2.3. Evidence 63
2.3.3 Data Analysis
Baseline Model
In order to test the predictions raised in the theoretical section, I proceed in three steps.
First, as a baseline framework, I mainly rely on random effects and fixed effects models with robust standard errors, respectively. Second, I account for partial adjustments in migrant selection by means of dynamic panel models. Third, I disentangle the impact of resource booms on income inequality and migrant selectivity based on a simultaneous equation model.
As part of the baseline setup restated below,
SELECT IV IT Yijt=αij +φ(RESOU RCESit−RESOU RCESjt) +ξX0ijt+ijt (2.38) I start out with a Hausman test in order to check whether the error components model is more efficient compared to the deviations-from-means estimator. In contrast to the fixed effects estimator, the random effects estimator treats fixed effects as part of a composite error term, αij + ijt = ηijt. Both, fixed and random effects estimators impose strict exogeneity13,
E(ijt|Xijt, RESOU RCESijt, αij) = 0 (2.39)
for t= 1, ..., T, but the random effects estimator additionally hinges on
E(αij|Xijt, RESOU RCESijt) = 0 (2.40)
As the null hypothesis of the Hausman test is rejected withχ2 = 35.55 for the baseline model, I henceforth mainly rely on the fixed effects estimator.
13I abstract from time-fixed effects in a first step. Further, for the sake of parsimony, I account for differences of oil revenues as RESOU RCESijt =RESOU RCESit−RESOU RCESjt as well as for other differenced variables.
The results of the baseline setup are reported in table 2.3 below. In particular, three different estimators are considered, a pooled OLS estimator in columns (1) - (4), a random effects estimator in columns (5) - (8) and a fixed effects estimator in columns (9) - (12), even though the results of the fixed effects estimator serve as the main ref-erence in light of the Hausman test. Both the random effects and the fixed effects models rely on country pair fixed effects, while I complementarily control for time fixed effects (columns (3), (4), (7), (8), (11), (12)) following Egger and Pfaffermayr (2003).
Moreover, I test for non-linearities in the relationship between migrant selectivity and resource revenues and the relationship between migrant selectivity and the polity2 in-dex (columns (1)-(12)). In addition, I test for pairwise interactions between oil revenues and the polity index (columns (6), (8)), between oil revenues and a civil war dummy (columns (2), (4) (10), (12)) and finally between the polity2 index and the civil war dummy (columns (3), (7)). The results do not refute the theoretical claim that resource booms foster brain drain effects. Apparently, oil revenues per capita, the main variable of interest, appears to be positively and significantly related to the selectivity of emigra-tion (in the absence of civil wars and a polity index equal to 0). In other words, a rise in relative oil abundance corresponds with an increase in brain drain effects, captured by the years of schooling of emigrants compared to the average years of schooling in the source country. This association appears to be qualitatively consistent through all model specifications. Moreover, the results display significant non-linearities in the rela-tionship between relative oil abundance and migrant selectivity. Namely, oil abundance sets the stage for brain drain effects, though this effect is decreasing in the level of oil abundance. I test the robustness of this finding with respect to the dynamic setup in section 2.3.3.2 below. Whether brain drain effects are mediated through distributional effects, as the theory suggests, is not clear-cut. In order to account for mediating effects through income inequality, I have to rely on a simultaneous equation model in section 2.3.3.3.
2.3. Evidence 65
With respect to covariates, the quantity as well as the selectivity of migration are negatively associated in the pooled OLS model as well as the random effects specifica-tions with time fixed effects. The larger the number of individuals migrating between two countries in one period, the lower the selectivity of emigration in the following period. This inverse relationship indicates that for low-skilled individuals existing com-munities and networks are much more important while high-skilled individuals appear to be more adaptable. In other words, the results suggest a quantity-selectivity-trade-off in migration. However, as opposed to the other specifications, the fixed effects estimates do not display any apparent selectivity-quantity tradeoff. Physical costs of migration are captured by distances between source and destination countries and are positively related to the selectivity of emigrating individuals. Migration costs are more easily borne by high-skilled individuals. Hence, the results are consistent with the theoretical predictions.
Moreover, the average income per capita in the source country seems to dampen brain drain effects which signifies that in developed countries individuals encounter lower poverty constraints of migration. Yet, the relationship is insignificant in several fixed effects specifications, especially while accounting for time fixed effects as well. An-other variable indicating development and institutional quality is the polity-index rang-ing from -10 (autocracy) to 10 (democracy). Apparently, the selectivity of migration and the polity index are not significantly associated while accounting for country pair fixed effects. Likewise, interacting oil revenues with a civil war dummy does not lead to significant estimates in the fixed effects specifications legitimized by the Hausman test either. In general, I expect that more developed countries with good institutions are less prone to a resource curse. Countries with good institutions are often able to ease the natural resource curse or even turn it into a blessing due to institutional quality (Van der Ploeg (2011)). “Norway is the world’s third largest petroleum exporter after Saudi Arabia and Russia, but is one of the least corrupt countries in the world and en-joys well developed institutions, far sighted management and market friendly policies.”
(Van der Ploeg (2011), p. 368) Therefore, even though the quality of institutions is not exogenous but depends on natural resource wealth (Isham and Busby (2005)), countries lacking in institutional quality may hardly turn the curse of natural resources into a blessing. This presumption is consistent with Sala-i-Martin and Subramanian (2003) who hypothesized that corruption and the transfer of money to elites is the main reason for the contraction of Nigeria’s economy in the course of resource findings. However, with respect to migration, better institutions might correspond with trade openness as well setting the stage for migration opportunities in the course of a Dutch disease. Yet, neither the interaction between oil revenues and the polity2 index nor the relationship between oil revenues and the civil war dummy are significant according to the estimates of the static panel model.
2.3. Evidence 67