Have an idea?

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

ACBC Design Efficiency

Dear Forum,

following up on I discussion I recently read about ACBC design efficiency,  I´d be grateful for your opinion on the efficiency of the following design.

In my ACBC (8 attributes, 4 prohibitions) I conducted the design test with the following results:
- Out of 300 robotic tests, only 1 respondent had a minium of 1 view/level. Thus, the design in general is efficient but partially sparse.
(Is that a sound conclusion?)

- The highest standard error for a level is 0.037 (average of 0.025) and thus below 0.05

- The average D-Efficiency is ~0.979

In this regard, I have a couple of questions.
*What is the maximum possible value for the D-Efficiency?

*In general these results indicate a valid and rather efficient design, right?

*After the real fielding, do you recommend to analyze these indicators again and where do I find the standard error and d-efficiency for the real data?

Thank you very much in advance for your support.

asked Oct 24, 2018 by Uwe

1 Answer

0 votes
Highest possible d-efficiency is 1.0.

Your results to me suggested a well-powered ACBC experiment, both in terms of enough information per respondent for very good estimation at the individual level, and enough information across respondents for population interferences.

The only way to estimate aggregate logit parameters on the real ACBC data (to check your standard errors from the aggregate model) would be to export your ACBC data to a .CHO file, to open that .CHO file in the Latent Class Standalone Module, and to estimate a 1-group latent class solution.  I would not bother to do that effort.
answered Oct 24, 2018 by Bryan Orme Platinum Sawtooth Software, Inc. (177,015 points)