Conference Sessions - Jesse H. Jones Graduate School of ...

FB02 MARKETING SCIENCE CONFERENCE – 2011
2 - Optimal Dynamic Pricing Strategies: Consumer Cross-category
Incidence/Purchase Quantity Decisions
Sri Devi Duvvuri, Assistant Professor, University at Buffalo,
Marketing, School of Management, 215F Jacobs Management Center,
Buffalo, NY, 14260, United States of America, sduvvuri@buffalo.edu,
Praveen Kopalle
The objective of this research is to determine optimal pricing policies from a retailer’s
perspective by taking into consideration the dynamics of consumers’ purchase
behavior across categories. While research on category management focuses on
developing pricing strategies within a category, our approach involves both withinand
cross-category interactions. In the consumer demand model, we investigate the
effects of including the quantity purchased on estimation and specification of
multivariate choice models. Specifically, we estimate a Hierarchical Bayes
econometric specification of a multivariate tobit model. A multivariate tobit model
simultaneously accounts for purchase incidence and quantity decisions of consumers
across multiple categories. The model will be estimated using Markov Chain Monte
Carlo (MCMC) methods such as Gibbs Sampling and Metropolis-Hastings Algorithm.
Scanner data for different categories are used for estimation. Using the estimated
parameters, we develop profit-maximizing dynamic pricing policies over time for the
various brands across the two categories.
3 - Market Delineation Strategies in Consumer Goods Market
Sebastian Gabel, Doctoral Student, RWTH Aachen, Virchowstr. 2,
Aachen, 52066, Germany, sebastian.gabel@rwth-aachen.de,
Raimund Bau
The term market delineation is closely associated with anti-trust research and
adjacent literature. Yet, knowledge of the relevant products in a market is of great
importance for marketing practitioners and researchers as well. In clearly
differentiated markets with small number of alternatives, marketing research and
practice can easily incorporate all alternatives into models and competitor strategies.
Problem arise when potential markets carry several thousand SKUs that (a) cannot be
consolidated on a small set of properties and (b) exhibit parallel product usage as well
as full substitution patterns. In these markets practitioners often use heuristics, such
as product market shares, to reduce their field of attention to a feasible set of
products. However, heuristics may lead to misguided management decisions. For
researchers using discrete choice models market delineation is of great importance if
the model estimation becomes computationally intensive due to a large number of
alternatives. Popular sampling strategies might not fully reflect consumer substitution
patterns of a particular firm’s products. Using a market simulation we evaluate how
different delineation strategies make it possible to reach both the practitioner’s and
researcher’s goals. We evaluate 1) Management heuristics based on aggregate data, 2)
Advanced descriptive individual-level data, 3) Causal time series analysis and 4)
Random sampling. Based on choice sets derived in each approach we analyze
potential bias in parameter estimates and elasticities in an aggregate logit model. A
comparison demonstrates which strategy yields best results under given
circumstances. Finally, the benefits of using the results of this simulation study will be
illustrated in an analysis of a German food market.
4 - The Influence of Willingness-to-pay on Consumer’s Cross Category
Purchase Behavior
Kamer Toker-Yildiz, PhD Student, University at Buffalo, Marketing,
School of Management, 215 Jacobs Management Center, Buffalo, NY,
14260, United States of America, ktyildiz@buffalo.edu,
Sri Devi Duvvuri, Minakshi Trivedi
Consumer’s cross-category purchase behavior has been investigated by marketing
researchers due to its important managerial and theoretical implications. This stream
of studies have shown that incorporating cross category effects allows a better
understanding of consumer’s purchase behavior. For a typical shopping trip,
consumers are likely to make multiple purchases and their purchase decisions depend
on various observed and unobserved factors. In addition to studying the effects of
observable variables (e.g., marketing mix), recent research has also documented
modeling efforts that also estimate the effects of unobservable factors. In this study,
we propose a cross-category model that allows us to measure how consumer’s
willingness-to-pay influences his actual purchase decision in a multi-category decision
making framework. It is important to note that willingness-to-pay is a
multidimensional variable and unobservable to the researcher. We build a
multivariate tobit model to accommodate both cross-category purchase behavior and
the willingness-to-pay construct. We use hierarchical Bayesian inference to estimate a
heterogeneous specification of our model. We use scanner panel data for organic and
conventional product types across multiple categories. Interestingly, we use survey
data from the households in the panel to further investigate and validate our
findings.
