I am trying to account for attribute non-attendance for my simulated choice data. Herefore, I am using the eaalogit command by Hole (2011). It estimates an endogenous attribute attendance model.
My used code:
- Code: Alles auswählen
eaalogit choiceb0p1 price brand, group(choiceset) id(ID) keaa(2) eaaspec(x1 x2) trace
The independent variable choiceb0p1 states that in my data simulation, in the choice no attention was paid towards variable brand, full attention towards price - b0p1. I would also like to vary this attention distribution so that choiceb1p0, choiceb1p1, or choiceb0p0 result following the same logic.
How I simulated the data: Price is normally distributed around the mean=40 with a standard deviation of 10. Brand level is a dummy variable, so there are only two brands available. The error is gumbel-distributed. The utility of an alternative is calculated based on a deterministic part from the attributes and their defined beta-levels and a random error. The choice then selects the utility-maximising alternative based on different attention levels.
The usual error:
- Code: Alles auswählen
could not calculate numerical derivatives -- discontinuous region with missing values encountered r(430)
Also, iterations stating 'not concave'.
Reading another post, I already traced the iterations and in the example above modeled price with full attention.
- Code: Alles auswählen
(eaaspec x1 one)
I attached a sample of my data set for you to understand better. It contains the choice decisions modeling the different attention levels. Only one is used per estimation obviously.
If there are any further uncertainties, please let me know! I am happy to provide more information if needed.
If anyone has a hint how to solve this issue or what I am doing incorrectly, please let me know. I am really new to Stata.
Thank you very much in advance for your time!
Kind regards,
Jenny