Have an idea?

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

Basian Statestical Tests for MaxDiff Study

Hi,

I was conducting the 95% confidence interval for the HB of a MaxDiff study that I am doing and used the instructions according to Chapter 12 of Bryan Orme and Keith Chrzan's 2017 book, Becoming an Expert in Conjoint Analysis: Choice Modeling for Pros.  from this link [https://www.sawtoothsoftware.com/169-support/technical-papers/cbc-related-papers/2033-statistical-testing]

I used the alpha draws and the results were reasonable for the unconstrained model. But when I tried the constrained models where I added several constraints, the results become unreasonable for some levels as the Interval dose not contain the model estimate values, Should I then use the mean beta draws instead? what is the justification for that? when I add covariates to me constrains what should I do to calculate the 95% Confident interval?

N.B. I used the standalone program (CBC/HB) to generate the data

Tahnks,
Regards'
asked Jan 8 by AMYN Bronze (2,115 points)

1 Answer

+2 votes
The method of Simultaneous Tying that we use in Sawtooth Software's constrained HB runs means that only the lower-level (beta) estimates should be used.  The upper-level (alpha estimates) should not.  They become erratic.

If you want to develop confidence intervals from alpha draws, then do not run constrained estimation via "simultaneous tying".
answered Jan 8 by Bryan Orme Platinum Sawtooth Software, Inc. (170,215 points)
edited Jan 8 by Bryan Orme
...