Hi everyone,
I am working with a farmers dataser to assess their crop production efficiency based on their inputs using Cobb Douglas stochastic frontier. The issue I have is there are numerous farmers with either production equal to zero or one of their inputs equal to zero (seeds, fertilizer, labour cost). The easiest option would be to simply drop those observations. However, such a solution would drastically limit my sample size. Indeed, I am using a stochastic frontier for each of five crops I considered. I was suggested to use a Box-Cox transformation to solve the issue. But I am still wondering whether this is a robust alternative. I also tried it but, while applying the Box-Cox, STATA tells me this "logprod_mil contains observations that are not strictly positive".
logprod_mil represents the millet production in logarithm.
Any help on how to deal with those zero values?
I am working with a farmers dataser to assess their crop production efficiency based on their inputs using Cobb Douglas stochastic frontier. The issue I have is there are numerous farmers with either production equal to zero or one of their inputs equal to zero (seeds, fertilizer, labour cost). The easiest option would be to simply drop those observations. However, such a solution would drastically limit my sample size. Indeed, I am using a stochastic frontier for each of five crops I considered. I was suggested to use a Box-Cox transformation to solve the issue. But I am still wondering whether this is a robust alternative. I also tried it but, while applying the Box-Cox, STATA tells me this "logprod_mil contains observations that are not strictly positive".
logprod_mil represents the millet production in logarithm.
Any help on how to deal with those zero values?
