«Abstract We examine the effects of precolonial and colonial legacies on the current economic growth rates of ex-colonies. We ﬁnd that precolonial ...»
From the ﬁgures, we do not observe a particular outlier that dominates the other observations and leads to the reported estimation results. By this, we infer that current economic growth needs to be linked to prehistoric legacies.19 We compute the counterfactual predictions for the economic growth using the estimations in Table 7. We focus on the Korean economic growth rate and use Model 4 to examine the effect of the state antiquity index AD 1950 on economic growth. According to the estimation results, the predicted growth rate is 5.75%, which is almost identical to the realized growth rate of 5.78%. We now maintain the values of the other variables and change the state antiquity index value AD 1950 of Korea from 8.18 to the average of the index values, namely 3.20. Form this, the modiﬁed growth rate is obtained as 3.66%.20 This is about 64% of the predicted and realized economic growth rate, implying that the state antiquity index AD 1950 explains about 1/3 of the Korean economic growth rate from 1960 to 2000. If we predict the economic growth rate by supposing that Korea had not been colonized, the economic growth rate is predicted as 5.62%.21 The growth rate reduces by 0.13 percentage points. That is, the portion contributed by the Japanese colonial government to the Korean economic growth rate is about 2.2% (= 0.13/5.75).
4.2 Robustness Tests using Geographical, Ethnic, and Religious Variables
We now examine how geographical, ethnic, and the Sub-Saharan Africa dummy variables affect the estimations. According to Sachs and Warner (1997) and Gallup, Sachs, and Mellinger (1999), geographical variables such as temperature, endemic, soil, and transportation system affect economic growth through health, productivity, and transaction costs. Furthermore, they argue that a low economic growth rate is mainly due to these geographical issues. In particular, Bloom and Sachs (1998) claim that malaria critically affects the economic growth rate, and that the Sub-Saharan African countries lose an economic growth rate of 1.3% every year from malaria. On the other hand, Easterly and Levine (1997) focus on the effect of ethnic diversity on economic growth and argue that ethnic fragmentation is the main cause of the low economic growth in the African countries. We accommodate these claims by including the malaria ecology index, landlocked country dummy, and ethnic diversity index variables in the original models.
Insert Table 8 around here.
19 Readers can refer to Spolaore and Wacziarg (2013) for a survey on this topic.
20 This is obtained as 5.75 + (3.20 − 8.18) × 0.42.
21 This is obtained as 5.75 + 0.46 × 0.35 − 0.49 × 0.59, where 5.75 is the predicted economic growth rate by Model 4 in Table 7, 0.35 is the colonial duration/100, and 0.59 is the log(immigrant ratio). The others are estimated coefﬁcients.
18 The estimation results are contained in Models 1 and 2 of Table 8. We ﬁnd no signiﬁcant relationship between these variables and economic growth. Although we do not report the estimation results for the sake of brevity, we also estimated the models by including other geographical variables such as the share of a country’s population in temperature ecozones, average temperature, average number of frost days per month in winter, latitude, the portion of the country’s land area within 100 km of the coast, the portion of the population residing in the area within 100 km of the coast, and so on. None of these variables were signiﬁcant and, thus, did not change the previous model estimation results qualitatively.
As another factor of economic growth, we include a religion variable to accommodate Max Weber’s hypothesis that religions have a close relationship with economic growth (see Barro and McCleary, 2003).
In Model 3 of Table 8, we include the portions of Catholics, Protestants, and Muslims in each country and estimate their effects on the economic growth rate. The religion variable is signiﬁcant at about the 9% level, but this does not change our previous model estimation results.
