R help - Oct 2014 - comparing two "half-normal production" stochastic frontier functions

Hi

I am using the function frontier::sfa (from the package frontier) to
estimate several "half-normal production" stochastic frontier
functions.

Now I want to compare the coefficients of the linear frontier function
and see if they are different.

According to my stackexchange (CrossValidated) question [1] I can
compare these as I can compare a normal linear regression. 

In R, I would uswe the function anova to do this model comparison -
correct?

Now this function does not accept objects of the type 'frontier' - so
how can I do this comparison in R?

To re-iterate, I want to know if the coefficients of the frontier line
(slope and intercept) are significantly different.

Below please find a reproducible example based on data provided in the
package, of what I did, and below the transcript.

Thanks,

Rainer

--8<---------------cut here---------------start------------->8---
library(frontier)
data(front41Data)
dat1 <- front41Data[1:30,]
dat2 <- front41Data[30:60,]
x1 <- sfa(log(output) ~ log(capital), data=dat1)
x2 <- sfa(log(output) ~ log(capital), data=dat2)
x1
x2
anova(x1, x2
--8<---------------cut here---------------end--------------->8---

,----
| > library(frontier)
| > data(front41Data)
| > dat1 <- front41Data[1:30,]
| > dat2 <- front41Data[30:60,]
| > x1 <- sfa(log(output) ~ log(capital), data=dat1)
| > x2 <- sfa(log(output) ~ log(capital), data=dat2)
| Warning message:
|   In sfa(log(output) ~ log(capital), data = dat2) : 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
| > x1
| 
| Call:
| sfa(formula = log(output) ~ log(capital), data = dat1)
| 
| Maximum likelihood estimates
|  (Intercept)  log(capital)       sigmaSq         gamma  
|       2.8646        0.2642        0.4364        0.8243  
| > x2
| 
| Call:
| sfa(formula = log(output) ~ log(capital), data = dat2)
| 
| Maximum likelihood estimates
|  (Intercept)  log(capital)       sigmaSq         gamma  
|       2.7035        0.4550        0.9736        0.9972  
| > 
| > anova(x1, x2)
| Error in UseMethod("anova") : 
|   no applicable method for 'anova' applied to an object of class
"frontier"
`----




Footnotes: 
[1] 
http://stats.stackexchange.com/questions/117319/comparing-coefficients-of-linear-stochastic-frontier-production-and-cost-funct

-- 
Rainer M. Krug
email: Rainer<at>krugs<dot>de
PGP: 0x0F52F982
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 494 bytes
Desc: not available
URL:
<https://stat.ethz.ch/pipermail/r-help/attachments/20141003/f9e25f8c/attachment.bin>

Dear Rainer

On 3 October 2014 14:51, Rainer M Krug <Rainer at krugs.de>
wrote:
> I am using the function frontier::sfa (from the package frontier) to
> estimate several "half-normal production" stochastic frontier
functions.
>
> Now I want to compare the coefficients of the linear frontier function
> and see if they are different.
>
> According to my stackexchange (CrossValidated) question [1] I can
> compare these as I can compare a normal linear regression.
>
> In R, I would uswe the function anova to do this model comparison -
> correct?
>
> Now this function does not accept objects of the type 'frontier' -
so
> how can I do this comparison in R?
>
> To re-iterate, I want to know if the coefficients of the frontier line
> (slope and intercept) are significantly different.
>
> Below please find a reproducible example based on data provided in the
> package, of what I did, and below the transcript.
>
> --8<---------------cut here---------------start------------->8---
> library(frontier)
> data(front41Data)
> dat1 <- front41Data[1:30,]
> dat2 <- front41Data[30:60,]
> x1 <- sfa(log(output) ~ log(capital), data=dat1)
> x2 <- sfa(log(output) ~ log(capital), data=dat2)
> x1
> x2
> anova(x1, x2
> --8<---------------cut here---------------end--------------->8---
library( "frontier" )
data( "front41Data" )

# estimate pooled model
mp <- sfa( log(output) ~ log(capital), data = front41Data )

# create a dummy variable
front41Data$dum <- rep( c( 1, 0 ), 30 )

# estimate model with different intercepts and different slopes
# but the same sigmsSq and the same gamma
md <- sfa( log(output) ~ log(capital)*dum, data = front41Data )

# likelihood ratio test
lrtest( mp, md )


If you have further questions regarding the frontier package, you may
also use the "help" forum at frontier's R-Forge site:

https://r-forge.r-project.org/projects/frontier/

...and please do not forget to cite the frontier package in your
publications (see output of the R command
'citation("frontier")').

Best regards,
Arne

-- 
Arne Henningsen
http://www.arne-henningsen.name