Here we replicate some of the main regressions reported in Table 9 to analyse their
robustness. We are particularly interested in finding out whether the significance of
schooling depends on the presence of certain outliers.Temple (1998, 1999) shows that
the coefficient for schooling varies considerably under certain specifications when
the countries with relatively high residuals are eliminated from the sample. We again
estimate regressions (9) and (10) by dropping a third and a half of countries with
the highest residuals (the results did not change when a lower number of countries
were eliminated). In Table A.2 we observe that omitting these countries from the
regressions produces small changes in the coefficients. If anything, the significance of