Bias correction in a dynamic panel data model of economic growth: The African dummy re-examined

The discrepancy between the observed and expected growth rates of African economies in cross-country or panel growth regressions is often summarised in a significant African dummy. However, the existence of this dummy may be an artifact of the panel data techniques used. The standard LSDV (least squares dummy variable) method produces a large bias in the estimate of the coefficient on the lagged dependent variable, which could generate the observed African dummy. The lagged dependent variable in a growth model is used to calculate the cross-country rate of convergence. If, however, the convergence rate is overestimated, then the Africa dummy would result due to the clustering of African economies at the lower end of the world cross-country income distribution. Correcting for the bias - using Kiviet’s (1995) algorithm - allows a fresh look at the apparent systematic underperformance of African countries relative to their growth predictions. Little evidence remains of such underperformance by African economies once the relevant bias in the dynamic panel has been accounted for.

O400, O410

growth regressions, dynamic panels, African dummy