EDA using Direct Manipulation Graphics     

                       Di Cook 
     Department of Statistics, Iowa State University
     

Question: What is the primary goal of exploratory data
analysis?

Answer: Insight. The clear (and often sudden) understanding of a
complex situation, the power or act of seeing into a situation;
Insight implies detecting and uncovering underlying structure in the
data.

"The primary goal of EDA is to maximize the analyst's insight into a
data set and into the underlying structure of a data set, while
providing all of the specific items that an analyst would want to
extract from a data set, such as:

   1. a good-fitting, parsimonious model
   2. a list of outliers
   3. a sense of robustness of conclusions
   4. estimates for parameters
   5. uncertainties for those estimates
   6. a ranked list of important factors
   7. conclusions as to whether individual factors are 
         statistically significant
   8. optimal settings" http://www.itl.nist.gov/div898/handbook/

With the evolution of fast, graphically enabled desktop computers,
exploratory data analysis has has evolved into a highly interactive,
real-time, dynamic and visual process.  To perceive multi-dimensional
structure, the user displays multiple views of the data and links
graphics objects in one plot with the other plots. Thus with basic
building blocks such as histograms, scatterplots, barcharts, mosaic
and double decker plots the statistician can pull out a lot of
information about high-dimensional relationships in data.

This talk will demonstrate the highly interactive data analysis
process as applied to a microarray data set (groan?) collected at
ISU. It's interesting data! There are treatments, repeated measures,
and model diagnostics. I began the analysis using graphics. A student
began the analysis using statistical modeling. The lists of
interesting genes that we produced had almost zero overlap! Why is
this so? This talk will discuss how this is possible, what we learn
from each approach and what we miss. And along the way we will see how
to use multivariate graphics, some new complex linking between plots,
and some surprising data findings.

This is joint work with Heike Hofmann, Eun-Kyung Lee, Hao Yang, Basil
Nikolau and Eve Wurtele.