Exploratory Data Analysis – Understanding Your Data through the Use of Graphics

:: Course Summary

:: Learning Objectives

• Understand the fundamentals of static graphing tools for both univariate and multivariate data

• Use graphical techniques to explore the nature of data distributions

• Understand scaling/aspect ratio

• Appreciate the problems of graphing large data sets

• Recognize the usefulness of interactive graphics and how to incorporate it into your data analysis

• Understand linked brushing and how to use this technique

• Use the tour technique to look for high-dimensional structure

• Take a dataset and use course procedures to search for interesting features

• Acquire the practical skills needed to apply course techniques to your data

• Learn to judge the quality of a graph

:: Target Audience

:: Prerequisite

:: Topics Covered

Static graphics

• Histograms – their use and misuse
• Quantile plotting – exploring the nature of the data distribution
• Box plots – looking at the data for many variables
• Jittering - when you have many data points of the same value
• Hexagon binning – a tool to use when your data set is large
• SPLOM – looking at more than one scatterplot
• Parallel coordinate plots – plotting many variables in one dimension
• Mosaic plots – what to do with multivariate categorical data

Dynamic/interactive graphics

• Conditioning your plots – revealing structure in subsets of your data
• Touring - taking a spin to visualize the nature of high dimensional data
• Selection & identification – presenting information about a point or clusters of points
• Linking & brushing – multiple views of the same data shown simultaneously
• Transformations and scaling your plots – should the axis be changed?
• Missing values - what is there and not there?

Some of the plotting software demonstrated in the course (all open source):

• R-lattice
• Ggobi
• Mondrian
• ggplot
• iplots

:: Course Content

The course is divided into two sections: the first (morning) session explores the use of static graphics, including the traditional histogram, cumulative distribution plots, Q-normal plots. Jittering, box plots, hexagon binning, and the use of mosaic plots for displaying categorical data will also be discussed. The graphing of both univariate and multivariate data sets is included during this session. Discussion will focus on how to apply these tools and their ‘proper’ treatment within a problem-solving environment. This session will also review the potential problems of graphing large data sets.

The second (afternoon) session is dedicated to dynamic and interactive graphics. The session introduces interactive and dynamic graphics, and includes topics such as sub-setting and conditioning, linked windows and graphical tours using these tools within the context of the familiar scatterplot, bar chart and parallel coordinate plots. Direct manipulation of the plots includes scaling, moving points, linked brushing and identification using categorical variables. A special section of the session is devoted to the important problem of dealing with missing data. This session introduces GGobi, an open source, dynamic-interactive graphics program.

Two class demonstrations will review the techniques discussed and show the procedures for generating the different types of graphical applications. These demonstrations are complemented with two tutorial sessions in which the participants work along with the instructor in order to analyze a data set using several of the graphing/plotting procedures presented. Participants are encouraged to use their own data.