Forum: help

RE: Effciency Effects Frontiers (EEF) [ Reply ] By: Arne Henningsen on 2017-03-15 09:02	[forum:43937]
Dear Luca Thanks for your message. Indeed this is a typo in the lecture notes. I have fixed this in the draft version of the second edition of my lecture notes ("[...] between 1.4 and 6.4 percentage points more efficient [...]") and I hope that I will be able to update the version on leanpub to the second edition soon. Best regards, Arne

RE: Effciency Effects Frontiers (EEF) [ Reply ]
By: Luca Elisei on 2017-03-14 17:49

[forum:43935]

Hi Arne,
I'm interested to know how you estimated the marginal effect on your lecture notes "Introduction to Econometric Production Analysis with R".
Your data show that apple producers who use an advisory service are between 6.3 and 6.4 percentage points more efficient than apple producers who do not use an advisory service.
I tried to replicate the commands but I'm not able to obtain the same result. Follow my commands and results :
dat$effCDAdv2 <- efficiencies( prodCDSfaAdv, margEff = TRUE )
hist( attr( dat$effCDAdv2, "margEff" ), 20 )
mean(attr( dat$effCDAdv2, "margEff" ))
0.04584022

what is wrong?

thanks for your support,

Luca

RE: Effciency Effects Frontiers (EEF) [ Reply ] By: Luca Elisei on 2017-01-09 14:53	[forum:43786]
Dear Arne, I Redefined the variable mcs (0-1), and the problem was solved. thank you very much for your support BR Luca

RE: Effciency Effects Frontiers (EEF) [ Reply ]
By: Arne Henningsen on 2017-01-07 16:35

[forum:43784]

Dear Luca

Your estimates do *not* show that the two management practices do *not* differ in terms of efficiency. However, the Wald test indicates that the estimated difference is not statistically significantly different from zero at the 10% significance level, i.e. the 90% confidence interval of the coefficient of 'mcs' includes the value zero.

You could do the following to check the robustness of your result:

1) Test the statistical significance of 'mcs' by using a likelihood-ratio test (i.e. comparing models with and without 'mcs' as explanatory z-variable), because the likelihood-ratio test may have better finite-sample properties than a Wald-test.

2) Redefine the variable that indicates the management practice as mcs2 = mcs - 1 so that that mcs2 is a traditional dummy variable with only zeros and ones. Theoretically, this should only affect the intercept of the inefficiency model but should not affect anything else. If it considerably affects other estimates, something is wrong (e.g. no convergence achieved in at least one of the estimations).

3) Re-estimate the model without an intercept in the inefficiency model (i.e. 'mcs' is the only explanatory variable in the inefficiency model), because a Wald test indicates that the intercept of the inefficiency model is statistically insignificant at 10% significance level, and use a likelihood-ratio test to check whether the data reject this.

4) Redefine the variable that indicates the management practice as mcs3 = 1 - mcs2 so that that mcs3 is one if mc2 is zero and the other way round and re-estimate the model with mcs3 instead of mcs or mcs2, both with and without intercept.

Best,
Arne

Effciency Effects Frontiers (EEF) [ Reply ]
By: Luca Elisei on 2017-01-06 17:21

[forum:43778]

Dear Arne,

I need to assess whether, the two management practices (mcs) differ in terms of efficiency.

I know that my model have a convergence problems (as show below).
My question is: Can I say that there is no difference in efficiency between the two management practices?

my model:
effcd <- sfa( log(va) ~ log(inv) + log(emp) | mcs , data=mydata)

where : va = value added;
inv = fixed capital;
emp= number of employees
mcs = managerial practices, as dummy variables (1;2)

effcd_mc <- sfa( log(va) ~ log(inv) + log(emp) | mcs, data=mydata)
Warning messages:
1: In log(va) : Si è prodotto un NaN
2: In log(inv) : Si è prodotto un NaN
3: In sfa(log(va) ~ log(inv) + log(emp) | mcs, data = my data) :
the parameter 'gamma' is close to the boundary of the parameter space [0,1]: this can cause convergence problems and can negatively affect the validity and reliability of statistical tests and might be caused by model misspecification
4: In sfa(log(va) ~ log(inv) + log(emp) | mcs.clusters, data = effPanel2) :
the covariance matrix of the maximum likelihood estimates is not invertible

summary( effcd_mc, extraPar = TRUE )
Efficiency Effects Frontier (see Battese & Coelli 1995)
Inefficiency decreases the endogenous variable (as in a production function)
The dependent variable is logged
Iterative ML estimation terminated after 126 iterations:
log likelihood values and parameters of two successive iterations
are within the tolerance limit

final maximum likelihood estimates
Estimate Std. Error z value Pr(>|z|)
(Intercept) 4.2843e+00 1.2722e-01 33.6760 < 2.2e-16 ***
log(inv) 4.8247e-02 1.1149e-02 4.3274 1.509e-05 ***
log(emp) 9.5765e-01 3.0062e-02 31.8561 < 2.2e-16 ***
Z_(Intercept)-1.2458e+03 8.2300e+02 -1.5137 0.130105
Z_mcs2 1.8560e+02 1.2242e+02 1.5161 0.129498
sigmaSq 2.9152e+02 1.9276e+02 1.5123 0.130452
gamma 9.9955e-01 3.1782e-04 3145.0108 < 2.2e-16 ***
sigmaSqU 2.9139e+02 1.9277e+02 1.5116 0.130637
sigmaSqV 1.3206e-01 6.5599e-03 20.1310 < 2.2e-16 ***
sigma 1.7074e+01 5.6450e+00 3.0246 0.002489 **
sigmaU 1.7070e+01 5.6464e+00 3.0232 0.002501 **
sigmaV 3.6340e-01 9.0258e-03 40.2620 < 2.2e-16 ***
lambdaSq 2.2065e+03 1.5488e+03 1.4247 0.154249
lambda 4.6974e+01 1.6486e+01 2.8494 0.004381 **
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
log likelihood value: -571.265

panel data
number of cross-sections = 242
number of time periods = 4
total number of observations = 955
thus there are 13 observations not in the panel

mean efficiency of each year
2011 2012 2013 2014
0.8000790 0.7973765 0.8063520 0.8164916

mean efficiency: 0.8051313

> lrtest( effcd_mc )
Likelihood ratio test

Model 1: OLS (no inefficiency)
Model 2: Efficiency Effects Frontier (EEF)
#Df LogLik Df Chisq Pr(>Chisq)
1 4 -628.96
2 7 -571.27 3 115.39 < 2.2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

***both models (Cobb-Douglas and Translog), whit or whitout Z intercept, lead to the same issue.

Thanks Arne,

BR,

Luca