Abstract
Using recent cross-sectional, public use, data set on 31 Chinese provinces, we empirically model the core determinants of life expectancy (population health status) using the Ordinary Least Squares (OLS) method, instrumental variables estimation and the relatively more efficient Hubert robust estimator. The empirical regression model results and diagnostic tests indicate that the core determinants of life expectancy are the real GDP per capita, illiteracy rates and daily visits to physicians. Using results of the robust regression estimator (mimic the instrumental variables model estimation), the statistically significant elasticities of life expectancy are 0.033 (t-ratio = 2.45) with respect to per capita real GDP, 0.41 (t-ratio = 2.54) with respect to daily visits to the physicians and is −0.026 (t-ratio = −2.26) with respect to the illiteracy rate. That is, income and daily visits to physicians are positively linked to life expectancy while the illiteracy rate stifles life expectancy production. Our findings are consistent with received theories. Policy implications are explored.
JEL Classification:
Notes
1 Huber’s (see, Hampel et al., Citation1986) extension of his results on estimation of a location parameter to linear regression involves computation of weighted least-squares estimates (redefined iteratively) of the form wi = min {1, c/|ri}, where ri is the ith residual and c a positive constant. The weights depend on the estimate (i.e. they are not fixed). Huber proposed M-estimators Tn, where Ί(Tn) = min{Ί(θ)|θϵⱷ}, where Ί(θ) = ρ((yi – xTiθ)/σ) for some function ρ: R→R+ and for a fixed σ. Assuming σ has a derivative (∂/∂r)ρ(r) = ψ(r), Tn satisfies the system of equations (with the p-vectors xi)
ψ((yi – xTiTn)/σ)xi = 0. The Huber-estimator defined by the weight wi above is a maximum likelihood when the residuals are distributed according to the distribution with density proportional to exp(–ρc(r)).
2 The least-squares estimator is defined by the function σ(r) = r2/2 in the derivation of Huber robust estimation discussed in footnote 1.