We checked the robustness of our findings, first by substituting the GSNI for the GII in our regression analysis (Table A5). GII and GSNI values were highly correlated (Figure A3) and, as expected, these regressions produced results very similar to those shown…