Methodology

As it has been already outlined before, the main aim of this study is to find out whether there is a significant influence of national culture on trade openness and if so, which dimensions of culture enhance or weaken the country´s preference to be open. The regression analysis allows a proper examination of these relationships. Nevertheless, to obtain reliable results, the regression analysis must use an accurate estimator. The cross-sectional character of the cultural data indicates that the Ordinary least square method (OLS) is the most appropriate estimator. The OLS estimates minimize the sum of squared residuals nevertheless, to achieve them the Gauss-Markov assumptions must be met. In the case of our model, there might be an issue with the exogeneity condition. This assumption secures that the exogenous variables are uncorrelated with the error term, E(ui|Xi) = 0 (Wooldridge 2002). The main possible origin of the endogeneity in this model is the simultaneity bias. I assume that the culture affects trade openness nevertheless, the reverse influence is possible as well.

To deal with the potential simultaneity bias I will use the instrumental variables (IV) for the cultural dimensions. More specifically, just three out of six cultural dimensions will be instrumented, because for the rest of the dimensions I was unable to find appropriate instruments. However, the use of at least these three instrumental variables will allow me to directly test the endogeneity by a Hausman test¹². The result of the Hausman test will

12 The Hausman test examines the significance of the distance between OLS and 2SLS estimation. If there is a significant difference between these estimates, the exogeneity assumption does not hold. In that case, the 2SLS estimate is preferred. Otherwise, the OLS estimator should be used because without endogeneity issue the OLS estimator gives consistent and efficient estimates, while in the case of 2SLS the standard errors are quite large (Wooldridge 2002).

50

show whether the simultaneity bias can be an issue in the regression analysis and if yes then it will be necessary to also find the rest of the IVs. All in all, I will use two estimators in the regression analysis, the OLS estimator and the Two-stage least square estimator (2SLS) that is used in the case of IV variable implementation and the Hausman test will decide which of the estimators gives a more precise estimation.

The good IV must be correlated with the related explanatory variable and it can affect the dependent variable just through its effect on this explanatory variable. In other words, the IV must satisfy the relevance and exclusion conditions. Based on the results, after examination of the relevance and exclusion condition, I chose the following instruments:

• The pronoun drop as an IV for individualism vs collectivism dimension. Kashima &

Kashima (1998) found out that the language rules tend to reflect the cultural aspects.

More specifically, they found out that the pronoun drop in language reflects the relationship between individual and group. If the pronoun drop is not possible, such as in English, the speakers emphasize the subject, and that leads to more individualistic tendencies in the country. If the language allows the omission of subject pronouns, the emphasis on a specific person is removed, and the society tends to be more collectivistic (Nash & Patel 2018).

For the use in the analysis, I constructed this instrumental variable as a dummy variable where the value 1 corresponded to languages allowing the pronoun drop, and value 0 to those that do not allow it. The respective data was retrieved from Kashima

& Kashima (1998) and Spencer & Luís (2012). The relevance test, with the consideration of all control variables, confirmed that the pronoun drop significantly affects the individualism vs collectivism dimension. The exclusion condition is hardly testable, nevertheless, I have not found any empirical or theoretical indication that would suggest other influence of pronoun drop on trade openness than its effect through individualism vs collectivism. Based on that, I consider the pronoun drop IV to be a suitable instrument for the individualism vs collectivism dimension.

• The use of multiple second-person pronouns as IV for uncertainty avoidance. The article of Kashima & Kashima (1998) was also an inspiration for the choice of second instrument. They found out, that the use of multiple second-person pronouns is highly

51

correlated with the level of uncertainty avoidance in a country. The authors argue that the individuals using language which distinguishes multiple second-person pronouns face constant decisional stress because they must correctly choose the appropriate pronoun based on the person´s social status. This leads to higher insecurity of an individual and his/her greater uncertainty avoidance (Nash & Patel 2018).

The IV is again introduced to the model as a dummy variable. The value 1 is assigned to countries whose language has multiple second-person pronouns and 0 for those that do not. The data was obtained from Kashima & Kashima (1998). The relevance test confirmed the significant influence of multiple second-person pronouns on the country´s level of uncertainty avoidance. Regarding the exclusion condition, similarly as before, I have not found any empirical or theoretical indicators that would suggest a direct impact of multiple second-person pronouns use on trade openness.

• The last IV used for the empirical analysis is ethnic fractionalization. Based on the research of Siegel et al. (2013), ethic fractionalization is highly correlated with the level of egalitarianism in a country, (Nash & Patel 2018). Societies with greater ethnic fractionalization tend to be less altruistic, they perceive a higher inequality in a country which is often related to social conflicts. In other words, the higher ethnic fractionalization correlates with high values of power distance (Nash & Patel 2018).

The data for the ethnic fractionalization index were retrieved from Fearon (2003). The benefit of this measure is that despite other possible definitions of ethnicity, he works with the ethnic categories as Black, White, Asian, etc. This form of defining the ethnicity is important because it stresses its predetermined character that cannot be altered by a free will of individual and strengthens the IV relevance in the context of exclusion condition. Regarding the relevance condition, the ethnic fractionalization index has an impact on the power distance dimension, nevertheless, this impact is significant only on a 10% level of significance what makes it a weaker instrument than the previous two IVs.

In addition to the possible endogeneity issue, another methodological aspect has to be taken into the account. Even though the cultural data have a cross-sectional character, the macroeconomic variables are rather of panel data character. It means that for accurate observation of the development of macroeconomic variables used in a model it is

52

convenient to observe their cross-country differences and their evolution in time. Without time consideration, the results obtained from regression analysis could be biased by some local crisis or a business cycle. The ideal solution would be to use a panel data estimator however, the character of cultural data does not allow for that. Since the endogeneity issue cannot be excluded prior to the analysis, the possible estimators that could be used require a time variance of the regressors. The fixed effects, as the cultural impact, would be cancelled out during the estimation. To solve this problem, I repeat the regression analysis for each of the years from the periods 1980-1989 and 2010-2019. If the significance of the exogenous variables is consistent through the years, I present just the results of regression analysis based on the average data of these periods. By this procedure, I should be able to control for possible time bias and identify potential changes in direction and strength of culture effect.