WTP space using bayes on individual base

[Jens Petersen]

Hello,

i try to calculate a WTP on individual base from choices. I am using Sawtooths Software for this (using hierachical bayes estimation data and calculate the WTP on individual base from the share preferences which are calculated from the utilities), but the results are a bit implausible. I have a few outliers (10% in total) on both sides (negative and positive). I want to make a group comparison but this outliers affect the means a lot and give me really flawed results. I can use medians to get rid of the outliers, yes. According to my understanding of statistics, this is not a good approach to make a comparison. I can not delete this outliers, because i can not give a good reason why i am doing this. the outliers are a result of the transformation of utilities to dollar values.

So I found all the discussion on wtp space and I want to try if this gives me better results. Before i transform the collected data to fit to stata, i have a few questions:

1) is it possible to get individual results (part worths) from Stata by using bayesmlogitwtp ?
2) where are this results?

Or is is not possible to get utilities on individual base with Stata? I tried this on the examples from Arne Risa Hole, but yes. It does not work. I am not sure if i am not able to work with Stata or if it is not possible to make such calculations with stata.

Maybe someone can help me with this. Thanks.

[Carlo Lazzaro]

Jens:
welcome to this forum.
As far as I know, the theoretical and Stata guru in this field is Arne Risa Hole (see: https://www.sheffield.ac.uk/economics/people/hole/stata).

[Jens Petersen]

Because i cannot edit the starting:

By "get individual utilities" I mean individual coefficients in wtp space for each respondent without loosing information. So exact the same calculations like sawtooth does with HB estimation, but in wtp space instead of preference space.

The segmentation data I want to use is a satisfaction model (7 point likert scales). I have 12 choice observations per respondent. Some segments of the satisfaction model are quite small, so i thought to group them on a 5point scale.

Segment	1	2	3	4	5
n	22	58	172	203	152
observations	264	696	2064	2436	1824

But i am not sure if i should run a estimation for each segment, or code the segmentation data ...???

[Carlo Lazzaro]

Jens:
sorry I cannot be more helpful than my previous reply.
Hopefully Arne will chime in.