SYMPOSIUM BACKGROUND

Research in electronic commerce (eCommerce) and related fields produces an increasing amount of data-related questions and problems. Take for instance online auction research: Online auctions have become a popular focal point of eCommerce practice and research. The popularity of these electronic markets provides researchers with a new-found ability to observe real economic agents interacting in technology enabled real-world settings. The availability of large amounts of bidding data enables and facilitates, for the first time, the empirical investigation of classical auction theory. eBay, the largest and most well-known online auction house, hosts several millions of auctions every year. Researchers can access the data from these auctions by designing and implementing latest web-data collection technology (i.e. web spiders), and can compile vast databases of bidding records in a matter of only hours or minutes. Yet, significant challenges remain in translating an enhanced understanding of such markets into improved designs.
 
Information technology has added challenging new dynamics that call for enhanced empirical investigational techniques. The introduction automated commerce agents (i.e. bidding agents for online auctions) have changed the speed, volume, and intensity of trades warranting new ways of assessing these matching mechanisms. Additionally, the complexity of IT-enabled trading mechanisms that support multi-unit and multi-attribute decisions reduce the transparency of the information and data generated. Finally, the availability of vast databases originating from eCommerce brings up many new methodological challenges. These challenges range from computational difficulties in handling large amounts of data to the challenges of dealing with new methodological and conceptual issues that are related to data structure, origin, and analysis and are at the intersection of Statistics, Economics and Information Systems. For instance, much of the current research in the eCommerce literature focuses on applying standard statistical methods and models. However, since many of these methods are developed for data of much smaller dimension, they do not scale up. Examples range from the very basic linear regression model to computationally-intensive Bayesian approaches. Oftentimes, methods are of iterative nature, requiring many runs through the entire set of data, which makes their application to large databases computationally infeasible.

Aside from computational issues, the direct application of standard statistical methods can lead to substantial information loss. Classical statistical methods, such as regression models, often do not capture a significant part of the information contained in data from an eCommerce environment. Take again online auction research as an example: Traditional research focuses on modeling bidding data only in a static way, in the sense that it looks only at a snapshot of the auction taken at the auction end and disregards data on its entire duration. Such a static relationship does not capture important dynamic aspects of the auction. In comparison, a dynamic approach looks at an auction throughout its entire duration and uses data on the auction progression to learn about its dynamics. Examples of dynamics that are not captured by a static model are a bidder’s reaction to another bidder’s action (which may be instantaneous or time-delayed). A higher price achieved for a product may be the result of an auction with a faster rate of incoming bids and incoming bid-increments; that is, faster bidding dynamics. New methodological approaches are necessary in order to capture bidding dynamics. In a recent series of working papers, Jank and Shmueli have shown that new statistical approaches, like Functional Data Analysis, can be useful in capturing dynamic information.

Information loss in eCommerce data, and in particular, in online auctions can also occur on a wider, more global scale. A bidder’s reaction to another bidder’s action may not only be the result of events within the auction but it may also be affected by events happening outside the auction, say in a different auction or in a different marketplace. Similar issues arise in other areas of empirical eCommerce research, such as the impact of online product review sites on consumer purchasing behavior or the impact of search engines on a company’s sales. Traditional statistical models do not take into account the dynamics and concurrency of events happening in a marketplace as large as the Internet. In a general sense, the static viewpoint impedes the IS researcher from being able to take a prescriptive “design science” perspective. In the latter the researcher takes an active role in guiding an economic mechanism towards a desirable objective (such as a higher allocative efficiency) as opposed to taking an ex post observational stance.

Besides the need for new computational and methodological advances, new conceptual thought is also necessary. Empirical research typically relies on the assumption that the collected data are a sample representative of the population of interest. Only if this is satisfied will analytical methods that are based on this assumption be valid in the sense of generalizability. Empirical research associated with eCommerce often relies on samples drawn from the Internet. In the example of online auctions, researchers attempt to explain different patterns of bidder behavior based on a sample of online auction data. While a sample of online auction data can be gathered with relative ease and speed, it is much harder to establish whether this sample is representative of the population of interest, or what type of biases are present. Furthermore, many issues related to sampling and non-sampling errors are of concern in web-sampling but have been completely overlooked in the literature.

The process of hypothesis formation, experimenting and testing makes the IS-Statistics relationship obvious: After obtaining a set of data representative of participants in online auctions, new light can be shed on many aspects of bidder behavior. Consider for instance the long-standing viewpoint in auction theory that bidders are homogenous (rational profit maximizing economic agents) in their behavior, each calculating an optimal bidding function given their valuation of an item. Given large and representative datasets of actual bidding behavior across a wide range of item categories, in combination with appropriate statistical inference tools, we can empirically test this long standing assumption and learn about how real bidders strategize. This knowledge in turn can be used to customize the auction’s design to better align the incentives of the various parties. In fact, given the right set of statistical tools, one can learn about how bidders’ strategies change over the course of the auction, where change can be motivated by new information made available within the auction or new information arriving from outside of the auction. Methods to model a bidder’s time-dependent behavior, and associated change, typically require a set of observations large enough. However, in many auctions bidders place only one bid, leaving only very sparse information. New statistical thought and methodology is necessary and can provide answers or at least guidelines to many of these questions and problems.

Based on these illustrative examples and on our own collaborative experience, we believe that closer interaction among researchers in this important interdisciplinary area will provide significant leverage in advancing this field of research and accelerate its practical application. Furthermore, such collaboration will prove fruitful to research development in Statistics, Information Systems and related fields. In this spirit, we propose to organize the first interdisciplinary academic symposium on Statistical Challenges and Opportunities in eCommerce Research, to take place in May 2005. Our intention is to bring together the leading researchers and collaborators from Statistics with the leading researchers in Information Systems that are active in the area of eCommerce. The purpose of this symposium is to help us better understand how these various lines of work connect to one another and how, together, they can contribute to the modernization and enhancement of empirical research methods for electronic commerce and our digital society at large.