14
There is evidence of conditional convergence, since countries with a low (log of the) level of
initial real GDP per active member of the population catch up and grow relatively fast.
Countries with a high log ratio of real public and private gross domestic investment to real
GDP averaged over 1970-89 grow faster. Countries with a large number of years in which
their economy is rated as open and whose citizens accept the rule of law more easily (on a
scale from 1 to 6) grow faster. Even taking account of these traditional growth determinants,
there is a strong negative effect of resource dependence (measured by the share of exports of
primary products in GNP in 1970) on growth. This is what has become known as the resource
curse. This pioneering study gives no role for institutions or bureaucratic quality in explaining
the curse. The second regression reported in Table 1 uses more countries, more years and an
index of institutional quality (on a scale from 0 to 1). Using the starting year 1965 rather than
1970, it confirms that resource rich economies experience slower growth and that institutional
quality is not significant at the 5 percent level (see, however, section 3.3).
These regressions are the cornerstone of many discussions of the resource curse, but
can be criticized on econometric grounds. For example, the share of resources in GNP
(dependence) is potentially endogenous and, if instrumented, it does not significantly affect
growth whereas subsoil resource wealth (abundance) does have a significant positive effect on
growth (Brunnschweiler and Bulte, 2008). However, natural resource wealth is also
endogenous as it is calculated as the present value of natural resource rents. If it is
instrumented with the more exogenous measure of economically recoverable reserves, there is
no evidence for either a curse or a blessing unless one allows for an indirect effect via
volatility (van der Ploeg and Poelhekke, 2010). Another issue is the negative correlation
between growth performance and resource dependence, which may merely be picking up
cross-country variations in income per capita. Alternatively, if the non-resource traded sector
declines and the wage premium for education falls, resource rich economies might invest less
in education and thus the growth rate falls. Hence, adding a control for education implies that
the negative coefficient on resource dependence should fall. Similar points apply to
intermediate variables such as wars or institutional quality, so one should be careful about
drawing inferences about the speed of convergence from the coefficient on initial income.
There may also be some omitted variable bias if a third factor say ‘underdevelopment’ is
driving income as then countries with a low income potential are measured as resource rich.
It is crucial to move from cross-country to panel data evidence to avoid omitted
variable bias arising from correlation between initial income per capita and the omitted initial
level of productivity (Parente and Prescott, 1994; Islam, 1995). If resource dependence is
expressed as a fraction of national income, cross-country regressions that do not control
properly for initial productivity under-estimate the speed of convergence and over-estimate
the share of capital in value added. Even though this requires reliable data on changing quality