Department of Marketing & Entrepreneurship

In gathering information for an intended purchase decision, consumers submit search phrases to online search engines. These search phrases directly express the consumers’ needs in their own words and thus provide valuable information to marketing managers. Interpreting consumers’ search phrases renders a better understanding of consumers’ purchase intentions, which is critical for marketing success. In this paper, we develop a model to connect the latent topics embedded in consumers’ search phrases to their website visits and purchase decisions. Our model captures the dynamics and heterogeneity in the latent topics searched by consumers along the path to purchase. Additionally, we apply topic models, which have been traditionally used to analyze long text documents, to short search phrases. Using a unique dataset provided by a hospitality firm and containing more than 8,000 search phrases submitted by the consumers, our model identifies five latent topics: “loyalty”, “convenience”, “luxury”, “economy”, and “location”, underlying the searches that led consumers to the firm’s website. Compared to a model with existing semantic heuristics such as the Latent Dirichlet Allocation or a model without any usage of the textual information in consumers’ search phrases, our model provides a better evaluation of a consumer’s position on the path to purchase and achieves much better predictive accuracy based on five-fold cross validations. We also extend our discussion on the aggregator websites and segments of consumers who respond to the firm’s ads. Marketing managers can use our method to extract structured information from consumers’ search phrases and better design offerings and promotions to target the right consumers.

Recently developed quasi-experimental approaches for estimating advertising effects exploit the fact that advertising decisions are made at a more aggregate level than which we measure response. In the identification of Super Bowl advertising effects, local sales are influenced by exogenous local variation in viewership of national ads. In border strategies, local markets on opposite sides of the DMA border face different advertising levels because of the advertising incentives elsewhere in their own DMAs. We develop a model of advertising allocation and conversion that unifies these strategies based on a simple layering structure. Building on Waldfogel (2003), we use this structure to motivate a cross-market instrumental variable estimator in which demand side observables from the entire DMA are used as instruments for the advertising level, conditional on observing those same demand side variables in the local sub-DMA region. The source of variation is similar to the border strategy, but the instrumental variable approach allows all markets to enter the analysis, thereby improving statistical power and avoiding concerns about local effects. It also relaxes the unconfoundedness restrictions that are common to quasi-experimental approaches yet inconsistent with the optimizing behavior firms should engage in. We illustrate these advantages and compare estimators in and empirical application.

Consumer payments elicited on slider scales can be systematically different from those elicited through text boxes because of the endpoint assimilation effect. When people use text boxes to make payments, they evaluate monetary values relative to the starting point of the response range. In contrast, when people use slider scales, they evaluate monetary values relative to the starting point as well as the endpoint of the response range. Consequently, payments elicited on slider scales tend to be assimilated towards the endpoint of the response range. This slider scale endpoint assimilation effect varies for ascending and descending payment formats. For ascending payment formats (e.g., eBay bids), slider scales elicit higher payments than text boxes. But for descending payment formats (e.g., Priceline bids), slider scales elicit lower payments than text boxes. This research not only documents how slider scales alter consumer payments, but it also explains how the mental number line affects financial decisions.

A consumer’s decision to engage in search depends on the beliefs the consumer has about an unknown product characteristic such as price. In this paper, we elicit the distribution of price beliefs and explicitly study their role in a consumer’s decision to search. We design an incentive-aligned online study where subjects search over the price of a homogeneous good, and provide prior price beliefs and updated beliefs after each search. Based on data collected from a nationally representative panel, we find substantial heterogeneity in prior price beliefs which is at odds with the rational expectations assumption. We explore the importance of accounting for price beliefs in two ways - first, we study the impact of assuming rational expectations on estimates of search costs. For both simultaneous and sequential search models, assuming rational expectations biases the search cost estimates the direction of bias depends on the subject’s prior beliefs. Importantly, while accounting for expected price beliefs is crucial to consistently estimating search costs, assuming that the standard deviation of the subject’s beliefs coincides with the true price distribution does not substantially bias the distribution of search cost. Second, we explore the importance of price beliefs in inferring how consumers search. Assuming rational expectations, we find that subjects engage in simultaneous search which is consistent with previous research. However, the decision to engage in an additional search depends on the updated price beliefs pointing to sequential search. We discuss the managerial relevance of these results and the implications for researchers.