Have an idea?

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

What conjoint method is most feasible for different competitor sets across regions?

Dear Sawtooth-Team,

in a B2B context we are conceptualizing a conjoint study to measure our clients brand strenght with two major restrictions:
- each region two be surveyed has a different set of competing brands
- per region the population size i.e. potential sample will be rather small (around 30-50)

To accommodate our clients need for estimation precision and to avoid the use of several conjoint designs with only few respondents each, we would like to try a single conjoint design that is feasible for all regions.

a) One idea was to use ACBCs ability to work with constructed lists and to feed only the relevant brands for each region from a long list of all brands to the corresponding respondents.
b) Another idea uses CBC and would limit the number of brands to, say, 5 and than uses conditional display to show relevant brand names and the corresponding prices for each region that way.

Regarding a): Can I apply ACBC to handle the described situation and, if so, what would I have to consider? I am sceptical because wouldn't the clients brands utility be boosted just because its the only brand that is "pre-selected" in each region?

Regarding b): With the conditional display the utility is estimated for a generic brand attribute level 1 that has different effect depending on the region. To ensure these differences are still pronounced in a HB estimation I can use prior df and variance to shift weight to individual responses rather than population averages, correct? Also I can apply covariates indicating the region (while I understand from numerous papers that the effect of covariates is usually very small). What else can I do?

Do you have an idea for an "alternative c)"?

Thanks for comments on my thoughts in this matter!
asked Jul 24, 2014 by alex.wendland Bronze (2,080 points)

1 Answer

0 votes
Best answer
To deal with the small sample size, it seems natural to want to estimate parameters across the full population (using HB), but also to customize the design (the brands seen) to be those relevant to each respondent.  

When you use a constructed list in ACBC to customize the levels to be those relevant to the respondent, you have the choice as an analyst regarding how missing levels to each respondent are handled: missing at random, missing inferior, or missing unavailable.  (See documentation for details, though "missing unavailable" is the most extreme way to deal with missing levels...meaning their utility is set to be most different (lower) from "included" levels for each respondent).  If your client’s brand is available in all regions, then it would appropriately be handled as such in the design.  If a conjoint study artificially builds awareness for each brand in the survey, then it is possible it could get an artificial leg up on the other brands due to its availability for all respondents.

I would recommend using region as a covariate in the estimation.
answered Jul 24, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (162,390 points)
selected Oct 8, 2014 by alex.wendland