I use data from 1993{2011, comprising of seven waves covering the entire duration of the health
insurance reform. Because the outcome variables and the health shock are in rst dierences, the
data from 1993 are used as the initial condition.12 Given that enrollment in the health insurance
program is at the household level, the unit of observation under analysis is the household. The
sample consists of rural households, dened as those in which the household head lives in a rural
county and has a rural hukou.13 Given that a health shock is dened in terms of the changes in
health between waves among key members of the household (as explained below), I focus on stable
households that have not experienced any change in household head.14 In order to capture health
shocks with a signicant economic impact, I exclude households where the head of the household
is more than 65 years of age.