Attrition in the sample. Because the health shock and other outcome variables in analysis are
dened using changes in variables between adjacent waves of the survey, it is important to consider
attrition within the selected sample. At the household level, about 8.3% on average cannot be
followed between the survey waves. This is a low attrition rate for a panel survey, particularly in
a developing country. Attrition is potentially worrisome if it is correlated with the independent
variable of interest, such that sample selection could then lead to biased estimates. I estimate
regressions of equation (3) on the overall sample used in the main analysis. The dependent variable
is an indicator variable equal to 1 if the household cannot be tracked in the following survey round
(and 0 otherwise). There is no evidence that attrition is correlated with the health shock or the
interaction term between the health shock and the reform: the coecient estimates 0 and 1 are
small in magnitude and statistically insignicantly dierent from zero.22
At the individual level, on the other hand, sample attrition is higher especially among children.
The main outcome variables on children used in the analysis are school attendance and work status.
For these variables, I explain the tracking of household children across survey rounds in Appendix
B.
Attrition in the sample. Because the health shock and other outcome variables in analysis arede ned using changes in variables between adjacent waves of the survey, it is important to considerattrition within the selected sample. At the household level, about 8.3% on average cannot befollowed between the survey waves. This is a low attrition rate for a panel survey, particularly ina developing country. Attrition is potentially worrisome if it is correlated with the independentvariable of interest, such that sample selection could then lead to biased estimates. I estimateregressions of equation (3) on the overall sample used in the main analysis. The dependent variableis an indicator variable equal to 1 if the household cannot be tracked in the following survey round(and 0 otherwise). There is no evidence that attrition is correlated with the health shock or theinteraction term between the health shock and the reform: the coecient estimates 0 and 1 aresmall in magnitude and statistically insigni cantly di erent from zero.22At the individual level, on the other hand, sample attrition is higher especially among children.The main outcome variables on children used in the analysis are school attendance and work status.For these variables, I explain the tracking of household children across survey rounds in AppendixB.
การแปล กรุณารอสักครู่..
