Content area
Full Text
Eur J Health Econ (2014) 15:515531 DOI 10.1007/s10198-013-0492-1
ORIGINAL PAPER
Identifying covariates of population health using extreme bound analysis
Fabrizio Carmignani Sriram Shankar
Eng Joo Tan Kam Ki Tang
Received: 17 April 2012 / Accepted: 23 May 2013 / Published online: 14 June 2013 Springer-Verlag Berlin Heidelberg 2013
AbstractBackground The literature is full of lively discussion on the determinants of population health outcomes. However, different papers focus on small and different sets of variables according to their research agenda. Because many of these variables are measures of different aspects of development and are thus correlated, the results for one variable can be sensitive to the inclusion/exclusion of others. Method We tested for the robustness of potential predictors of population health using the extreme bounds analysis. Population health was measured by life expectancy at birth and infant mortality rate.
Results We found that only about half a dozen variables are robust predictors for life expectancy and infant mortality rate. Among them, adolescent fertility rate, improved water sources, and gender equality are the most robust. All institutional variables and environment variables are systematically non-robust predictors of population health. Conclusion The results highlight the importance of robustness tests in identifying predictors or potential determinants of population health, and cast doubts on the ndings of previous studies that fail to do so.
Keywords Population health Regression Extreme
bounds analysis Robustness
JEL Classication I10 I19 C21
Introduction
Good population health is regarded as a key policy objective in both industrial and developing countries, and, therefore, it is not surprising to nd a lively literature that empirically investigates the determinants of health outcomes. Studies in this literature typically use a standard regression framework of the type:
hi Wia ei; 1
where h is a generic indicator of health such as life expectancy, W is a set of potential determinants of health, e is a stochastic disturbance, and a is the set of coefcients to be estimated. Data usually come in cross-sectional (or eventually panel) format, with i denoting the generic country.
Or [47] provides an early example of this approach. He regresses a measure of premature mortality on proxies for the quality of the medical system and environmental factors, and nds a signicant positive impact of occupational status and public...