To Use BugsXLA6 for Conjoint Analysis

[krishna mohan]

Hi

First time using BugsXLA. I have been using Sawtooth when I needed to build HBayes / Random Effects models where individual level parameters needed to be estimated.

My data is structured as follows:

ID task stim x1 x2 x3 x4 x5 y1 y2 y3 yall y

Data was collected from 30 students, each student completed 8 tasks. In each task there were 3 different product stimuli created by combining different levels of x1 to x5 (all factors, x1 has 2 levels, x2 to x4 have three levels each).

The choice made by the student is recorded in the Variate y which has been recoded into three different binary variables y1 (when y=1, else =0), y2 (when y=2, else =0), and y3 (when y=3, else =0).

My dependent variables variables I wanted to use is y, but BugsXLA6 required >2 variates, hence I did the recode.

Here are my model specifications (the pictures may be too small, I will attach the images with my data file in a CSV format.

1. I have chosen y1,y2,y3 as my response variables

2. I have chosen the multinomial + logit as my distribution and link function

3. My random factors are x1 to x5 plus ID since I'm interested in individual level parameters

4. My priors for the mean of x1-x5 and ID is a Normal Distribution and the Tau is Half-N which came as the defaulta.

My big question: Am I setting this up and doing it right? If not, what would I change? Is my attempt at attempting HBayes model analysis even realistic?

Any help would be greatly appreciated.

Krishna

[Phil Woodward]

Hi, I cannot comment on how appropriate your model is for your data, but it appears you could use the categorical rather than multinomial error distribution and then enter y rather than the constructed variables y1,y2,y3.  If the responses are ordered than you could use Ord Cat. Phil

[krishna mohan]

Thanks Phil for your feedback. I will follow through on your some of your suggestions and check what happens.

Krishna