If the only tool you have is
a hammer, you tend to see every problem as a nail.
Abraham Maslow, 1966
DCM - Discrete Choice Modeling
|Discrete choice modeling has broad
uses in marketing research applications due to its potential to get
insight into the perceptions and behavior of customers. The most
distinctive advantages of choice-based methods can be summarized as
- Choice is a natural manifestation of human behavior.
- Choice is an action, in contrast to a statement.
- Choice imposes minimal constraints on the response.
- Choice is scale-free.
- Choice is non-verbal.
- Choices are less culturally, conventionally or habitually
dependent than ratings or statements.
- Overstatements or understatements are suppressed.
The general belief is choice-based models can give results much closer
to real behavior than ratings based models. However, a choice is less
informative than a stated value. More choice data must be collected to
obtain an equivalent amount of information. Models for discrete events
quite differ from models for metric values and imply different
limitations. Results have often a format different from that known from
traditional metric approaches. This all leads to uneasiness in marketing
project managers who plan marketing research surveys and apply the
The objective of these web pages is to provide end-users of marketing
research with an overview of selected DCM-related methods, to encourage
exploitation of the methods, and to bridge the gap between the academic
and commercial approaches to the subject.
|The following aspects of DCM are dealt with.
- Basic assumptions, conditions, properties and prospects.
- Pros and cons of various approaches.
- Topics concerning applications.
- Understanding and use of selected types of results.
The described methods make part of the services offered by the g82,
s.r.o., a Czech based marketing research and consultation company.
The most popular methods thriving on DCM are variants of choice based
conjoint that have displaced metric conjoint methods to a large extent. If
you would like to get acquainted with the basics of goals and uses of
conjoint analysis you might wish to see a lightweight white-boarding video
(5'36'') from Sawtooth Software on YouTube.
- Instead of a summary of subjective statements, the importance of
image attributes can be derived from the purchase behavior using
the standard image questionnaire data and the analytical model of
conjoint value analysis. An example is presented.
- Handling of the "None" constant alternative in
a preference share simulator of substitutables using a simplified
two-level nested logit.
- A MaxDiff-based method to estimate preferences between product
concepts and the concept aspects has been developed. The method can
handle a higher number of attributes than a CBC. A description,
some results, and a live
demo questionnaire are available.
- Question for Ranked
Grid Analysis (RGA) is designed as a two-dimensional check-box
grid with ranking. Ranking is possible in either horizontal or vertical
or both dimensions. Any number of exclusive choices can be added.
Examples of estimated brand and feature profiles from two market
research studies show ranking leads to higher sensitivity and
discrimination between brand-feature combinations. Link to a demo
questionnaire is available.
many CBC tasks should be asked? Two easy rules of thumb and one
based on the information entropy are presented.
- Estimation of a possible
cannibalization maximum using CBC. The multinomial model is
extended with a nest containing items involved in cannibalization and
set as perfectly correlated.
- A relaxed
non-compensatory modeling and simulation has been introduced as a
compromise between fully non-compensatory and compensatory models. An
Excel-based choice simulator with three simulation models is available
Trial and Repeat Dissipation simulator has been made public.
Besides the Excel-based sales simulator, the download includes an
enlightening .pdf file.
- An example of a
non-compensatory modeling and preference simulation, based on
estimation of thresholds of acceptability of attribute levels.
- A simple comparative example of obtaining preference
and acceptability for non-compensatory modeling.
- A standalone example of clustering based on relative perceptions
obtained by ranking
choices with ties from a large scale question battery.
- A method for DCM analysis of ranked
choices with ties, typically from Q-sort exercises or large scale
- An example of
stated and evoked perceptions of attribute levels obtained by
Scale Discrete Choice Analysis has been amended with estimation of
perceptual thresholds and a way to encompass larger number of
- Object Image Analysis, a DCM-based solution, has been
introduced. Results from a test of four consumer electronics brands
(Apple, Lenovo, Samsung and Sony) are presented. A live updated
version of the questionnaire in English is available.
- Based on latest experience the page on Common
Scale Discrete Choice Analysis and live questionnaire
demo have been updated.
- DCM Blog Cz with anonymous acces
for DCM sympathizers speaking Czech has been started.
- An optimal consideration
set at the time of making decision consists of 3 alternatives. A
proof is given.
- DCM Portfolio
Optimization algorithm has been enhanced to obtain a persuasive
subset of items by taking account of the confusion due to a too wide
range of offers (aka "tyranny of choice").
- Overview of the Common
Scale Discrete Choice Analysis allowing comparison of level
preferences across attributes. A live questionnaire
demo is available.
- Example of Elasticity
of Substitution used as generalized importance of quantitative CBC
- A robust CBC based estimation method of an Optimal
- A live demo of a parametric
CBC for a bank loan study.
Any findings or conclusions expressed on these pages are
solely the opinion of the author and may not be shared by the marketing
research company G82, s.r.o.