Hey,
i am using a data set about the health efficiency in germany. As output measure I use the inverted log of the mortality rate (so it is an desirable output for health efficency). As input factors I use the number of general practitioners, number of specialists, number of hospital beds, number of dialysis devices and an indicator speciality diversity to represent the heterogenity of medical service provision within a district. These are ideas of Herwartz and Schley in "Improving health care service provision by adapting to regional diversity: An efficiency analysis for the case of Germany".
I want to use a stochastic frontier analysis to analyse the efficiency of the districts in germany. My tests proposes to use a time varying model from Kumbhakar (1990) or the time varying model from Battese and Coelli (1992).
I am using a cobb douglas specification.
I use the following code:
sfpanel $Output $loglist, model(bc92)
scalar ll_bc92 = e(ll)
predict ineff_BC, u
predict eff_BC, bc
summarize eff_BC
This code works well, I get the following results
Time-varying decay model (truncated-normal) Number of obs = 3540
Group variable: id Number of groups = 354
Time variable: time Obs per group: min = 10
Prob > chi2 = 0.0000
Log likelihood = 6076.9113 Wald chi2(5) = 74.14
----------------------------------------------------------------------------------------
LogSMRInv | Coefficient Std. err. z P>|z| [95% conf. interval]
-----------------------+----------------------------------------------------------------
LogGP | -.0155696 .0143281 -1.09 0.277 -.0436521 .0125129
LogSpecialists | .011152 .0103765 1.07 0.282 -.0091856 .0314896
LogBeds | - . 0571862 .007225 -7.92 0.000 -.071347 -.0430254
LogDialyseAnzahl | .0015767 .0070368 0.22 0.823 -.0122151 .0153686
LogSpecialityDiversity | .0136673 .0057841 2.36 0.018 .0023306 .0250039
_cons | -2.527519 .075975 -33.27 0.000 -2.676428 -2.378611
-----------------------+----------------------------------------------------------------
/lnsigma2 | -4.517443 .0721543 -62.61 0.000 -4.658863 -4.376023
/ilgtgamma | 2.069327 .0857345 24.14 0.000 1.90129 2.237363
/mu | .3156293 .0181876 17.35 0.000 .2799822 .3512764
/eta | -.0037632 .0012755 -2.95 0.003 -.0062631 -.0012633
-------------+----------------------------------------------------------------
sigma2 | .0109169 .0007877 .0094772 .0125753
gamma | .887886 .0085344 .8700375 .9035549
sigma_u2 | .009693 .0007879 .0081487 .0112373
sigma_v2 | .0012239 .0000308 .0011636 .0012843
------------------------------------------------------------------------------
The efficiency results are around 0,73 (73%)
Variable | Obs Mean Std. dev. Min Max
-------------+---------------------------------------------------------
eff_BC | 3,540 .7364692 .0714501 .5491924 .9892062
Can I use this models since some of the coefficients are negative? I dont understand the effect from the variables onto the mortality rate (health output). For example more hospital beds do not improve the health output. Does this mean there are enough beds?
And are the results or the coefficients significant? I think i need to interpret the z-value.
i am using a data set about the health efficiency in germany. As output measure I use the inverted log of the mortality rate (so it is an desirable output for health efficency). As input factors I use the number of general practitioners, number of specialists, number of hospital beds, number of dialysis devices and an indicator speciality diversity to represent the heterogenity of medical service provision within a district. These are ideas of Herwartz and Schley in "Improving health care service provision by adapting to regional diversity: An efficiency analysis for the case of Germany".
I want to use a stochastic frontier analysis to analyse the efficiency of the districts in germany. My tests proposes to use a time varying model from Kumbhakar (1990) or the time varying model from Battese and Coelli (1992).
I am using a cobb douglas specification.
I use the following code:
sfpanel $Output $loglist, model(bc92)
scalar ll_bc92 = e(ll)
predict ineff_BC, u
predict eff_BC, bc
summarize eff_BC
This code works well, I get the following results
Time-varying decay model (truncated-normal) Number of obs = 3540
Group variable: id Number of groups = 354
Time variable: time Obs per group: min = 10
Prob > chi2 = 0.0000
Log likelihood = 6076.9113 Wald chi2(5) = 74.14
----------------------------------------------------------------------------------------
LogSMRInv | Coefficient Std. err. z P>|z| [95% conf. interval]
-----------------------+----------------------------------------------------------------
LogGP | -.0155696 .0143281 -1.09 0.277 -.0436521 .0125129
LogSpecialists | .011152 .0103765 1.07 0.282 -.0091856 .0314896
LogBeds | - . 0571862 .007225 -7.92 0.000 -.071347 -.0430254
LogDialyseAnzahl | .0015767 .0070368 0.22 0.823 -.0122151 .0153686
LogSpecialityDiversity | .0136673 .0057841 2.36 0.018 .0023306 .0250039
_cons | -2.527519 .075975 -33.27 0.000 -2.676428 -2.378611
-----------------------+----------------------------------------------------------------
/lnsigma2 | -4.517443 .0721543 -62.61 0.000 -4.658863 -4.376023
/ilgtgamma | 2.069327 .0857345 24.14 0.000 1.90129 2.237363
/mu | .3156293 .0181876 17.35 0.000 .2799822 .3512764
/eta | -.0037632 .0012755 -2.95 0.003 -.0062631 -.0012633
-------------+----------------------------------------------------------------
sigma2 | .0109169 .0007877 .0094772 .0125753
gamma | .887886 .0085344 .8700375 .9035549
sigma_u2 | .009693 .0007879 .0081487 .0112373
sigma_v2 | .0012239 .0000308 .0011636 .0012843
------------------------------------------------------------------------------
The efficiency results are around 0,73 (73%)
Variable | Obs Mean Std. dev. Min Max
-------------+---------------------------------------------------------
eff_BC | 3,540 .7364692 .0714501 .5491924 .9892062
Can I use this models since some of the coefficients are negative? I dont understand the effect from the variables onto the mortality rate (health output). For example more hospital beds do not improve the health output. Does this mean there are enough beds?
And are the results or the coefficients significant? I think i need to interpret the z-value.
