Catarina Sismeiro

A Model of Web Site Browsing Behavior Estimated on Clickstream Data

Using the clickstream data recorded in Web server log files, the authors develop and estimate a model of the browsing behavior of visitors to a Web site. Two basic aspects of browsing behavior are examined: (1) the visitors decisions to continue browsing (by submitting an additional page request) or to exit the site and (2) the length of time spent viewing each page. The authors propose a type II tobit model that captures both aspects of browsing behavior and handles the limitations of server log-file data. The authors fit the model to the individual-level browsing decisions of a random sample of 5000 visitors to the Web site of an Internet automotive company. Empirical results show that visitors propensity to continue browsing changes dynamically as a function of the depth of a given site visit and the number of repeat visits to the site. The dynamics are consistent both with “within-site lock-in” or site “stickiness” and with learning that carries over repeat visits. In particular, repeat visits lead to reduced page-view propensities but not to reduced page-view durations. The results also reveal browsing patterns that may reflect visitors time-saving strategies. Finally, the authors report that simple site metrics computed at the aggregate level diverge substantially from individual-level modeling results, which indicates the need for Web site analyses to control for cross-sectional heterogeneity.

Modeling Purchase Behavior at an E-Commerce Web Site: A Task-Completion Approach

The authors develop and estimate a model of online buying using clickstream data from a Web site that sells cars. The model predicts online buying by linking the purchase decision to what visitors do and to what they are exposed to while at the site. To overcome the challenges of predicting Internet buying, the authors decompose the purchase process into the completion of sequential nominal user tasks and account for heterogeneity across visitors at the county level. Using a sequence of binary probits, the authors model the visitors decision of whether to complete each task for the first time, given that the visitor has completed the previous tasks at least once. The results indicate that visitors browsing experiences and navigational behavior predict task completion for all decision levels. The results also indicate that the number of repeat visits per se is not diagnostic of buying propensity and that a sites offering of sophisticated decision aids does not guarantee increased conversion rates. The authors also compare the predictive performance of the task-completion approach with single-stage benchmark models in a holdout sample. The proposed approach provides superior prediction and better identifies likely buyers, especially early in the task sequence. The authors also discuss implications for Web site managers.

Using Multimarket Data to Predict Brand Performance in Markets for Which No or Poor Data Exist

The authors show how multimarket data can be used to make predictions about brand performance in markets for which no or poor data exist. To obtain these predictions, the authors propose a model for market similarity that incorporates the structure of the U.S. retailing industry and the geographic location of markets. The model makes use of the idea that if two markets have the same retailers or are located close to each other, then branded goods in these markets should have similar sales performance (other factors being held constant). In holdout samples, the proposed spatial prediction method improves greatly on naive predictors such as global-market averages, nearest neighbor predictors, or local averages. In addition, the authors show that the spatial model gives more plausible estimates of price elasticities. It does so for two reasons. First, the spatial model helps solve an omitted variables problem by allowing for unobserved factors with a cross-market structure. An example of such unobserved factors is the shelf-space allocations made at the retail-chain level. Second, the model deals with uninformative estimates of price elasticities by drawing them toward their local averages. The authors discuss other substantive issues as well as future research.

Perception Spillovers Across Competing Brands: A Disaggregate Model of How and When

Drawing on the accessibility–diagnosticity framework and previous literature on branding and order of entry, the authors hypothesize that perception spillovers can also occur across directly competing products that do not share a common brand name. The authors suggest two mechanisms (prior perception spillover and dynamic perception spillover) and one moderating variable (product/brand similarity). To test for spillovers across competing brands, the authors develop a structural Bayesian learning model and estimate it using prescription choice and marketing communication data from a panel of physicians. From their model results, the authors find evidence of prior and dynamic perception spillovers across competing brands only when brands are sufficiently similar. In contrast, they find no evidence of spillover effects across brands that are highly dissimilar. Finally, several policy experiments illustrate the strength and significance of competitive spillovers for product diffusion, and from the results, the authors derive strategic implications for order-of-entry effects and the entry of “me-too” products.

Variability in the Incidence of miRNAs and Genes in Fragile Sites and the Role of Repeats and CpG Islands in the Distribution of Genetic Material

Background Chromosomal fragile sites are heritable specific loci especially prone to breakage. Some of them are associated with human genetic disorders and several studies have demonstrated their importance in genome instability in cancer. MicroRNAs (miRNAs) are small non-coding RNAs responsible of post-transcriptional gene regulation and their involvement in several diseases such as cancer has been widely demonstrated. The altered expression of miRNAs is sometimes due to chromosomal rearrangements and epigenetic events, thus it is essential to study miRNAs in the context of their genomic locations, in order to find significant correlations between their aberrant expression and the phenotype. Principal Findings Here we use statistical models to study the incidence of human miRNA genes on fragile sites and their association with cancer-specific translocation breakpoints, repetitive elements, and CpG islands. Our results show that, on average, fragile sites are denser in miRNAs and also in protein coding genes. However, the distribution of miRNAs and protein coding genes in fragile versus non-fragile sites depends on chromosome. We find also a positive correlation between fragility and repeats, and between miRNAs and CpG islands. Conclusion Our results show that the relationship between site fragility and miRNA density is far more complex than previously thought. For example, we find that protein coding genes seem to be following similar patterns as miRNAs, if considered their overall distribution. However, once we allow for differences at the chromosome level in our statistical analysis, we find that distribution of miRNA and protein coding genes in fragile sites is very different from that of miRNA. This is a novel result that we believe may help discover new potential correlations between the localization of miRNAs and their crucial role in biological processes and in the development of diseases.

