Running an 'empty' stochastic production frontier model w/technical inefficiency effects

[Amy Thom]

I've fitted a BC92 stochastic production frontier model (see below), and for my results report I need to run an 'empty' iteration of the model, i.e. a null version without predictors (single restriction). Stata's help files have fallen short when it comes to how to do this... does anyone in the forum know the command(s) please? Many thanks!

frontier ln_qtymilk ln_cows1 ln_feed1 ln_totassetresid ln_labindex, uhet(ln_exp ln_zonemilk ln_breedindex1 dv_off_farm_inc)

[Scott Merryman]

Can you provide some more details. I assume you mean something other than just

frontier ln_qtymilk

[Amy Thom]

Hello Scott, and thank you for your reply.

Apologies if there wasn't enough info in my first post. Below is an example table summarizing different versions of a similar inefficiency sub-model, using the 'empty' sub-model as a starting point for comparison of subsequent sub-models. Note: I only need to produce the empty half normal model.

So the frontier model has been fitted, and this table speaks to the fitting of the inefficiency sub-model. I would expect the Stata command to look something like:

frontier ln_qtymilk ln_cows1 ln_feed1 ln_totassetresid ln_labindex, uhet(*)

* but I don't know what belong inside the brackets! I have tried 1, null, empty, etc etc and Stata won't run it. The help section hasn't provided any guidance, unfortunately.

Table 4b: Selecting variables for the inefficiency model using frontier command in Stata

Alternative inefficiency sub-models	LLH value	LR test	Coefficient	z-ratio
Empty – truncated normal	-176.68
Empty – half normal	-176.88
Experience	-174.07	5.62	0.517	2.17
Distance to market	-173.36	7.04	0.088	2.41
D_modern breeds	-173.92	5.92	0.751	2.26
D_off-farm job	-175.55	2.66	-0.553	-1.60
D_diversification	-176.42	0.92	0.112	0.96
D_credit	-176.32	1.12	-0.349	-1.06
D_altitude	-176.76	0.12	-0.142	-0.48
Herd size	-176.62	0.52	0.025	0.71
Household size	-176.15	1.46	0.080	1.18
			AIC	BIC
Experience, distance	-169.06	15.64	354.12	377.93
Experience, modern breeds	-172.07	9.62	360.15	383.96
Experience, off-farm job	-172.85	8.06	361.70	385.51
Distance, modern breeds	-170.37	13.02	356.74	380.56
Distance, off-farm jobs	-172.89	7.98	361.78	385.60
Modern breeds, off-farm jobs	-173.46	6.84	362.92	386.73

*** signify [IMG]file:///C:/Users/amyet/AppData/Local/Temp/msohtmlclip1/01/clip_image001.png[/IMG] , ** indicate [IMG]file:///C:/Users/amyet/AppData/Local/Temp/msohtmlclip1/01/clip_image002.png[/IMG] and * indicates [IMG]file:///C:/Users/amyet/AppData/Local/Temp/msohtmlclip1/01/clip_image003.png[/IMG]

[Amy Thom]

Hello Statalists,

I'm still battling to figure out how to run an 'empty' model so that I can then run LR tests for model selection. I was given the following advice by someone who uses Front4.1 to run stochastic frontier production functions:

"To do it in Stata, you will have to run your production function with Stata's maximum likelihood equivalent of the 'regress' command, and then you fit the production function to obtain the empty model."

So the question is, what is Stata's ML equivalent of regress?

Thanks for any suggestions!

Regards,
Amy

[Andrew Musau]

So the question is, what is Stata's ML equivalent of regress?

regress estimates the linear model using OLS whereas glm (by default) does this using maximum likelihood. See

help glm

[Amy Thom]

Thank you very much, Andrew. I was able to generate this 'empty' model using the glm command as you advised.

However, when I then tried to run LR tests between the empty model and my frontier sub-models, Stata throws an error message reminding me that I cannot run LR tests between models with different estimators. Any suggestions on how to work around this?

[Andrew Musau]

Use the -force- option

lrtest m1 m2, force

[Amy Thom]

Thank you very much, Andrew. Your advice has been spot-on and extremely useful to me. I truly appreciate your suggestions!