13 Guiso et al. (2008a) estimate the effect of bilateral trust on bilateral trade using genetic distances of the population to instrument for a country’s level of trust. Giuliano, Spilimbergo, and Giovanni Tonon (2006) claim that genetic distance is just a proxy for transportation costs, which are not fully captured by the log distance between two countries. In their view, trust might simply be the result of international trade, with little or no effect from cultural characteristics.

14 Mean distance to trading partners’ for say, country i, averages the distance of country i to each of its trading partners in 2000. We use a non-weighted average. Trading partners are defined as countries for which country i had positive trade flows in 2000, which corresponds to the year of our trade data.

15 See Disdier and Head (2008) who survey the literature on the effect of distance on international trade. They find that the distance effect on international trade is stable over time, immune to econometric methodologies and different sample sizes. They conclude that the distance effect is also immune to the inclusion of different covariates. Campbell (2010) shows that trade is persistent and this explains somewhat the reasons why the ‘significance of distance in the gravity equation has not decreased apace with declines in transport costs.’

16 Frankel and Romer (1999) used geographical variables to obtain an exogenous component of bilateral international trade that they used to identify the effect of trade on income. Rodriguez and Rodrik (2000) found that Frankel and Romers’ results were not robust to the inclusion of additional geographical variables in their main regression. Consequently, we test the robustness of our results with the addition of regional dummy variables, distance to the equator (a geographical variable often used in cross-country regressions), and a landlocked dummy variable to (1). See also Frankel and Rose (2002) who use the same instruments as Frankel and Romer (1999).

17 Other related studies find that the structure of the population explains variance on trust. For example, some empirical studies find that older people are more trusting (see Putnam (1995), Glaeser et al. (2000), Alesina and La Ferrara (2002) and Berggren and Jordahl (2006)).

18 We also perform tests of constant error variance and do not reject the null hypothesis of homoscedasticity across all our specifications. However, we also estimate all our models and perform diagnostics IV inference that are robust to the presence of heteroscedasticity. Specifically, we estimate the models using an efficient Generalized Moments estimator and perform tests of the over-identifying restrictions that are robust to heteroscedasticity of unknown form. However, in the context of robust first-step regression, the Cragg-Donald F-test statistic used to test for weak instruments is only valid under homoscedasticity. These results are available upon request (see also footnote 20 below).

19 Data on generalized trust and religiosity is from Christian Bjornskov. He collected the trust scores from the 1997 and 1999–2001 waves of the World Values Survey. We complemented this data with generalized trust data from the AsianBarometer, AfroBarometer, and LatinoBarómetro surveys for countries not available in Christian Bjornskov’s trust data set. The data on population, monarchy, post-communist states and the shares of the population affiliated to different religions come from the CIA’s World Factbook (also the source for nation population). The data on fractionalization comes from the MacroDataGuide. Income inequality is from Deininger and Squire (1996). The data on international trade is from Head et al. (2010). Data on terms of trade are from (Gaulier et al., 2008) and data on mean distance to trade partners is based on the distance between capital cities of the countries and corresponding trading partners. We thank Christian Bjornskov, who kindly provided us with the trust and religiosity data (Bjornskov, 2007; Bjornskov & Berggren, 2011). Appendix I provides more details of the sources of our data and the description of the covariates used in the analysis.

20 We have a sample of 127 countries, but terms of trade data are only available for 122 countries.

21 In all the specifications presented in Table 2, we do not reject a constant error variance, so |t-values| reported in Table 2 are computed using standard errors that are non-robust to heteroscedasticity. However, we also estimated the models presented in Table 2 with robust to heteroscedasticity standard errors. The robust to heteroscedasticity |t-values| for the trade coefficient were (2.35)***, (1.82)*, (0.96) and (2.57)*** for model I throughout IV, respectively. Clearly, Table 2 shows that the addition of the regional dummy variables does not affect the OLS trade coefficient value but does decrease its statistical significance. On the other hand, the addition of the internal distance variable decreases its value and its statistical significance. It is worth noting that using robust standard errors does not make these additional covariates statistically significant determinants of trust. A table (Table 2(a)) with robust standard errors, not reported in the paper, is available upon request.

