Appendix A

Appendix A. Summary Statistics
Observations / Mean / Standard Deviation / Minimum / Maximum
Life Expectancy / 1,710 / 60.70 / 11.73 / 31.61 / 79.54
Infant Mortality / 1,702 / 66.36 / 52.59 / 2.300 / 230.2
Openness / 1,291 / 4.078 / 0.672 / 0.702 / 6.048
GDP per capita / 1,289 / 7.855 / 1.060 / 5.139 / 10.44
Population Growth / 1,578 / 0.105 / 0.0880 / -0.243 / 1.096
Female Education / 846 / 4.418 / 6.389 / 0 / 50.90
Democracy / 1,288 / 0.0210 / 7.441 / -10 / 10
Ethnic Fractionalization / 1,489 / 47.00 / 27.47 / 0 / 98.40
Economic Growth / 995 / 0.313 / 0.407 / -0.456 / 6.284
Income Inequality / 580 / 40.27 / 9.758 / 17.83 / 64.30
Alternative Income Inequality / 1,103 / -0.964 / 8.589 / -24.04 / 40.70

1

Appendix B

Our findings are generally encouraging for the Dreze-Sen hypothesis, but one concern is the small sample size. The GINI data for income inequality suffer from limited availability both spatially and temporally, meaning that our sample consists of only 236 observations once cases for which observations on income inequality are not available. Additionally, it is possible that the estimated effect may be related to the methodology used to construct the GINI measure, raising the question of whether the effect is robust to estimation of GINI using an alternative measure of domestic income inequality.

For these reasons, we also generate an additional set of results which use the same framework as models 4-1 and 4-2, but now use an alternative measure of income inequality. To construct our alternative measure we downloaded data on fixed telephone lines per 100 citizens from the World Bank’s World Development Indicators. We chose this measure because we anticipate that the number of fixed telephone lines will be correlated with the level of income inequality. In other words, the demand for telephone service should be higher in countries where a more equitable distribution of income generates a more robust private market for these services. The challenge, however, is that the level of overall economic development- rather than inequality alone- is a key predictor of higher levels of, and growth in, this type of telecommunications infrastructure (see as an example Sridhar and Sridhar 2004).[A]

To address this limitation, we constructed our alternative measure of inequality in the following manner. The first step was to construct a bivariate regression estimating the impact of GDP per capita on our measure of fixed telephone lines, after which we then predicted and captured the residuals from this model. The goal was to create a measure of telephone lines per 100 individuals which was independent from the level of development, and therefore representative of the second explanatory factor of income inequality. A basic correlational analysis of the residuals and our two explanatory factors supports our intuition. The residuals are only marginally correlated with the level of development (-0.07), whereas the original measure was highly correlated with GDP per capita (0.777). For the residuals to be consistent with our interpretation of income inequality, they would be anticipated to be negatively correlated with our existing measure of Income Inequality, since greater telephone line penetration would be indicative of lower levels of inequality, while increases in the GINI measure represent increases in inequality. Reassuringly, this is in fact what we observe, as the telephone residuals are correlated with Income Inequality at -0.407.

Having constructed our alternative measure of income inequality, we once again interact it with Openness and substitute the constituent and interactive terms into our original models 3-3 and 3-4, which are now represented in models A-1 and A-2.

If Dreze and Sen are correct, then the interaction will be negative and the coefficient will become larger as the value of the alternative measure of Income Inequality (i.e., telephone residual) increases, indicating that trade reduces infant mortality in nations where majority of citizens have access to telephones. The coefficient for the interaction for life expectancy will be the opposite; the coefficient will become increasingly positive as the amount of telephone lines increases. Based upon the results of these estimations we also then construct the necessary conditional coefficients and confidence intervals to assess the impact and statistical significance of Openness conditional on our alternative measure.

Table B. Robustness Tests

(B-1) / (B-2) / (B-3)
Infant Mortality / Life Expectancy / Undernourishment
Openness / -8.084** / 3.480*** / -18.31***
(3.222) / (0.657) / (6.507)
Alternative Inequality Measure / 5.290*** / -1.233***
(1.378) / (0.280)
Openness X Alternative Inequality Measure⁺ / -0.945*** / 0.263***
(0.298) / (0.0604)
Income Inequality / -1.193**
(0.565)
Openness * Income Inequality⁺ / 0.336**
(0.141)
GDP per capita / -4.466 / -0.318 / -4.861***
(3.275) / (0.667) / (1.523)
Female Education / -3.549*** / 0.605*** / -0.145
(0.497) / (0.102) / (0.213)
Population Growth / -1.564 / 2.542 / 20.90*
(16.83) / (3.420) / (12.39)
Democracy / -0.874*** / 0.0921** / -0.0506
(0.198) / (0.0402) / (0.130)
Ethnic Fractionalization / 1.263*** / -0.161** / 0.0544
(0.347) / (0.0705) / (0.0481)
Constant / 85.50*** / 53.51*** / 118.6***
(31.82) / (6.461) / (26.33)
Observations / 490 / 490 / 136
R-squared / 0.476 / 0.348 / 0.429
Number of Countries / 69 / 69 / 53
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
⁺ Conditional coefficient of interactive effect evaluated in accompanying figures

Figure A shows the conditional coefficient of Openness upon Infant Mortality and Life Expectancy from models B-1 and B-2.

Figure A.

The results we see in Figure A are consistent with those we observed in Figure 1, keeping in mind that increases in our alternative measure now represent decreases in overall inequality. We see that the greatest decreases in infant mortality as a function of trade openness occurs at the lowest level of income inequality; at the 90th percentile of the observed distribution, a single unit increase in trade decreases infant mortality by approximately 15 deaths per 1,000 births. This effect decreases as inequality increases, and becomes statistically significant between the 50th and 60th percentiles. A similar effect is observed Life Expectancy from model A-2: the largest increase in life expectancy occurs at the lowest levels of income inequality, and attenuates as inequality increases. The effect also becomes statistically insignificant at the 20th percentile of the distribution.

Having used a number of different measures relating to inequality, we also explore an alternative dependent variable more directly related to proportion of the poor within a country: Undernourishment.[B] This variable measures the percentage of the population whose food intake is below what is necessary to meet daily needs, and was obtained from the World Bank’s World Development Indicators.[C] We anticipate that the effect of trade and income inequality should be similar for undernourishment as it is for infant mortality and life expectancy; increasing trade should reduce the prevalence of undernourishment, but the effect should be conditional upon the prevailing level of income inequality. The results of this estimation, presented in Model B-3 with the calculated conditional coefficient presented in Figure B, are consistent with this expectation. Increasing trade does reduce undernourishment, but this effect diminishes as income inequality increases, eventually yielding no benefit.

Figure B.

1

[A]Sridhar, Kala Seetharam & Sridhar, Varadharajan. (2004). “Telecommunications Infrastructure and Economic Growth: Evidence from Developing Countries.” New Delhi National Institute of Public Finance and Policy Working Paper.

[B]We thank an anonymous reviewer for this suggestion.

[C]Accessed December 5, 2015.