دانلود رایگان مقاله رویکرد تجمع بایاس اصلاح برای رابطه سلامت درآمد

Abstract

A large amount of data consisting of 148 countries for the years 1970 to 2010 is analysed in the context of the health–income relationship. The literature suggests that the biased income–health effect obtained with macro data can be a result of the aggregation of individual concave income functions on average health. This aggregation problem is analysed in detail, and a bias-correcting method is proposed to overcome it. The results with new model alternatives show that they correct the income effects on average health in the right direction; that is, they produce smaller parameter estimates than biased models. Augmenting the results with the quantile regression approach, which is sensitive to health differences between countries, indicates that the poorest countries’ income gradient is still much larger than that of rich countries. However, the median life expectancy effect of the log of GDP per capita across the countries decreased during the sample decennials. The results for income inequality measured with the Gini coefficient indicate that the effects of inequality on health are still significant in the poorest countries but non-significant among rich countries after the year 2000. We argue that the proposed bias-correcting method retains the interest in macro health modelling and offers new model alternatives in other contexts.

5. Discussion and conclusions

A large amount of data consisting of 148 countries in the years 1970–2010 was analysed in the context of the health–income relationship. The current literature emphasizes individual data, deriving results for health and incomes. However, the aggregative approach is still active, because country-level data on GDP per capita, income inequality, and average health status are still widening and gaining a longer time span. Both at the individual and at the aggregative data level, some results indicate that the absolute income effect (AIH) on health is still strong but the inequality effect (IIH) is disappearing from developed countries. The literature also suggests that part of the inequality effect obtained with macro data is a result of a concave mean income function on average health. We showed that this estimation strategy is seriously biased because of incorrect aggregation, which we analysed in detail. A method based on the first-order Taylor approximation is suggested to overcome this aggregation-induced errors-in-variables bias. Two bias-correcting model alternatives are provided that correct for aggregation bias and still preserve the individual-level interpretation of estimated income effects on average health. The results show that bias-correcting models produce quite different results for the log GDP per capita effects on life expectancy across the sample countries in the years 1970–2010 from the biased noncorrected reference model. Especially in the period from 1985 to 2005, the biased model with yearly cross-sections gives income effect estimates that are too large compared with the bias-corrected ones. However, the income inequality effects estimated with the GINI coefficients are not affected by the model alternatives. Across the models the inequality effects on life expectancy are still negative and are not significant in statistical terms after the late 1990s. To achieve more transparent income and inequality effects on life expectancy distribution across the sample countries, the bias-correcting model was also estimated with the quantile regression approach, which is sensitive to the life expectancy data distribution. It produces income regression effects at different quantiles of the life expectancy model error distribution. The coefficients for the log of GDP per capita were significant throughout the four decades in both OLS and quantile regressions. In the decade of 1970–1979, when lnGDPc increased by 1%, life expectancy increased on average by 0.061 years in the OLS regressions and 0.051 years in the median LAD regressions. We know that the income gradient is higher for poorer countries than for richer ones. However, the coefficient values of lnGDPc fall over the years for all the countries. In the last decade (2000–2010), the respective lnGDPc values were 0.047 and 0.044 in OLS and median LAD, respectively. The results with quantile regression other than the median for four different decennials in the sample show that the poorest countries’ income gradient is still much higher than that of the rich countries.