Table 6 reports the coecient estimates when the outcome variables are the changes in house-
hold food consumption across four nutritional intakes: calories, proteins, fats and carbohydrates.
If households are fully insured against idiosyncratic shocks, consumption growth should not be
correlated with changes in health once growth in county-level resources is controlled for, and con-
sequently, the coecient on health shock should be zero.
I nd that prior to the reform, negative increments to health are not associated with reductions
in the intake of macronutrients. The point estimates are small and not statistically dierent from
zero. Therefore, I cannot reject the hypothesis of full insurance against severe health shocks before
the reform. This is perhaps not so surprising, given that households were already able to smooth
household income against health shocks to specic members.
In Table 6, the coecient estimates on the interaction of the health shock and reform is the
estimated consumption smoothing eects of the reform (parameter 1). None of the point estimates
are signicant. Adding the coecients in the rst and second rows gives the predicted net eects
of the health shock after the reform. The t-test reported in the third column implies that after
the reform, I still cannot reject the hypothesis that households are fully insured against health
shocks.25
One potential concern is that consumption growth may be shifted by changes in the state of
health (even with full insurance) when the marginal utility of consumption depends on health
status, either directly or indirectly through induced changes in leisure if consumption and leisure
are not additively separable. One advantage of the CHNS data over other household panel data