Help with Stochastic Frontier Analysis in Stata, bc95 model

[Bernice Monkah]

Dear Stata users,

If you could please help me with this.

I am working on 'the determinants of health systems in Sub-Saharan Africa' and testing whether some socio-economic and environmental factors effect the efficiency of the system for my project. It is from the period of 2001-2015 and include some other countries as benchmark counties. Where Health adjusted life expectancy is my output -- my inputs are :Health expenditure, GDP per capita, ,Unemployment, Immunization of measles, diphtheria and pertussis, Access to sanitation facilities, Population density -- and my exogenous variables are: Public share of health expenditure, Out-of-pocket expenditure, Urbanization, Gini coefficient, Corruption, Government effectiveness, Rule of law.

I am using the Battese & Coelli(1995) model, where I can simultaneously parameterize the inputs and exogenous determinants of efficiency. I have gone through the Stata journal by Belotti et al. (2013) as well. I took the log of some of the inputs that were not already in % form. Every time I try and run the model I keep running into errors.

My question(s) is
1. Am I missing something in the command, that is causing this such output?
2. Is there a better way to get results for this besides the model I am using?

Anything will be very helpful.

Thank you.

sfpanel lnhale lnhexp lngdp dpt3 san unemply lnpopden , model(bc95) ort(o) emean( phexp oophexp urban corrupt goveff rol gini )


initial:       Log likelihood = -743368.14
Iteration 0:   Log likelihood = -743368.14 
could not calculate numerical derivatives -- flat or discontinuous region encountered
could not calculate numerical derivatives -- flat or discontinuous region encountered

Inefficiency effects model (truncated-normal)        Number of obs =       242
Group variable: code                              Number of groups =        96
Time variable: year                             Obs per group: min =         1
                                                               avg =       2.5
                                                               max =         3

                                                     Prob > chi2   =         .
Log likelihood = -2.237e+04                          Wald chi2(0)  =         .

------------------------------------------------------------------------------
      lnhale |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
Frontier     |
      lnhexp |   24.16933          .        .       .            .           .
       lngdp |   35.60164          .        .       .            .           .
        dpt3 |   338.7429          .        .       .            .           .
         san |   256.3927          .        .       .            .           .
     unemply |   28.79692          .        .       .            .           .
    lnpopden |   15.92289          .        .       .            .           .
       _cons |   5.733198          .        .       .            .           .
-------------+----------------------------------------------------------------
Mu           |
       phexp |  -228.3673          .        .       .            .           .
     oophexp |  -121.7882          .        .       .            .           .
       urban |  -216.8518          .        .       .            .           .
     corrupt |   .3903584          .        .       .            .           .
      goveff |   .3277178          .        .       .            .           .
         rol |   .4112223          .        .       .            .           .
        gini |   -148.933          .        .       .            .           .
       _cons |  -2.930592          .        .       .            .           .
-------------+----------------------------------------------------------------
Usigma       |
       _cons |   182.3525          .        .       .            .           .
-------------+----------------------------------------------------------------
Vsigma       |
       _cons |   182.3525          .        .       .            .           .
-------------+----------------------------------------------------------------
     sigma_u |   3.96e+39          .        .       .            .           .
     sigma_v |   3.96e+39          .        .       .            .           .
      lambda |          1          .        .       .            .           .
------------------------------------------------------------------------------

.

[Chiangi Aza]

Try the syntax below:

sfpanel lnhale lnhexp lngdp dpt3 san unemply lnpopden year, model(bc95) dist(tn) emean( phexp oophexp urban corrupt goveff rol gini) ort(o)

There was no trend variable-year, or you change that to your time variable; distribution-(tn); the orientation, ort(o) comes at the end of the syntax

I hope it helps.

[Haoyu Wang]

Another new bird in sfa, but I've met with similar problem. May you try add term i.dmu in your xvarlist.

Good luck!

[Mandy Smith]

Another new bird in sfa, but I've met with similar problem. May you try add term i.dmu in your xvarlist.

Good luck!

could you explain it more detailed？please

[Alexander Lotz]

Dear Bernice, did you find a solution to your problem, eventually?
I am using the -sfcross- command (also suggested by Belotti et al. (2013)), and get the same errors with Stata showing only dots for Std. Err., z, P>|z| etc.
Thank you in advance.

[Akbar Muhammad]

Hi Alexander Lotz Have you find the solution? mind to share about it? thanks in advance