The numbers of districts selected for each of the crops are 153 for rice, 80 for pearl millet and 88 for sorghum. These districts cut across the agriculturally-important states of the country (rather than being selected from certain states). For the criteria used in selection of districts, refer to the data appendix. Table A.4 lists the districts considered in each state, for every crop. As previously mentioned, the districts in- cluded in the ICRISAT database are those that existed as of 1966. However, the climatic dataset has been created taking into account the district boundaries as of 2002, which are remarkably different from those of 1966. The districts that comprise the panel-sample have been selected on the basis of the districts that existed in the ICRISAT database, and the climatic variables for these districts have been approxi- mated from the district to which the largest area of the parent district was allocated19 (provided that it is more than 50% of the total area of the parent district) (Kumar and Somanathan, 2009).
In the presence of AR cross-sectional dependence (the outcomes are correlated across districts in a given year), along with heteroscedasticity, FGLS (feasible gen- eralized least squares) with fixed effects was found to be an appropriate method of estimation. However, one of the drawbacks of FGLS estimation is that it produces overly optimistic standard error estimates. Moreover, the estimates are only feasible if
N < T, i.e., the number of observations are less than the number of time period, which
is not the case for any of the three crops. To correct this, panel-corrected standard error (PCSE) estimates are obtained, where the parameters are estimated using a Prais–Winsten (or OLS) regression. Equations have been estimated with district and year fixed effects, district fixed effects and district-by-year fixed effects.
For each of the crops, it was observed that the errors exhibited the presence of heteroscedasticity, and contemporaneous correlation. A Prais–Winsten regression was thus estimated, under two different assumptions on correlation:
(1) Within panels, there is AR (1) autocorrelation and the coefficient of the AR (1) process is common to all of the panels, and
(2) Within panels, there is AR (1) autocorrelation and that the coefficient of the AR (1) process is specific to each panel (i.e., panel-specific AR (1) autocorrelation) (Cameron and Trivedi, 2009)
The regression equation which is estimated for all three crops is as follows:
The numbers of districts selected for each of the crops are 153 for rice, 80 for pearl millet and 88 for sorghum. These districts cut across the agriculturally-important states of the country (rather than being selected from certain states). For the criteria used in selection of districts, refer to the data appendix. Table A.4 lists the districts considered in each state, for every crop. As previously mentioned, the districts in- cluded in the ICRISAT database are those that existed as of 1966. However, the climatic dataset has been created taking into account the district boundaries as of 2002, which are remarkably different from those of 1966. The districts that comprise the panel-sample have been selected on the basis of the districts that existed in the ICRISAT database, and the climatic variables for these districts have been approxi- mated from the district to which the largest area of the parent district was allocated19 (provided that it is more than 50% of the total area of the parent district) (Kumar and Somanathan, 2009).In the presence of AR cross-sectional dependence (the outcomes are correlated across districts in a given year), along with heteroscedasticity, FGLS (feasible gen- eralized least squares) with fixed effects was found to be an appropriate method of estimation. However, one of the drawbacks of FGLS estimation is that it produces overly optimistic standard error estimates. Moreover, the estimates are only feasible ifN < T, i.e., the number of observations are less than the number of time period, which
is not the case for any of the three crops. To correct this, panel-corrected standard error (PCSE) estimates are obtained, where the parameters are estimated using a Prais–Winsten (or OLS) regression. Equations have been estimated with district and year fixed effects, district fixed effects and district-by-year fixed effects.
For each of the crops, it was observed that the errors exhibited the presence of heteroscedasticity, and contemporaneous correlation. A Prais–Winsten regression was thus estimated, under two different assumptions on correlation:
(1) Within panels, there is AR (1) autocorrelation and the coefficient of the AR (1) process is common to all of the panels, and
(2) Within panels, there is AR (1) autocorrelation and that the coefficient of the AR (1) process is specific to each panel (i.e., panel-specific AR (1) autocorrelation) (Cameron and Trivedi, 2009)
The regression equation which is estimated for all three crops is as follows:
การแปล กรุณารอสักครู่..