22 Under conditional homoscedasticity, these endogeneity test statistics are numerically equal to a Hausman test statistic; see Hayashi (2000).

23 In Table 6, we only report the coefficient estimates of the excluded instruments, which are the coefficients of interest from the first-step regression. However, we will make all the coefficient values estimates from this step available upon request.

24 We also estimated the models using an efficient two-step Generalized Method of Moments estimator, which produces estimates that are robust to heteroscedasticity, but are not robust to the presence of weak instruments. Our findings with this estimator are very similar to those obtained with the 2SLS and LIML estimators. We will make the GMM estimates available upon request.

25 A test of the joint statistical significance of the regional dummy variables in the context of the LIML estimator for model II indicates that they do not have an effect on trust. The value of the test statistic is 1.22 with a p-value of 0.8754. Even though these regional variables turned out to be insignificant, it is important to check the robustness of the results to their inclusion in our models because these variables are proxies for cultural proximity, and cultural proximity can lead to more trade. Chan (2007) also adds regional dummy variables to his regressions of trust on international trade.

26 Alternatively, we estimated models dropping these variables one at a time. In most of the estimations, the trade coefficient was either very similar in value or a bit larger and continued to be statistically significant at the 5% level.

27 Interestingly, Guiso et al. (2009) report a four-fold increase in the size of the IV-GMM coefficient, relative to the OLS estimates, of the effect of bilateral trust on bilateral trade for a sample of European countries.

28 Weak instruments tests use the Cragg-Donald F-test statistic. When there is one endogenous variable this statistic is the usual F-statistic for the test of excluded instruments from the first-stage regression. The test compares this statistic to the Stock and Yogo (2005) critical values, which depend on the estimator used. Critical values are available for maximal relative bias of the estimator (relative to the bias of the OLS) and maximal rejection rate for a given test size of, say 5%. Stata, the software we use, provides the Stock and Yogo (2005) critical values for the 2SLS, the LIML and Fuller’s modified LIML estimators. Here we use the critical values for 10% maximal LIML size (6.46) and the critical values for 5% maximal Fuller relative bias (9.61). Critical values for maximal relative bias are not available for the standard LIML estimator because the LIML estimator does not have a finite expected value so the concept of bias does not apply. See also Staiger and Stock (1997), Stock, Wright, and Yogo (2002) and Baum, Shaffer, and Stillman (2007) and the references therein.

29 We thank an anonymous referee for bringing to our attention the work of Rodriguez and Rodrik (2000), Nunn and Puga (2012) and Nunn and Wantchekon (2011).

30 Empirical cross-country studies of economic growth often include this variable in their set of covariates.

31 We also used within country distance in the set of excluded instruments. Results were essentially the same as those obtained with country area. This is not surprising since these two variables are highly correlated in our sample.

32 We apply the difference-in-Sargan test of orthogonality, also known as the C-statistic. Because we have two surplus over-identifying restrictions, we are able to check for the orthogonality of one instrument at a time, assuming that at least one of the other two remaining instruments is orthogonal. Table 8 reports these tests results. As the table shows, all three of our instruments pass these orthogonality tests. These orthogonality tests results are also robust to the inclusion/exclusion of different covariates and to the assumption of the presence of weak instruments. For a discussion of the Sargan tests of orthogonality, see Hayashi (2000).

33 We cannot identify the effect of trade on trust in the new version of model III due to the high collinearity between within-country distance and area.

34 The Anderson Rubin Wald and the Stock-Wright tests turned out statistically insignificant at the conventional 5% or 10% levels.

35 Results from these robustness check are available upon request.

36 Bottazzi et al. (2016) examines the effect of generalized trust on direct foreign investment flows, without a full consideration of the possible reverse causality.

Additional information

Notes on contributors

Robert Riley

Rob Riley is a Full Professor at the Economics Department, University of St. Thomas, St. Paul, MN. Currently he is serving as the Vice Provost for Academic Affairs at the University of St. Thomas.

Luz Saavedra

Luz Saavedra is an Associate Professor at the Economics Department, University of St. Thomas, St. Paul, MN.