Physicians' Persistence and Its Implications for Their Response to Promotion of Prescription Drugs

Motivated by the medical literature findings that physicians are inertial, we seek to understand (1) whether physicians exhibit structural persistence in drug choice (structural persistence occurs when the drug chosen for a patient depends structurally on the drug previously prescribed by the physician to other patients) and (2) whether persistence, if present, is a physician-specific characteristic or a physician state that can change over time. We further explore the role of promotional tools on persistence and drug choice, and we investigate whether physicians who exhibit persistence respond differently to three forms of sales promotion: one-to-one meetings (detailing), out-of-office meetings, and symposium meetings. n nOur results show significant levels of physician persistence in drug choice. We find that persistence is mostly a cross-sectional physician feature. Nonpersistent physicians appear to be responsive to detailing and symposium meetings, whereas persistent physicians seem to be responsive only to symposium meetings. Out-of-office meetings, such as golf or lunch, have no effect on physicians drug choice. We also find that (1) older physicians and those who work in smaller practices are more likely to be persistent and (2) physicians who are more willing to receive sales force representatives have a lower likelihood of being persistent. Finally, we discuss implications for public policy from our rich set of results.

Using visual and text features for direct marketing on multimedia messaging services domain

Traditionally, direct marketing companies have relied on pre-testing to select the best offers to send to their audience. Companies systematically dispatch the offers under consideration to a limited sample of potential buyers, rank them with respect to their performance and, based on this ranking, decide which offers to send to the wider population. Though this pre-testing process is simple and widely used, recently the industry has been under increased pressure to further optimize learning, in particular when facing severe time and learning space constraints. The main contribution of the present work is to demonstrate that direct marketing firms can exploit the information on visual content to optimize the learning phase. This paper proposes a two-phase learning strategy based on a cascade of regression methods that takes advantage of the visual and text features to improve and accelerate the learning process. Experiments in the domain of a commercial Multimedia Messaging Service (MMS) show the effectiveness of the proposed methods and a significant improvement over traditional learning techniques. The proposed approach can be used in any multimedia direct marketing domain in which offers comprise both a visual and text component.

Exploiting visual and text features for direct marketing learning in time and space constrained domains

Traditionally, direct marketing companies have relied on pre-testing to select the best offers to send to their audiences. Companies systematically dispatch the offers under consideration to a limited sample of potential buyers, rank them with respect to their performance and, based on this ranking, decide which offers to send to the wider population. Though this pre-testing process is simple and widely used, recently the direct marketing industry has been under increased pressure to further optimize learning, in particular when facing severe time and space constraints. Taking into account the multimedia nature of offers, which typically comprise both a visual and text component, we propose a two-phase learning strategy based on a cascade of regression methods. This proposed approach takes advantage of visual and text features to improve and accelerate the learning process. Experiments in the domain of a commercial multimedia messaging service show the effectiveness of the proposed methods that improve on classical learning techniques. The main contribution of the present work is to demonstrate that direct marketing firms can exploit the information on visual content to optimize the learning phase. The proposed methods can be used in any multimedia direct marketing domains in which offers are composed by image and text.

Automatic content targeting on mobile phones

The mobile phone industry has reached a saturation point. With low growth rates and fewer new customers available to acquire, competition among mobile operators is now focused on attracting competitors customers. This leads to a significant downward price pressure, the inability by mobile phone providers in deriving reasonable returns from basic telephony services, and an increasing reliance on value added services (VAS) for revenue growth. There are today thousands of such services available for companies to sell to their customers daily. These services include, for example, the provision of sports information, ring-tones, personalized news, weather forecast, and financial trends. Because of the many possible offers, and of the limited contact opportunities (operators tend to cap the number of commercial messages sent to their users and phones have limited-size screens), data mining can play an important role in optimizing message targeting. In this paper we describe our experience in developing a successful automatic system to target users with the most relevant offers. We describe the proposed data mining methods and report on their performance. In addition, we discuss several experiments we implemented on live data. These experiments have been useful to tailor our approach to the specific characteristics of the market under study. We believe this is a very interesting domain for data miners though it is still fairly unexplored. This is despite the availability of very large and detailed logs of customer activity.

Can Branded Drugs Benefit from Generic Entry? Switching to Non-Bioequivalent Molecules and the Role of Physician Response to Detailing and Prices

Patent expiration represents a turning point for the brand losing patent protection as bioequivalent generic versions of the drug quickly enter the market at reduced prices. In this paper, we study how physician characteristics and their prescribing decisions impact the competition among molecules of a therapeutic class, once generic versions of one of these molecules enter the market. Our results suggest that to understand the diffusion of generics in the category marketers should (1) determine the size of physician segments sensitive to marketing activity and prices, and (2) assess the marketing activity of all pharmaceutical firms, whether bioequivalent or not. We further discuss the managerial implications of our results.