Preston curve
[4] According to Preston, the independent increases in life expectancy have been greatest in the poor countries, although he also believed that a good portion of the potential gains from better medical technology have not been realized.
[4] Several poor countries in Sub-Saharan Africa have actually seen declines in life expectancy in the 1990s and 2000s as a result of the HIV/AIDS epidemic, even if their per capita incomes have increased during this time.
[5] Analysis of more recent data, for example by Michael Spence and Maureen Lewis, suggests that the "fit" of the relationship has become stronger in the decades since Preston's study.
Those below the curve, such as South Africa or Zimbabwe, have life expectancy levels that are lower than would be predicted based on per capita income alone.
If the relationship is driven by other factors, if it is spurious, or if it is in fact health that leads to higher income, then this policy outcome will no longer be true.
[3] The existence of the Preston curve has been used by Lant Pritchett and Larry Summers to argue that poor countries should focus on economic growth, and that health improvements will come about spontaneously as a result of increases in income.
[9] However, the upward shifts of the Preston curve still imply that the main portion of gains in life expectancy has come about as a result of improved health technology rather than just increases in per capita income.
[6] In particular, per capita incomes between countries have generally diverged over time, while life expectancies, and other health indicators such as the infant mortality rates, have converged (this trend was interrupted in the 1990s with the outbreak of the AIDS epidemic in Sub-Saharan Africa).
[6] Healthier children spend more time at school and learn faster, thus acquiring more human capital which translates into higher growth rates of incomes later in life.
[9] This strategy requires identification of an "instrument" – i.e. a variable which correlates with per capita income but not with the error term in the linear regression.
