Knowledge that Transforms

In many categories consumers display cyclical buying: they repeatedly purchase in the category for several periods, followed by several periods of not buying. We believe that the cyclicality is a manifestation of cross-category substitution by the consumer, caused by “variety-seeking” tendencies as well as by the firm's marketing activities in all relevant categories. We propose a Markov regime-switching random coefficient logit model to represent these behaviors as stochastic switching between high and low category purchase tendencies. The main feature of the proposed model is that it divides the stream of purchase decisions of a consumer into distinct regimes with different parameter values that characterize high versus low purchase tendencies. In an empirical application of the model to purchases of yogurt-buying households, we find that as many as 38.3% households display cyclicality between high and low yogurt-purchasing tendencies. Predictions from our proposed model track observed yogurt purchases of households over time closely, and the model also fits better than two benchmark models. Alternating between high and low purchase tendencies may correspond with changing levels of consumer inventory in a substitute category. If one ignores this phenomenon, a correlation between yogurt inventory and the error term in utility arises, leading to biased estimates. Also, we show that cyclicality in buying has a key implication for a firm's price promotion strategies: a price reduction that is offered to a household during its high purchasing tendency period will result in greater increases in sales than one that is offered during its low purchasing period. This opens up a new dimension for enhancing the effectiveness of promotions—customized timing of price reductions.

The World Wide Web contains a vast corpus of consumer-generated content that holds invaluable insights for improving the product and service offerings of firms. Yet the typical method for extracting diagnostic information from online content—text mining—has limitations. As a starting point, we propose analyzing a sample of comments before initiating text mining. Using a combination of real data and simulations, we demonstrate that a sampling procedure that selects respondents whose comments contain a large amount of information is superior to the two most popular sampling methods—simple random sampling and stratified random sampling—-in gaining insights from the data. In addition, we derive a method that determines the probability of observing diagnostic information repeated a specific number of times in the population, which will enable managers to base sample size decisions on the trade-off between obtaining additional diagnostic information and the added expense of a larger sample. We provide an illustration of one of the methods using a real data set from a website containing qualitative comments about staying at a hotel and demonstrate how sampling qualitative comments can be a useful first step in text mining.

This paper studies the tendency to use negative ads. For this purpose, we focus on an interesting industry (political campaigns) and an intriguing empirical regularity (the tendency to “go negative” is higher in close races). We present a model of electoral competition in which ads inform voters either of the good traits of the candidate or of the bad traits of his opponent. We find that in equilibrium, the proportion of negative ads depends on both voters' knowledge and the candidate's budget. Furthermore, for an interesting subset of the parameter space, negativity increases in both knowledge and budget. Using data on the elections for the U.S. House of Representative in 2000, 2002, and 2004, we examine the model and its implications. Using nonstructural estimation, we find that negativity indeed increases in both voters' knowledge and the candidate's budget. Furthermore, we also find that knowledge and budget mediate the effect of closeness on negativity. Using structural estimation, we reinforce these findings. Specifically, we find that the model's parameters are within the subset of the parameter space discussed above. Thus, the evidence implies that the model is not only helpful in identifying variables that were ignored by previous studies (i.e., knowledge and budget) but also in explaining an intriguing empirical regularity.

What can be learned about marketing phenomena from descriptive, structural, and experimental empirical models? Is structure implicit in a descriptive empirical model? What is a “reduced-form model?” What is a natural experiment, and what can one infer from a study that uses experimental data? Having clear answers to these questions can improve empirical dialog. This paper defines descriptive, structural, and experimental empirical work, provides examples, discusses their similarities and differences, and comments on their strengths and weaknesses. An important theme is that the marketing question and the data available should determine the methods used, and not the other way around. Most of the examples discussed reference linear models that are widely employed in the marketing literature. Many of the points, however, extend to the development and interpretation of cutting-edge nonlinear, dynamic, or nonparametric models used in marketing.

We develop and test an incentive-compatible Conjoint Poker (CP) game. The preference data collected in the context of this game are comparable to incentive-compatible choice-based conjoint (CBC) analysis data. We develop a statistical efficiency measure and an algorithm to construct efficient CP designs. We compare incentive-compatible CP to incentive-compatible CBC in a series of three experiments (one online study and two eye-tracking studies). Our results suggest that CP induces respondents to consider more of the profile-related information presented to them compared with CBC.

