the reform indicator, controlling for county and year xed eects, when the outcome variables is a
dichotomous variable indicating whether the head or the spouse of the head reported any illness in
the past four weeks (both severe and non-severe). The estimated coecient on reform is small and
insignicantly dierent from zero, suggesting no evidence that the reform changes the percentage
share of households with illness. In column (4), among households reporting some illness, I regress
the duration of illness on the reform indicator, controlling for the year and county xed eects. I
also nd no evidence that the reform shifts the duration of illness. Columns (7){(8) conrm this
nding by repeating the analysis on the health shock measured dened on the head, the spouse of
the head and other working-age household members. Overall, these results indicate that the reform
is unlikely to aect the composition of households with illness.