Applied

Multivariate
Analysis
Introduction

1
Nature of Multivariate
Analysis
Typicallyexploratory, not confirmatory
Often focused on simplification
Often focused on revealing structure in
dimensions that our eyes and
imaginations dont fully support.

2
Adequate Preparation?
Basic course in statistical science
STA 671
SAS exposure
Linear algebra (?)

3
Begin Reviewing and Reading
Basic data steps in SAS
Chapter 1 in AMD
Chapter 2 in AMD
Well begin with Chapter 4

4
Potential Topics Covered
PrincipalComponents Analysis (PCA)
Factor Analysis (FA)
Discriminant Analysis (DA)
Multidimensional Scaling (MDS)
Cluster Analysis (CA)
Canonical Correlations Analysis (CCA)
Multivariate Analysis of Variance
(MANOVA)
5
Why Multivariate?
Typically more than one measurement
is taken on a given experimental unit
Need to consider all the
measurements together so that one
can understand how they are related
Need to consider all the
measurements together so that one
can extract essential structure

6
In Chromatography
Ty pical Chromatogram f or PCB
14

12

10

Percent of Substance
8

0
0 10 20 30 40 50 60 70 80 90 100
Isomer/Congener

one observation

7
In Neuroimaging

MRI Time Series f or a Single Voxel

4

Magnitude
0

-1

-2

-3
0 20 40 60 80 100 120 140 160 180
Time

one observation

8
In Social Science Research

Education level
Your opinion on welfare
Your opinion on social security
Your opinion on .

one (joint) observation

9
Distinguishing Midges

Suppose we are interested in measuring

the wing length and the antenna length.
10
Distinguishing Midges
What can you do with both variables
that you cant do with just one of
them?
2.1
Coordinate Axis Projections 2.1 Relabeling the Coordinates
2.05 2.05
Species Af Species Af
2 2

1.95 1.95

1.9 1.9
Wing Length

Wing Length
1.85 1.85

1.8 1.8

1.75 1.75

1.7 1.7

1.65 Species Apf 1.65 Species Apf

1.6 1.6
1.1 1.2 1.3 1.4 1.5 1.6 1.1 1.2 1.3 1.4 1.5 1.6
Antenna Length Antenna Length

11
Measuring Heads
Are these data truly two-dimensional?

Scatterplot of LTN vs. LTG 135

Scatterplot of LTN vs. LTG
135

130 130

125 125

LTN
LTN

120
120

115
115

110
110 120 125 130 135 140 145 150 155
120 125 130 135 140 145 150 155
LTG
LTG

Not the usual regression line .

12
Our Approach in STA 677
Emphasize De-emphasize
Intuition Theoretical
SAS basis
Formal proofs
Geometry
Interpretations
Data Analysis

13
Getting on the Computers
Here

14
Personal SAS License
Lorinda Wang
[email protected]
SStars Lab
213d M I King 0039
Phone 859 257-2204
Fax 859 323-1266

15
Organizational Details
Please get the textbook (required)
Look at Readme.txt on the text CD
Notes posted on the class website
Take a look at the syllabus

16
Basic Vocabulary
Variance More than one kind of
variability will emerge.
Covariance
Correlation

17
Additional Vocabulary
Eigenvalues
Eigenvectors
Projections
Matrix Notation

18
Discovering Linear
Combinations
Log on to the computer in front of you and
access our course web site.

Find the data set helmet.xls and open it.

Compute (.707*LTN)+(.707*LTG) (use Excel)

What did you just do geometrically?

19
Discovering Linear
CombinationsHelmet Data
15

10
Equal Wts
5 On LTG, LTN

0
LTN

-5

-10

-15
LTG is WTD > LTN
-20
-20 -15 -10 -5 0 5 10 15
LTG
20
Discovery Exercise Continued

Find the variance of LTN, LTG (use Excel).

Find the variance of (.707*LTN)+(.707*LTG) ---

equal weights.

Find the variance of (.50*LTN)+(.85*LTG) ---

unequal weights with LTG weighted more.

21
Discovery Exercise
What did you find and does it make sense?
Var(LTN)= 15.37
Var(LTG)= 31.84
Var(707)= 38.11
Var(5085)= 39.19

Thisis no accident. And this is what

Principal Components is all about.

22
Encounter With SAS
Save the helmet file to your hard disk.

Exit Excel and start up SAS.

Watch the demonstration on how to

bring the Excel file into SAS.

Repeat this yourself.

23
Encounter With SAS
Itis easy to transfer the AMD .txt data
to Excel files. If you dont know how
and want to know, just ask.

So you can always bring your data in

as Excel files if you want.

That is what Ill do in front of the class.

24
Coming Up
Principal Components
Analysis

25