Interval estimation

[T_FIELD]

Dear Tom and colleagues,

Could you help me?
I would like to do CVM (contingent valuation method) analysis.
So, following UG-383 in Users Guide 810 and ver3p217.rpf, I perform interval estimation using LDV instruction.

However, I can't obtain good results. Could you tell me why?
I attach the rpf file and data file.

Thank you for your cooperation.

[TomDoan]

Your "dependent variable" (AGE) doesn't seem to have anything to do with the upper and lower limits, and is of a completely different scale. The dependent variable is used for limited purposes, but one is to get an initial guess for the variance, and here it ends up being off by four orders of magnitude which causes problems. If you use anything with the scale right (average of low and up; or up or low), it will work.

[T_FIELD]

Dear Tom,

Thank you so much for your advice.
As shown in the attached rpf, I got a good estimation.

By the way, in UG-383 in Users Guide 810, we find,

***************************************************
Note that you still need a dependent variable series, though its only real purpose is to show which entries should be used in estimation: any entry with a missing value in the dependent variable is skipped. The following, for instance, is from Verbeek (2008, example7.2.4). The data were generated by starting with an initial bid (BID1).
***************************************************

The first sentence made me adopt a variable irrelevant to the "upper" and "lower" variable.
But, actually, the last sentence is so important. I think you should emphasize this point more.

Anyway, your comment is so helpful.
Thank you so much.

Happy holidays,

[T_FIELD]

Dear Tom,

Sorry, I guessed wrong.
One problem still remains.
Why do I get zero variance for SIGMA?

Any comments are welcome.

Thank you for your cooperation.

[TomDoan]

It looks like minus the Hessian for the interval estimation isn't necessarily positive definite, the way it is for other models that can be done by LDV. I assume you are planning to do a model with something other than just a constant. It's quite possible that once you have a model with better explanatory power that the issue will go away.