We show our model estimation results in Model 4 of Table 8 in terms of the Sub-Saharan African countries. Barro (1991) reports that the Sub-Saharan Africa dummy is statistically signiﬁcant in explaining the economic growth rate, and since then, researchers have attempted to identify the roles of this variable. Englebert (2000) and Block (2001) report that the average economic growth rate of the Sub-Saharan African countries is lower than the other countries by 1.1%∼1.8%. Sachs and Warner (1997) and Bhattacharyya (2009) explain this low economic growth rate by resorting to geographical aspects, while Easterly and Levine (1997) claim ethnic diversity as its cause. Then, Temple and Johnson (1998) focus on social capability, and Price (2003) focuses on colonial legacies as possible causes, while Acemoglu, Johnson, and Robinson (2001, 2002) examine the qualities of institutions as the main cause of the low growth rate. Nevertheless, the Sub-Saharan Africa dummy is still statistically signiﬁcant, even if these variables are included in their models.
On the other hand, our estimation of Model 4 shows that the Sub-Saharan Africa dummy is no longer signiﬁcant. This model estimation implies that the technology level AD 1500 variable is able to explain the low economic growth rate of the Sub-Saharan African countries. As we discussed earlier, the low precolonial legacy levels of the Sub-Saharan African countries are the main causes of their low growths.
This is evident from the fact that the Sub-Saharan dummy variable becomes nonsigniﬁcant after including the proxy variable for the precolonial legacies. To examine this aspect in more detail, we estimate Model 5 in Table 8. Here, we include the landlocked dummy and ethnic diversity variables, but exclude the technology level AD 1500 variable. As expected, the Sub-Saharan Africa dummy becomes statistically signiﬁcant at the 5% level. The estimated coefﬁcient value is about −1.23, which is similar to the estimate in prior literature 19 (e.g., Bloom and Sachs, 1998). We can see that the landlocked dummy and ethnic diversity variables are signiﬁcant at the 1% and 10% levels, respectively, but they fail to exclude the Sub-Saharan country dummy variable from this model.22 We further examine the roles of social capability and the poor quality of institutions by including the Adelman and Morris index and the quality of law as a proxy for the quality of institutions. Because the institutional quality may have a reverse causality problem, we estimate the model using the 2SLS estimation method. Hall and Jones (1999) suggest using latitude as a proper instrumental variable for the quality of law. As an additional instrumental variable, we employ the near coast variable, which is the percentage of the land area within 100 km of the ice-free coast of each country. This latter variable satisﬁes the exclusion condition. Furthermore, because the hypotheses related to this estimation do not apply to ex-colonies only, we estimate the model using the worldwide samples. As a result, the immigrant ratio and colonial durations are excluded from the set of explanatory variables. These speciﬁcations are summarized in Models 6 and
7. Here, we can see that the variables are all statistically signiﬁcant at the 1% level. That is, high levels of social capability and law quality lead to high economic growth rates. Nevertheless, we can also see that these two variables fail to make the Sub-Saharan dummy variable nonsigniﬁcant. The Sub-Saharan dummy is still negative and statistically signiﬁcant at the 5% level. Therefore, we can say that these two variables are not the most important factors, although they do lead to a low economic growth rate.23 Model 8 estimates the original model by using the state antiquity index AD 1950 as an alternative proxy for the precolonial legacies. We can see that the Sub-Saharan dummy is no longer statistically signiﬁcant and that the state antiquity index AD 1950 has a positive relationship with the economic growth rate. This estimation supports the ﬁnding that the most important factor for the low growth rates of the Sub-Saharan African countries is more closely related to their low levels of colonial legacies than their ethnic diversities, colonial legacies, or geographical factors.
22 Diamond’s (1997) hypothesis can also be associated with this estimation result that the precolonial legacies make the geographical factors nonsigniﬁcant. He argues that geographical factors indirectly inﬂuence economic growth through history, so that the geographical disadvantage of the Sub-Sahara African countries cannot alone resolve every problem associated with low economic growth rate. Over the course of time, the geographical factors have already inﬂuenced the other factors, and have again acted as the causes of the current low economic growth rates.
