Problems with lclogitml2 and estimating the WTP using lclogitwtp2

[Maike Linhoff]

Dear all,

I am running a discrete choice model using lclogit2 and lclogitml2. I tested the model and decided to run it with 3 classes and to make the price-coefficient class-invariant or in other words, to include price as homogeneous coefficients that is identical across classes.
I figuered out that it is not possible to use the postestimation command lclogitwtp2 when using the above described specification.
Error:
lclogitwtp2, cost(price) post
__000002 not found


I tried to implement a constraint

constraint 1 [Class1]price = [Class2]price = [Class3]price

, which, however, only constrains the price-coefficient of the first and second class to be class-invariant.

I also tried to implement 2 constraints

constraint 1 [Class1]price = [Class2]price = [Class3]price

constraint 2 [Class1]price= [Class3]price

, which also did not work.

The entire code applied is the following

lclogit2 choice, rand(price rate_2 strike_level channel_radio channel_randp ///
    certification_gov certification_ngo ASC) group(idchoice) id(id0) nclasses(3)seed(1234567890)
matrix start3 =e(b)

constraint 1 [Class1]price = [Class2]price = [Class3]price
lclogitml2 choice, rand(price rate_2 strike_level channel_radio channel_randp ///
    certification_gov certification_ngo ASC) group(idchoice) id(id0) nclasses(3) ///
    from(start3)constraint(1) seed(1234567890)

lclogitwtp2, cost(price) post

Is there a proper way to estimate the WTP when using a latent class model with 3 classes and a class-invariant price coefficient using the constraint option?

Thanks in advance!!
Maike

[Hong Il Yoo]

Hi Maike, thanks for bringing this to my attention. There is a little bug in the loop which causes the problem when price is the only class-invariant coefficient. I will fix that and release a new version soon. As far as I can check, however, -lclogitwtp2- works fine with the constraint approach that you have suggested. Please make sure that each constraint defines one restriction, instead of two different restrictions. Also, if you'd like to impose both constraints 1 and 2, you should specify -constraint(1 2)- option.

Edited: OK I now have a revised version of -lclogitwtp2.ado- on my hard disk. I will let you know once it becomes available for -ssc install-. If you need the file urgently, you can email me at [email protected].

[Hong Il Yoo]

The updated version of -lclogitwtp2.ado- is available for downloads now, thanks to Kit Baum as usual. Please type -ssc install lclogit2, replace- at the command prompt.

[Maike Linhoff]

Dear Hong II Yoo,

thank you a lot for your immediate action, now it works perfectly!!

[Wizaso Munthali]

Dear Hong II Yoo,

I'm using the lclogit2 command to determine latent classes from Best Worst Scaling data. My study solicits improved cookstove preferences from 28 attributes. I have succeeded in running the lclogit2 command for a 5-class model. And I have both the response variable effects per class and covariate effects per class. I am now running the lclogitml2 command to get standard error values. Each iteration so far is taking about 2 to 3 hrs to appear. Why is it taking so long to produce output? I have a total of 50,400 observations with 34 variables. How can I speed up the analysis? Take note I am not a statistician so please try and explain this to me in a simple manner. Thank you.