Marketing expenditures in the form of pricing, product development, promotion, and channel development are made to maximize profits. A challenge in evaluating the effectiveness of these expenditures is that decisions such as whether to lower prices or run promotions are made based on managers' knowledge of how sensitive consumers are to these marketing activities. Although marketing control variables are explanatory of sales, they are often set in anticipation of a market response, which reflects strategic behavior on the part of a firm. A challenge in developing a model of strategic behavior is that the process by which marketing expenditures are made is often not directly observable. We propose tests for comparing supply-side model formulations in which input variables are strategically determined. In these models, the joint likelihood of demand (y) and supply (x) can be factored into a conditional factor of demand given supply and into a marginal factor of supply. We illustrate our approach using data from a services company that operates in multiple geographic regions.

We present a dynamic factor-analytic choice model to capture evolution of brand positions in latent attribute space. Our dynamic model allows researchers to investigate brand positioning in new categories or mature categories affected by structural change such as entry. We argue that even for mature categories not affected by structural change, the assumption of stable attributes may be untenable. We allow for evolution in attributes by modeling individual-level time-specific attributes as arising from dynamic means. The dynamic attribute means are modeled as a Bayesian dynamic linear model (DLM). The DLM is nested within a factor-analytic choice model. Our approach makes efficient use of the data by leveraging estimates from previous and future periods to estimate current period attributes. We demonstrate the robustness of our model with data that simulate a variety of dynamic scenarios, including stationary behavior. We show that misspecified attribute dynamics induce temporal heteroskedasticty and correlation between the preference weights and the error term. Applying the model to a panel data set on household purchases in the malt beverage category, we find considerable evidence for dynamics in the latent brand attributes. From a managerial perspective, we find advertising expenditures help explain variation in the dynamic attribute means.

Experimental and survey-based research suggests that consumers often rely on their intuition and cognitive shortcuts to make decisions. Intuition and cognitive shortcuts can lead to suboptimal decisions and, especially in high-stakes decisions, to legitimate welfare concerns. In this paper, we propose an extension of a Bayesian learning model that allows us to quantify the impact of salience—the fact that some pieces of information are easier to retrieve from memory than others—on physician learning. We show, using data on actual prescriptions for real patients, that physicians' belief formation is strongly influenced by salience effects. Feedback from switching patients—the ones the physician decided to switch to a clinically equivalent treatment—receives considerably more weight than feedback from other patients. In the category we study, salience effects slowed down physicians' speed of learning and the adoption of a new treatment, which raises welfare concerns. For managers, our findings suggest that firms that are able to eliminate, or at least reduce, salience effects to a greater extent than their competitors can speed up the adoption of new treatments. We explore the implications of these results and suggest alternative applications of our model that are relevant for policy makers and managers.

Sponsored search advertising is ascendant—Forrester Research reports expenditures rose 28% in 2007 to $8.1 billion and will continue to rise at a 26% compound annual growth rate [VanBoskirk, S. 2007. U.S. interactive marketing forecast, 2007 to 2012. Forrester Research (October 10)], approaching half the level of television advertising and making sponsored search one of the major advertising trends to affect the marketing landscape. Yet little empirical research exists to explore how the interaction of various agents (searchers, advertisers, and the search engine) in keyword markets affects consumer welfare and firm profits. The dynamic structural model we propose serves as a foundation to explore these outcomes. We fit this model to a proprietary data set provided by an anonymous search engine. These data include consumer search and clicking behavior, advertiser bidding behavior, and search engine information such as keyword pricing and website design. With respect to advertisers, we find evidence of dynamic bidding behavior. Advertiser value for clicks on their links averages about 26 cents. Given the typical $22 retail price of the software products advertised on the considered search engine, this implies a conversion rate (sales per click) of about 1.2%, well within common estimates of 1%–2% [Narcisse, E. 2007. Magid: Casual free to pay conversion rate too low. GameDaily.com (September 20)]. With respect to consumers, we find that frequent clickers place a greater emphasis on the position of the sponsored advertising link. We further find that about 10% of consumers do 90% of the clicks. We then conduct several policy simulations to illustrate the effects of changes in search engine policy. First, we find the search engine obtains revenue gains of 1% by sharing individual-level information with advertisers and enabling them to vary their bids by consumer segment. This also improves advertiser revenue by 6% and consumer welfare by 1.6%. Second, we find that a switch from a first- to second-price auction results in truth telling (advertiser bids rise to advertiser valuations). However, the second-price auction has little impact on search engine profits. Third, consumer search tools lead to a platform revenue increase of 2.9% and an increase of consumer welfare by 3.8%. However, these tools, by reducing advertising exposures, lower advertiser profits by 2.1%.