Transcript Choice modelling in marketing- some examples Hans S. Solgaard

Choice modelling in marketingsome examples
Hans S. Solgaard,
Department of Environmental and Business Economics
June 2008
Overview
a. What is marketing?
b. The complexity of the choice context in marketing
c.
Brief history of choice models in marketing.
d. Some marketing examples
-Choice of TV program
-An integrated model of purchase incidence, brand choice and quantity decisions
-A model of store choice
Marketing
Because the purpose of business is to create and keep
customers it has only two central functions – marketing
and innovation. The basic function of marketing is to
attract and retain customers at a profit
(P. Drucker; 1999).
Therefore:
It is one of the most important areas of research in
marketing to understand and measure the effects of
consumer/customer choice
Complex choice context
-many choice alternatives
-high number of attributes, features and characteristics to
to characterize alternatives and decision makers.
-consumers are heterogenous
Brief history
Despite the complexity of consumer choice the choice
modelling literature was dominated by fairly simple
stochastic brand choice model until beginning of the 80s.
- parsimonious in behavioral assumptions and in
parameterization.
- contains no decision variables
Brief history
Most widely used models:
-The NBD (negative bionomial distribution) model
Purchase incidence described by Poisson process (memoryless process!?)
Heterogeniety modelled by Gamma distribution.
-The Multinomial-Dirichlet model.
-Beta-bionomial model.
-Linear learning model.
Brief history
BIG BANG in choice modelling was the development of the
random utility model formulated as a multinomial logit
model by McFadden et al. (1973).
An initial appeal of the MNL model was due to it being
stochastic and yet admitting decision variables like price,
quality, promotion etc.
Utility directly incorporated in the model.
Utility specified as a stochastic function consisting of
deterministic component and a random component.
Example
”A model of audience choice of local TV news program”
Solgaard (1979, 1984)
The model provides:
a description of the associative relationship between viewers’
actual choices of news programs on the one hand and
attributes of the programs on the other. I derive this model from
a model of individual viewer choice behavior that specifies a
viewer’s probability of choosing a particular program on a given
occasion as a function of the viewer’s relative preference
toward that program. The audience model is operationalized
using a multinomial logit model.
Example
Examples of factors that may influence the choice
decision:
(1) news program attributes such as cast members (anchorman, weatherman,
sportscaster), program contents, the sequence of presentation of the various
program segments, photographic coverage, scheduling (i.e., time of day the
program is offered) and adjacent programs);
(2) the general image of the TV station;
(3) the quality of TV reception;
(4) the news of the day.
Example
Preference towards a news program:
uji = Aji + Sji, j=1,2 ... J,
where
uji = measure of preference assigned to program j by individual i,
Aji = measure of individual i’s attitude toward program j,
Sji = unobserved random component,
J = number of alternative local TV news programs.
Example
Preference of an arbitrary viewer may be written as:
Example
Attitude towards a news program is specified in the form of a multiattribute attitude model:
Example
The preference function can then be specified as:
The event that an arbitrary viewer will choose program j
Example
Assuming the random terms follow the Type I Extreme
Value Distribution results in the MNL model:
Example, results.
Economic models of choice
Assume the existence of a scalar measure of consumer
utility that can be used to generate a preference ranking of
the choice alternatives.
Consumers are assumed to make choices that are
consistent with the concept of constrained utility
maximization:
max u(q), subject to p´q ≤ y
Where x denotes a vector of quantities, p denotes the price of each item and y is the budget
constraint (i.e. the max expenditure a consumer is willing to make in a product category)
Example
An integrated model of purchase incidence, brand choice and purchase quantity,
(Chiang, 1991 and Chintagunta, 1993, also see Hanemann, 1984)
-Focuses on a single product category.
-Shopping basket separated into two groups: one containing the product
category of interest, the other all other goods purchased. All other goods are
treated as a composite good. The composite good is assumed essential and
therefore always bought.
-Consumers/households make decisions based on product attributes. A
consumer’s (subjective) evaluation of each brand in the category is
summarized in an index, ψ, - a function of product attributes, marketing mix
variables and consumer characterisitcs.
Example
Consider a store visit at time t, - made by household i. Where j refers to brand
j. The household’s problem is to maximize u subject to the budget constraint:
subject to:
Where z is the composite good.
Example
Two possible solutions:
a. Non-purchase with the budget spent entirely on the composite good, and
b. Brand choice and quantity demanded.
Solution: Non-purchase if the reservation price/shadow price, say R, is below the quality adjusted
prices for the brands in the considered category.
R < pj/ψj for all j
Purchase if the reservation price/shadow price, say R, is above the quality adjusted price for
at least one of the brands in the category.
R > pj/ψj and R < pk/ ψk k≠j
Explicit specification of the functional forms for u, ψ and the distribution for ε are of course requisite for
empirical applications.
Example
Explicit specification of the functional forms for u, ψ and the
distribution for ε are of course requisite for empirical applications.
Specifications:
u may be specified as the indirect translog utility function.
Ψ may be given the form:
exp(aij + Σxijtsβs + εijt)
Where aij represent hh i’s intrinsic preference for brand j, xijts is the value of
the sth quality attribute af brand j for hh i on store visit t, and βs the associated
parameter.
These specifications leads to flexible and useful models for each of
the choice decisions, for the brand choice it results in the MNL.
Store format choice example
Solgaard and Hansen (2003)
Objective:
(1)To model the store choice decision of supermarket
shoppers so as to be able to investigate the sensitivity of
the store choice decision to changes in shopper’s
perceptions of the choice determinants.
(2) to discuss problems involved in operationalizing
store choice models, using the framework of the
multinomial logit model, and to suggest alternative
model specifications to remedy the identified problems
Example
Determinants of store choice:
-Store image i.e. perceptions of store values:
Quality/service level
Price level
Samples
Assortment
Accessibility
-Distance
-HH descriptors
Example Store choice
Probability that consumer i will select store format j on a
particular purchase occasion:
Example Store choice
Problems with the standard logit model.
1. The coefficients of variables that enter the model are
assumed to be same for all consumers.
2. The standard logit exhibits the IIA (independence from
irrelevant alternatives) property.
A random coefficients logit model can remedy these
problems.
Example Store choice
Operationalization of the store choice model as a random
coefficients logit model:
Where Xij is a vector of observed store attribute perceptions and household
descriptors and βij is a vector of unobserved coefficients for each hh that
varies randomly over the households according to a distribution G. The
term eij is an unobserved random term independent of X and β, and
distributed IID Type I extreme value.
Example store choice
Since eij is IID extreme value, as in the standard logit model
We estimate the model applying Bayesian estimation.
Example Store choice, results
References:
Chiang, J. (1991), A Simultaneous approach to the whether, what and how much to buy
questions, Marketing Science, vol. 10, no. 4, pp 297-315
Chintagunta, P. K., (1993), Investigating purchase incidence, brand choice and purchase
quantity decisions of households, Marketing Science, vol. 12, no. 2, pp 184-208
Hanemann, W.M. (1984), Discrete/continous models of consumer demand, Econometrica, vol.
52, no. 3, pp 541-561.
Solgaard, H.S. (1979), Modelling choice of local TV news program, Unpublished Ph.D.
dissertation, Graduate School of Business and Public Administration, Cornell University,
Ithaca NY.
Solgaard, H.S. (1984), A model of audience choice of local TV news program, International
Journal of Research in Marketing, pp 141-151.
Solgaard, H.S. and T. Hansen (2003), A hierarchical Bayes model of choice between
supermarket formats, Journal of Retailing and Consumer Services, vol. 10, pp 169-180.