Research Highlights

To Explain or To Predict?

My research examines the fundamental and practical differences between using statistical and other empirical methods for prediction compared to causal explanation and to description. Although the discussion of explanation vs. prediction has been actively pursued in the philosophy of science, the statistics literature has not considered it in a holistic way. Yet, statistical modeling can be and is used for each of these goals.

Statistical Strategy

I have been tackling some "big picture" questions related to using statistical methods in practice.

My major focus is on assessing the differences between explanatory, predictive and descriptive modeling and statistical modeling in terms of the statistical modeling process (from data collection and goal definition to model use). My paper To Explain or To Predict? discusses the distinction from a statistical point of view. The paper Predictive Analytics in Information Systems Research examines the value of predictive modeling to theory building, testing, and validation, illustrated in information systems research which is monopolized by explanatory modeling.

Quality Control

Runs and Scans in Industrial statistics

During my Ph.D. I developed a method for computing exact probabilities for random variables that arise when runs or scans are used. A run is a sequence of consecutive successes in a series of Bernoulli trials. A scan is a “window” of consecutive Bernoulli trials that includes at least a given number of successes. Runs and scans are applied in various fields. Although they are easy to understand and use, the random variables that arise tend to have characteristics (e.g. probability functions, moments) that are complicated for computation.

Count Data Model

Shmueli et al. (2005) revived a useful discrete distribution called the COM-Poisson (the Conway–Maxwell–Poisson) and introduced its statistical and probabilistic properties. This distribution is a two-parameter extension of the Poisson distribution that generalizes some well-known discrete distributions (Poisson, Bernoulli and geometric).

Online Auctions

Empirical research of online auctions such as eBay has been dominated by researchers from economics and information systems. Together with colleagues from Information Systems and statistics, I have been working on developing statistical methods for visualizing, collecting, modeling, predicting and analyzing such data. Bid data (and other types of eCommerce data) have non-standard structures and therefore require careful and specialized methods.

previous pauseresume next

At a Glance

Galit Shmueli is Tsing Hua Distinguished Professor at the Institute of Service Science, and Director of the Center for Service Innovation & Analytics at the College of Technology Management, National Tsing Hua University, Taiwan.