I use the sfkk command for stochastic frontier analysis. I have more than 6000 household farmers as observations in an aggregate farming system and cross-sectional analysis for the year 2016-17. My dependent variable is wheat_productivity and frontier variables are 1) Irrigated farming dummy (is equal one if household farmer farming irrigated land, 0 otherwise), 2) cultivated_land, 3) working_capital & 4) hired_labor_cost. These frontier variables are endogenous variables as well. The control variables are education_level & Economically_active_family_member. I allocated "distance_from_community_to_nearest_drivable_r aod" as an inefficiency variable. As the model have four endogenous variables, so I allocated four IVs which are 1) Irrigation_Infrastruction_Dummy IV for Irrigated_Farming_Dummy, 2) HH_Asset_Value IV for cultivated_land, 3) Non-farm_income is IV for Working Capital, & 4) HH_Size is IV for Hired_Labor_Cost.
When I regress the following command:
sfkk wheat_productivity irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost Education_Level Economically_Active_family_member, prod u(Distance_from_community_to_nearest_drivable_raod ) en(irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost) i(Irrigation_Infrastruction_Dummy HH_Asset_Value Non-farm_income HH_Size) header delve compare nicely timer.
I produced the results for irrigated farming and rainfed farming by subsample analysis and this is producing results for aggregate farming with only difference that here I included one additional variable which irrigated_farming_dummy. When I regress so I get the error: initial: log likelihood = -<inf> (could not be evaluated)
Due to this problem I cannot produce the final results for aggregate farming, which is important for my differential analysis of irrigated farming and rainfed farming. I also used different commands for solution such as by adding "diff" or "tech" as following:
sfkk wheat_productivity irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost Education_Level Economically_Active_family_member, prod u(Distance_from_community_to_nearest_drivable_raod ) en(irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost) i(Irrigation_Infrastruction_Dummy HH_Asset_Value Non-farm_income HH_Size) header delve compare nicely timer diff
OR
sfkk wheat_productivity irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost Education_Level Economically_Active_family_member, prod u(Distance_from_community_to_nearest_drivable_raod ) en(irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost) i(Irrigation_Infrastruction_Dummy HH_Asset_Value Non-farm_income HH_Size) header delve compare nicely timer tech (bfgs 10 nr 5).
In the end I was not able to get the results. It seems the problem is in choosing a good initial value for likelihood or something like this. In the last 4 months I have worked on consecutively but not able to solve this problem and now looking for your valuable assistance.
Thanks in advance
When I regress the following command:
sfkk wheat_productivity irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost Education_Level Economically_Active_family_member, prod u(Distance_from_community_to_nearest_drivable_raod ) en(irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost) i(Irrigation_Infrastruction_Dummy HH_Asset_Value Non-farm_income HH_Size) header delve compare nicely timer.
I produced the results for irrigated farming and rainfed farming by subsample analysis and this is producing results for aggregate farming with only difference that here I included one additional variable which irrigated_farming_dummy. When I regress so I get the error: initial: log likelihood = -<inf> (could not be evaluated)
Due to this problem I cannot produce the final results for aggregate farming, which is important for my differential analysis of irrigated farming and rainfed farming. I also used different commands for solution such as by adding "diff" or "tech" as following:
sfkk wheat_productivity irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost Education_Level Economically_Active_family_member, prod u(Distance_from_community_to_nearest_drivable_raod ) en(irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost) i(Irrigation_Infrastruction_Dummy HH_Asset_Value Non-farm_income HH_Size) header delve compare nicely timer diff
OR
sfkk wheat_productivity irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost Education_Level Economically_Active_family_member, prod u(Distance_from_community_to_nearest_drivable_raod ) en(irrigated_farming_dummy Cultivated_Land Working_Capital Hired_Labor_Cost) i(Irrigation_Infrastruction_Dummy HH_Asset_Value Non-farm_income HH_Size) header delve compare nicely timer tech (bfgs 10 nr 5).
In the end I was not able to get the results. It seems the problem is in choosing a good initial value for likelihood or something like this. In the last 4 months I have worked on consecutively but not able to solve this problem and now looking for your valuable assistance.
Thanks in advance
