Have an idea?

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

Difference between Discrete Choice Model and Conjoint?


I've run into this online comment:

I'll paste the relevant bit:
"The difference between discrete choice models and conjoint models is that discrete choice models present experimental replications of the market with the focus on making accurate predictions regarding the market, while conjoint models do not, using product profiles to estimate underlying utilities (or partworths) instead.

As a result of their focus on market replication, discrete choice models can be counted on for accurate predictions."

Does anyone have a concrete example of what a survey for "discrete choice models" would look like? Is he talking about providing a certain set of circumstances, and respondents select which actually existing product they would purchase in that given scenario?
asked Nov 19, 2014 by Phil

2 Answers

+1 vote

A very similar question came up on this forum last week and you can find the discussion here:  https://sawtoothsoftware.com/forum/7064/difference-between-dcm-%26-cbc.

The comment you cite is vague and a little confusing.  It is also not quite accurate, since it conflates discrete choice models (models with categorical dependent variables) with discrete choice experiments (discrete choice models whose independent variables conform to an  experimental design).  Most conjoint analysts today use some variety of choice-based conjoint models, which are themselves members of the class of discrete choice experiments.  That said, there are also varieties of discrete choice models that are not also conjoint models, because they do not decompose utilities into an attribute-and-levels sort of structure.  

Because these methods have been invented and developed in different fields (psychology, econometrics, marketing) the nomenclature around them has become confusing, partially overlapping and even partially contradictory.  As a result I try not to get hung up on semantics, about which you're not going to get even the experts to agree, and rather focus on finding the methods that address the managerial questions clients need to have answered.
answered Nov 19, 2014 by Keith Chrzan Platinum Sawtooth Software, Inc. (50,675 points)
+1 vote
I think the author is referring to a subset of discrete choice experiments in which the actual products on the marketplace are more closely represented, rather than in a traditional conjoint experiment wherein the attributes are left to vary freely to create all sorts of product concepts that may or may not necessarily be represented in the marketplace.

As an example, imagine a marketplace in which the products could be broken down into six different attributes (with the sixth attribute being price), each with 4 levels.  There are 4^5 = 1024 product concepts (aside from price differences) that could be presented to respondents.

However, with a discrete choice experiment that aimed to more closely align with actual product definitions on the marketplace, maybe the actual products as existing in the marketplace are defined as fixed alternatives on the first 4 attributes (say, 8 possible fixed product alternatives, each described using just one level of each of the first 4 attributes), the 5th attribute perhaps is allowed to vary across its 4 levels, and the 6th attribute (price) also varies (perhaps in a conditional pricing design, so that the ranges of prices are customized for each of the 8 possible fixed alternatives).

With such a design as above, respondents tend to be seeing products in the CBC (discrete choice) survey that are more realistic (and recognizable) as those that currently exist in the marketplace.  This has the benefit of producing a more realistic CBC interview, where the preference captured for the existing products in the marketplace are potentially more accurate.  However, it has the drawback that if the researcher wants to examine what other combinations (other than the 8 fixed product alternatives) could be constructed using the first four attributes, it would be impossible to examine that given the data.

There are hybrids of the same idea above.  For example, maybe four of the product alternatives are fixed across all their attributes; two of the product alternatives are allowed to vary freely on just two of the attributes; and another two are allowed to vary across all attributes.

All the designs I've described above may be accomplished with Sawtooth Software's CBC system with its Advanced Design Module.
answered Nov 19, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (131,390 points)
DCE and CBC: Differences in underlying theory?