[Limdep Nlogit List] Question on coefficient stability

[Jiong Yan]

We have a question about coefficient stability in nested logit model:
interaction of attribute with attitudinal variable.

*In the base model,
*The coefficient of convenience attribute is 0.0882+
*In the interacted model,
*The coefficient of convenience attribute is: 0.6281***
The coefficient of sceptical attitude*convenience attribute is: -0.0931**
The combined value of the convenience attribute and interaction variable
=0.6281+ (-0.0931*5.78) = 0.09.
*5.78 is the sample mean of sceptical attitude dimension.

(For utility estimation, the basic formula is:
Uo=b*a
When including attitudinal variables, the estimation formula is:
Uo=ba*a+bi*a*v (v=attitudinal variable)
            A one step change in "a" on "U" is
    ΔU = ba+ bi*v (v=attitudinal variable)
ΔU depends on respondent characteristic v (here v refers to an attitudinal
variable). To give a one-number impression of ΔU, we multiply bi with
(=mean of v), i.e. we document the effect for an "average" respondent.

Our question is that for average respondent the parameter remains stable,
just re-distributed between the attributes and interaction variables, which
is necessarily collinear. *Could this be regarded as a case of model
misspecification or unstability of main effect?

*Thank you very much for your help.

Best regards.

Jiong Yan

-- 
Jiong Yan
Ph.D Candidate
Department of Agricultural Economics and Rural Development
Georg-August-University Goettingen

Phone+49 (0)551/39-4579
E-mail:yjiong at gwdg.de <E-mail%3Ayjiong at gwdg.de>



-- 
Jiong Yan
Ph.D Candidate
Department of Agricultural Economics and Rural Development
Georg-August-University Goettingen

Phone+49 (0)551/39-4579
E-mail:yjiong at gwdg.de <E-mail%3Ayjiong at gwdg.de>