|
I am working with a large data set (approximately 50K observations) and trying to running a Maximum likelihood estimation on 5 unknowns in Stata. I encountered an error message of "Numerical Overflow". How can I overcome this? I am trying to run a Stochastic Frontier analysis using the built in Stata command "frontier". The dependent variable is log of output and the independent variable is log of intermediate inputs, capital, labour, and utlities. |
|||||
|
|
After a day of searching, I found out that the issue was due to starting values. Thought I should just post the answer for future reference. The frontier command in Stata obtains its starting values using method of moments. The initial values might have produced negative infinity for the log likelihood. To get around the problem I needed to specify the starting values myself, which were obtained from a linear regression. |
|||
|
|
|
How many variables do you have? Are you using a built in command? Five years ago I used mlogit (multinomial logit) in Stata 8 to model 60,000 observations and about 40 variables with no difficulty. I'd recommend searching the Stata websites (http://www.stata.com/) and even asking technical support. For example, here are two answers: http://www.stata.com/statalist/archive/2007-04/msg00390.html http://statalist.1588530.n2.nabble.com/st-Reproducing-xtlogit-with-xtmelogit-td3418488.html Stata provide the best technical support of any stats software I know. |
|||
|
|