44
■ FB02
Legends Ballroom II
UGC-II (Quest for Comprehension and Integration)
Cluster: Internet and Interactive Marketing
Invited Session
Chair: Manish Tripathi, Emory University, Atlanta, GA,
United States of America, manish_tripathi@bus.emory.edu
1 - Listening in on Online Conversations: Measuring Consumer
Sentiment with Social Media
David Schweidel, University of Wisconsin, Wisconsin School of
Business, Madison, WI, United States of America,
dschweidel@bus.wisc.edu, Wendy W. Moe
Businesses are increasingly turning to social media as a listening tool in an effort to
gain insights into how consumers perceive their brand. Some recent research has
examined the predictive ability of both the volume of social media contributions and
the sentiment expressed therein. Much of this work, however, focuses either on
comments contributed to a single online venue or on a simple aggregation of
comments across venues and domains (e.g., total number of contributions across
venues or average sentiment expressed across all venues). In this research, we probe
the differences in sentiment that manifest across multiple types of venues, including
blogs, forums and social networks. We also investigate differences in sentiment that
may exist between contributors with and without direct experience. To this end, we
model the posted online sentiment pertaining to a single brand across all social media
venues. Our analysis reveals notable differences across venues and time in the
sentiment that contributors express, suggesting that measures of social media will
depend on the source from which they are gathered. By controlling for the effects of
venue, domain, experience and other factors, we proposed an adjusted online
sentiment metric that represents underlying consumer sentiment better than metrics
based on contributions to a single online venue or a simple aggregation of all online
contributions.
2 - Social Tag Maps: A New Approach for Understanding Brand
Association Networks
Hyoryung Nam, University of Maryland, College Park, MD,
United States of America, Hnam@rhsmith.umd.edu
We present a novel technique for inferring consumers’ brand associative networks
using data from social tags. Social tags are a manifestation of a consumer’s mental
representation of a particular product, brand or firm; and hence provide a rich source
of information for assessing how consumers think and conceive different products
and brands individually as well as related to each other. Given the massive rise in
user generated content in the recent years, many websites provide such information
in a readily accessible manner. Marketers have much to gain from processing this
information to understand the revealed rather than stated preferences of consumers
for their products and brands. We demonstrate that the social tag maps have
significant benefits for understanding brand associations as compared to conventional
techniques such as metaphor elicitation techniques or brand concept maps. In
particular, social tags provide a unique ability to obtain information about dynamic
evolution of brand associations and competitive associative maps.
3 - The Quest for Content: The Role of User Generated Links in
Online Content
Shachar Reichman, Tel-Aviv University, Israel, sr@post.tau.ac.il,
Jacob Goldenberg, Gal Oestreicher
Online content is often presented as a product network, where nodes are product
pages linked by hyperlinks (e.g., books on Amazon.com). These links are often
generated by collaborative filtering algorithms and are based on aggregated data.
Recently, websites have begun to offer social networks and user generated links
alongside the product network, creating a dual network structure. We focus on the
role of this dual network structure and specifically of the user generated links in
facilitating content exploration. We first analyze the YouTube.com dual network and
show that user pages have unique structural properties (e.g. high betweenness
centrality) and act as better content brokers in the dual network. Next, we create a
synthetic dual network and show that random rewiring of the product network
cannot replicate the brokering effect of the user generated links of the real world dual
network. We conduct five internet experiments to study the effect of different
structures on the efficiency of the exploration process (time to desirable outcome),
and effectiveness (average product rating and overall satisfaction). Our first two
experiments show that the dual network structure leads to faster access to “good”
content and to overall higher satisfaction. We subsequently extend the experiments
to include dynamic structures, in which the underlying site structure changes as a
function of time and in response to participants’ satisfaction. Our findings suggest that
using a dynamic mechanism leads to an increase in consumers’ ratings of the
products as well as to higher overall satisfaction.