23 For brevity, we do not report the other estimation results using the other proxy variables for the quality of institutions. Instead of the reported index, we also estimated the model using the indices of government effectiveness, government corruption, and constraint on executive (e.g., La Porta, Lopez-de-Silane, Shleifer, and Vishny, 1999). From these estimations, we obtain the same qualitative results as in our prior estimations. That is, these variables are signiﬁcantly related to current economic growth at the 1% level, but they do not succeed in making the Sub-Saharan Africa dummy variable nonsigniﬁcant. In other words, the dummy is negative and signiﬁcant at the 1% level.
204.3 Robustness Tests using Macroeconomic Variables and Coup Frequency
We now examine the estimation results when nine macroeconomic and coup frequency variables are included in the model. Prior studies have found that all of these are signiﬁcant explanatory variables for economic growth (e.g., Barro, 1991; Levine and Renelt 1992; Sachs and Warner 1997; Sala-i-Martin, 1997a, 1997b; Sala-i-Martin, Doppelhofer, and Miller, 2004). Models 1 to 5 in Table 9 estimate that the average investment rate, coup frequency, real exchange rate overvaluation, and population increase rate are significantly related to the economic growth rate, whereas the average inﬂation rate is nonsigniﬁcant. Model 6 includes these signiﬁcant variables, the four colonial ruler dummies, and three continent dummies. Models 7 and 9 replace the technology level AD 1500 variable with the state antiquity index AD 1950 and the population density, respectively. Although these models are speciﬁed by employing many explanatory variables, the precolonial legacy indices and colonial duration variable are statistically signiﬁcant and maintain the same signs as before.
Insert Table 9 around here.
We speciﬁcally examine the core estimation results. The most interesting estimations are those of Model 1 and Models 6 to 9 of Table 9. If the precolonial legacies contribute to the economic growth by increasing investments, the precolonial legacy variable has to be nonsigniﬁcant after including the investment rate variable. On the contrary, the precolonial legacy index maintains the same sign and remains statistically signiﬁcant. This implies that the precolonial legacies must have affected economic growth in a way other than through the contribution of investment to economic growth. In the same way, the colonial duration variable maintains a negative sign and remains statistically signiﬁcant. This implies that the colonial experience still negatively affects economic growth, although its overall impact on the investment increase is uncertain.
Insert Table 10 around here.
Then, we estimate models using other macroeconomic variables. Table 10 estimates the model using the life expectancy, real exchange rate distortion, the ratio of natural resource exports to GDP, and the share of government investments. Models 1 to 4 of Table 10 estimate the models by separately including these variables and show that the variables are all signiﬁcant. Model 5 includes all the variables simultaneously, and Model 6 adds the continent and colonial dummies to the three signiﬁcant variables, except for the share of government investments, which is nonsigniﬁcant in Model 5. Models 7 and 8 replace the technology level AD 1500 variable with the state antiquity index AD 1950 and the population density AD 1500 variable, 21 respectively. Irrespective of these changes, the precolonial legacy and colonial duration variables are significant, with the same signs as before. Furthermore, they are all signiﬁcant at the 1% level, except Model 7, in which the colonial duration variable is signiﬁcant at the 5% level. This implies that the model estimation results on the precolonial and colonial legacies are robust to the inclusion of these macroeconomic and coup frequency variables.
4.4 Robustness Tests using Income Distributive Variables
Finally, we examine how the model estimation results are affected by including income distribution variables. The literature shows that in cross-sectional data, particularly for less developed countries, these variables have positive relationships with the economic growth rate (e.g., Galor, 2009; Neves and Silva, 2014).
For our examination, we use two income distribution variables: the land Gini coefﬁcient of the 1960s, as provided by Deininger and Olinto (1999), and the middle-class ratio of Perotti (1996). According to the prior literature, the land Gini coefﬁcient and the middle-class ratio are negatively and positively related with the economic growth rate, respectively. The estimations in Table 11 conﬁrm this expectation. At the same time, including the income distribution variables does not change the main results of this study. That is, the precolonial legacy and colonial duration variables are still positively and negatively related with the economic growth rate, respectively, at the 1% level, even after including the income distribution variable.