**Gareth James**

**
Home**
**Bio**
**Research****
Teaching
****CV****
Personal**

I am Professor and head of the Statistics group in the Marshall School of Business at the University of Southern California. I come from New Zealand and in 1994 completed Bachelor of Science and Bachelor of Commerce degrees at the University of Auckland. In 1998 I received my Ph.D. from the Department of Statistics at Stanford University. My advisor at Stanford was Trevor Hastie. I have been in the Information and Operations Management department at USC since I graduated.

I have three main areas of research.

**Functional Data Analysis.** The key tenet of this
Functional Data Analysis (FDA) is to treat the measurements of a function or
curve not as multiple data points but as a single observation of the
function as a whole. This approach allows one to more fully exploit the
structure of the data. FDA is an inherently multidisciplinary area and is
becoming increasingly important as technological changes make it more common
to observe functional data. A common FDA situation involves a functional
regression problem where one might observe a response Y and a functional
predictor X(t) which is measured over time or some other domain. The goal
would then be to build a model to predict Y based on X(t). Fitting such a
model is more challenging than for standard linear regression because the
predictor is now an infinite dimensional object.

**High Dimensional Regression.
**Traditionally statistics has involved getting information from
relatively small data sets involving perhaps on the order of a hundred
observations and ten predictors or independent variables.
However, recent technological advances in areas as diverse as web
based advertising, finance, supermarket bar code readers (linked to customer
cards) and even micro-arrays in genetics, have led to an entirely new type
of data called *High Dimensional Data*.
This data typically has anywhere from ten to a few hundred
observations but possibly up to tens of thousands of variables.
Dealing with such data poses very significant statistical and
computational challenges. Trying to find the one or two important variables among thousands with only
say 10 observations is roughly analogous to the traditional “finding a
needle in a hay stack” with the added challenge that you only get 10 guesses
before your time is up. HDD has
become one of the most important areas of research in statistics.

**Statistical Problems in Marketing.** There are many
interesting statistical problems in the marketing field. My major goal here
is to incorporate new ideas from the statistical literature to provide
solutions to practical marketing problems. For example, in one paper I used
methods from the functional data analysis literature to accurately predict
*market penetration* of 21 new products over 70 different countries.
In another paper my coauthors and I suggested a new statistical methodology
for predicting the trajectory of new technologies over time. We collected
data over time for an extensive set of technologies and showed that our
approach was overall more accurate than well known laws such as Moore's Law.