In a second stage of the analysis, we investigate what state-level variables might
explain the observed state fixed effects in the estimated micro-level equations. The state dummy coefficients are saved for each year and then pooled to generate a panel
with T=2. The panel of state fixed effects coefficients is merged with a panel of data
on state-level GNP, inequality, education expenditure and other relevant predictors.
We have time series data on these predictor variables (see Besley and Burgess 2002,
2004; we are grateful to these authors for allowing us to use the data they have
compiled). We have constructed five-year averages of the predictors, over the five
years preceding the data of the NFHS survey (i.e. 1987/8-92/3 for NFHS1 and
1993/4-98/9 for NFHS2).