Have an idea?

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

Should I try to avoid combinations that are clearly "better"?

I'm trying to run a Discrete Choice Conjoint Analysis for a financial insurance product. When coming up with the cards for the experiment design how do you avoid combinations that are clearly "better"?

For instance if your attributes are

**Benefit Amount** - ($1000, $2000, $3000)

**Benefit Period**  - (10, 20, 30)

**Price** - ($5, $10, $20)

You could potentially get a combination of

$1000, 10, $20


$3000, 10, $5

which seems like a no brainer to select the second option. Should I be trying to avoid this, and if so, how?
asked Nov 22, 2017 by anonymous

1 Answer

0 votes
You could avoid this if you like by using prohibitions.  Be careful doing this, however, because if you put in too many prohibitions you'll start to seriously decrease your ability to estimate utilities precisely.  

Part of what goes into the precision of utility estimates in any statistical model is the uncorrelatedness/independence/orthogonality of the predictor variables.  In choice experiments we use designs that make for independent predictors.  The more prohibitions you add, the less independent are your predictors, and the lower is the precision of your utility estimates (as measured by their standard errors).

If you add prohibitions just make sure to test your design (our Lighthouse Studio software has a built-in design test functionality for choice-based conjoint experiments) and examine the quality of the design and the size of the standard errors as described in the help menu.
answered Nov 22, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (74,725 points)