2 Model Specification

In this section, I establish two general regression models in which GDP per capita and GDP growth rate are respectively regressed on the same set of explanatory variables, that is,

Y=Xβ+Zθ+Wγ+u

Where Y represents GDP per capita in the first model and GDP growth rate in the second, X is the set of variables of natural resources, Z is the set of variables of infrastructure and education and W is the set of variables representing the extent of openness and reform.

The underlying logic behind the model specification is as follows. First, the effects of natural resources on income level and economic growth can be reflected by the sign of β. The positive sign means that natural resources do matter for the regional economic development so that resource-abundant provinces can take this advantage to achieve higher development level and speed, or in other words, natural resources are still the bottleneck for regional development in resource-scarce provinces, while the negative sign may imply existence of the curse of nature and the insignificance of the parameters may imply that regional economies have progressed to such a stage that natural resources no longer exert any influence on economic performance.

Second, the improvement in infrastructure can overcome the bottleneck of natural resources. Infrastructure here is supposed to consist of two kinds, hardware and software. The hardware infrastructure includes transportation, communication, and other physical facilities, and the software infrastructure refers mainly to human resources, that is, education. The improvement in both hardware and software infrastructure can bring barren regions out of the resource trap by increasing the accessibility to markets, reducing transaction costs and enhancing labor productivity. In practice, infrastructure construction has been a critical component in the Western Development project implemented by the central government since 2000, aiming to promote the development of the lagging western provinces. So, if infrastructure has the effects mentioned above, the sign of the coefficients on the set of variable, Z, is expected to be positive.

Third, regional development disparities can also be explained by various degrees of openness and market reform among provinces. Globalization and marketization have always been two themes of the economic reform since the 1980s. And the essence of the reform is the institutional change, from the planned economic institutions transformed to the market economic institutions. Therefore, the variables of openness and market reform can be regarded essentially as the proxies of the institutional change. So we have institutions, together with natural resources and infrastructure, account for regional economic development.

The estimation of both regression models is stepwise. The barebone model only with the explanatory variables of natural resources is first estimated. Next, the variables for infrastructure, first with hardware and then software, are added in the regression. If infrastructure can free regional economies from the restriction of natural resources, we would see the coefficients on natural resources decreased. And finally, the fully specified model with natural resources, infrastructure and openness and reform is estimated. Again, we would observe the influence of natural resource dwindled in the full model.

Also, the initial value of GDP per capita in the period studied is included in all the regression. In the regressions with GDP growth rate being the dependent variable, its inclusion has been used in many studies to examine convergence or divergence of economic development, with negative sign implying convergence and vice versa. In the regressions with GDP per capita is the dependent variable, the inclusion is aimed to check the influence of the initial income level on the future scale, and Bao et al. (2002) also use it as the proxy for initial capital stock which is assumed to affect the income level positively.