Historical factors were also incorporated. Large plantations on which slaves produced sugar were established in Latin America, and in North America land suitable for growing crops gave rise to a less unequal agricultural middle class. That is why wheat-to-sugar land ratios could be deployed as a proxy for inequality.
Causes – what machine learning doesn’t tell us
Social scientists prefer finding causal inferences, rather than just applying machine learning to get accurate predictions. After all, predictive methods tell us almost nothing about the actual causes of the findings. It also might be the case that other factors affected both poverty and development outcomes.
So explanatory variables connected to long-term growth were included such as schooling, demography and geographic characteristics. Again: the percentage of the population living in poverty was again significant in predicting how the country is doing. At the same time, the Gini coefficient turned out not to be good predictor.
The results imply that reducing Bolivia’s poverty level (50 %) to Uruguay's level (10 %) would erase the 20 % difference in secondary education enrollment rate between these two states. The increase in per capita GDP would be similar. In developed countries, poverty in its absolute form is not present and this means the effect would be not as significant.
-jk-