Have an idea?

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

Linear CBC HB estimation

I am considering doing a CBC study that will contain 5 attributes - 2 of which I would like to estimate (in LightHouse Studio's HB analysis) as 1 linear utility per attribute (instead of part worth utilities for each attribute level).

The 2 attributes will be something like:
1. Price (4 levels: $24,000, $27,000, $30,000, $33,000)
2. Warranty (3 levels: 1 year, 3 years, 5 years)

My questions are:

(1) I would like to show relative attribute level preference (that is, zero centered diffs) for each tested level of Price and Warranty - as would be obtained from HB analysis if Price and Warranty were estimated as part worth utilities. Since I am estimating  Price and Warranty as linear utilities, zero centered diff results only contain 1 column for Price and 1 column for Warranty (instead of separate columns for each attribute level). Is it common/acceptable to run HB analysis 2 different ways for reporting - once with Price and Warranty as part worths to obtain zero centered diffs for each attribute level (to use for level preference reporting) and once as linear (to obtain raw utilities to use in preference share simulations/reporting)?

(2) When using an exponent in an Excel simulator - at the respondent level, you multiply the raw part worth utilities for each attribute level by the exponent and then proceed with preference share computations as usual. How do you apply the exponent with linear utility estimation since you do not have part worth utilities (for example, do you multiply the 1 linear coefficient for the attribute by the exponent before proceeding with preference share computations)?
 
Thank you.
asked May 17 by anonymous

1 Answer

0 votes
It's not common to report utilities estimated one way (with part-worth coding) and use utilities estimated with linear terms for warranty and price in a market simulator.  Despite the fact that this is not a common thing to do, depending on your aims it could be a reasonable thing.

It looks like you are on good footing to remember that you need to use the raw HB utilities rather than zero-centered diffs to conduct market simulations.  

The exponent is a multiplier that you apply to all utilities (prior to the summation and exponentiation step of the logit equation) , whether they are estimated as part-worth (dummy-coded) or linear terms.
answered May 18 by Bryan Orme Platinum Sawtooth Software, Inc. (163,315 points)
...