Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Calculating the percent certainty for BWS-Case 2 in hierarchical Bayesian

Hi,

I have just come to know that the percent certainty is an important statistic reported in academic studies and I wanted to calculate it for a profile-case BWS.

I was referred to the following post to understand the procedure:
https://sawtoothsoftware.com/forum/12252/rlh-and-percent-certainty

I found the post very helpful in explaining the meaning of RLH and the Percent Certaininty through an example and how to calculate them for a logit model for a CBC task.

When trying to apply it for the HB I am confused about how to get the likelihood for each respondent and since I do not have any alternatives (but 6 levels to choose one from as best and one as worst) how would this fit in the worst ln?

I looked at the results of the HB from the stand-alone program and I found that they report the percent certainty for each iteration done before and after convergence. Can this be of any use when calculating the overall Percent Certainty of the model?

Sorry for asking the question again and thanks for your patience.

Regard's
asked Feb 11 by AMYN Bronze (2,980 points)

1 Answer

+1 vote
 
Best answer
Well, in each question you DO have alternatives.  If you have 6 items (levels) to choose from, then you have one choice of the best against the other 5 (so the probability for the calculation of the null likelihood is 1/6) and you have another observation of the choice of the worst among the six (again 1/6).

This gives you the raw material you need to calculate the likelihood of the null model, which you will need for your rho-squared formula.
answered Feb 11 by Keith Chrzan Platinum Sawtooth Software, Inc. (90,475 points)
selected Feb 13 by AMYN
Thanks for the explanation
...