Stochastic Frontier Analysis - Dummy interpretation
Hello my fellow Stata-friends,
let me thank you first for joining in to my discussion within this post. I've been estimating various models within the realm of stochastic frontier analysis to determine the technical efficiency of the agricultural production in my area, by utilizing a comprehensive panel data set. All of this was done for a Master research project with 150h workload so please bare with my lack of knowledge. I read 3 relevant books from the famous Kumbhakar and various publications. I estimated the CMAD, Half-Normal, Fixed Effects and Random Effects model to obtain an overall estimate for mean efficiency. I then further explored exogenous determinants for the inefficiencies on the farm level, by including several determinants in a one-stage approach (code below). By nature, most of my determinants are categorial measures (e.g. management-form 1-conventional; 2-transition; 3-ecological). The functional form for the frontier is a translog form which gives the opportunity to directly interpret in percentage. To determine the effect on the mean inefficiency, the marginal effects of the determinants were calculated (as discussed in Kumbhakar). The interpretation for the continuous variables (e.g. age) is straight forward. However, I cant find any information how to interprete the dummies created for all my categorial information. I wonder if the marginal effect interpretation still holds (e.g. omitting management-form 1; the marginal effect being from 0 to 1 for the variable coding the transition farms).
I hope my problem was elaborated clearly enough. I would really appreciate some help from you guys, as I trust that there are several more capable statistician within this forum.
I look forward to your replies and I wish all of you a nice and hopefully sunny day!
Best regards,
Kevin
global varTRL log_capital log_cr_inputs log_intermediates log_labor log_size log_capital_sq log_cr_inputs_sq log_intermediates_sq log_labor_sq log_size_sq log_int_cap_cr_inp log_int_cap_interm log_int_cap_labor log_int_cap_size log_int_cr_interm log_int_cr_labor log_int_cr_size log_int_interm_labor log_int_interm_size log_int_labor_size log_int_cap_t log_int_cr_t log_int_interm_t log_int_labor_t log_int_size_t t t_sq
global DETvars managform2 managform3 acqcharacter2 age education2 education3 education4 education5 gender2 location2 location3 location4 altitude2 altitude3 log_netinvest debt_asset_ratio fracfamlabor fracownsize
version 10.1
sfmodel log_outcrop, prod dist(h) frontier($varTRL) usigmas($DETvars) vsigmas()
ml max, difficult gtol(1e-5) nrtol(1e-5)
sf_predict, bc(bc_w) marginal
sf_transform
let me thank you first for joining in to my discussion within this post. I've been estimating various models within the realm of stochastic frontier analysis to determine the technical efficiency of the agricultural production in my area, by utilizing a comprehensive panel data set. All of this was done for a Master research project with 150h workload so please bare with my lack of knowledge. I read 3 relevant books from the famous Kumbhakar and various publications. I estimated the CMAD, Half-Normal, Fixed Effects and Random Effects model to obtain an overall estimate for mean efficiency. I then further explored exogenous determinants for the inefficiencies on the farm level, by including several determinants in a one-stage approach (code below). By nature, most of my determinants are categorial measures (e.g. management-form 1-conventional; 2-transition; 3-ecological). The functional form for the frontier is a translog form which gives the opportunity to directly interpret in percentage. To determine the effect on the mean inefficiency, the marginal effects of the determinants were calculated (as discussed in Kumbhakar). The interpretation for the continuous variables (e.g. age) is straight forward. However, I cant find any information how to interprete the dummies created for all my categorial information. I wonder if the marginal effect interpretation still holds (e.g. omitting management-form 1; the marginal effect being from 0 to 1 for the variable coding the transition farms).
I hope my problem was elaborated clearly enough. I would really appreciate some help from you guys, as I trust that there are several more capable statistician within this forum.
I look forward to your replies and I wish all of you a nice and hopefully sunny day!
Best regards,
Kevin
global varTRL log_capital log_cr_inputs log_intermediates log_labor log_size log_capital_sq log_cr_inputs_sq log_intermediates_sq log_labor_sq log_size_sq log_int_cap_cr_inp log_int_cap_interm log_int_cap_labor log_int_cap_size log_int_cr_interm log_int_cr_labor log_int_cr_size log_int_interm_labor log_int_interm_size log_int_labor_size log_int_cap_t log_int_cr_t log_int_interm_t log_int_labor_t log_int_size_t t t_sq
global DETvars managform2 managform3 acqcharacter2 age education2 education3 education4 education5 gender2 location2 location3 location4 altitude2 altitude3 log_netinvest debt_asset_ratio fracfamlabor fracownsize
version 10.1
sfmodel log_outcrop, prod dist(h) frontier($varTRL) usigmas($DETvars) vsigmas()
ml max, difficult gtol(1e-5) nrtol(1e-5)
sf_predict, bc(bc_w) marginal
sf_transform
Comment