PHD Syllabus Multivariate Analysis

Syllabus for Multivariate Analysis
(26:630:670)
Rutgers Business School Newark• PHD in Management Program
Professor J. Douglas Carroll
Spring 2004
• PHONE: (973) 353-5814 (Office)

• E-MAIL: [email protected]
• OFFICE HOURS: 4-5PM Tuesdays or by appointment. E-mail communication is strongly encouraged.
• OFFICE: 125 Management Education Center
• Class meets: Tuesdays, 6-9PM in MEC117
• Text: Analyzing Multivariate Data by James M. Lattin, J. Douglas Carroll, and Paul E. Green (Belmont,
CA: Duxbury Press).
• Optional Supplementary Text: Carroll, J. D., & Green, P. E. (1997). Mathematical Tools for Applied
Multivariate Analysis. San Diego, CA: Academic Press (with contributions by A. D. Chaturvedi).
Teaching Assistant: Ulas Akkucuk

Office Hours: By appointment
E-mail: [email protected]
Class 1.
Chapter 1: Introduction. Overview of book and course.
Chapter 2: Review of basics of vector and matrix algebra and applicability to multivariate
analysis (MVA).
Class 2.
Chapter 2 and supplementary material.
Review of Singular Value Decomposition (SVD) of general matrices, Eigenstructure

(eigenvector and eigenvalue) decomposition of square matrices, and application of
eigenstructure decomposition of square symmetric “product moment” matrices to
calculation of SVD of a general matrix. Statistical applications of eigenstructure and
SVD analysis of various matrices; use of SVD to determine rank, compute
determinant, and define inverse (or various generalized inverses, including the right or
left pseudo-inverse) of various special matrices.
Class 3.
Chapter 3: Regression Analysis.
Overview of multiple linear regression analysis and its applications. Both ordinary
regression analysis and partial regression analysis, in which the regression of one set
of variables on another after the effect of a third set of variables has been “partialed
out,” or subtracted, will be dealt with, as well as methods of testing significance of
various regression models, or of more complex models as compared to less complex
ones.
Class 4.
Chapter 4: Principal Components Analysis.
Overview of principal components analysis (PCA). Relationship to multivariate

normal distribution and to eigenstructure of covariance or correlation matrix and to
SVD of original (mean centered or standardized) matrix of observations by variables.
The latter discussion will introduce the notion of component scores (closely related to
that of factor scores in factor analysis).
Class 5.
Chapter 5: Exploratory Factor Analysis.
Will primarily focus on principal axis form of exploratory factor analysis (FA) and its
relationship to PCA. Will include a discussion of computation of factor scores as well
as factor loadings, as well as various approaches to estimation of “communalities”.
Class 6 & 7.
Chapter 7: Multidimensional Scaling.
Metric and nonmetric models and methods of “two-way” multidimensional scaling

(MDS). Three-way MDS; specifically the INDSCAL model and the SINDSCAL
method of three-way (or individual differences) MDS, and its implementation.
Methods of multidimensional analysis of preferential choice (or other “dominance”)
data. MDS as an alternative to PCA or exploratory FA, as well as a method of
analysis of directly judged or otherwise measured proximities (similarities or
dissimilarities, measures of association, concurrence, overlap, friendship or other
indicators of affinity, or other measures of “closeness”) of a set of stimuli or other
objects (e.g., products or brands in marketing studies).
Please Note: Take home midterm exam will be distributed in Class 7. Will be due by time of
meeting of Class 9. Alternatively, some students may substitute a project in which a
substantive data set relevant to that student’s current research or planned dissertation
research is analyzed via methods discussed in this course. Substitution of such a
project will require permission of Prof. Carroll, and, if permitted, the project will be
due not later than the meeting of Class 11.
Class 8.
Chapter 8: Cluster Analysis.
Emphasis will be placed on hierarchical clustering, specifically single, complete and

average linkage, Ward's method, and Kth nearest neighbor clustering, based on direct
or derived measures of similarity or dissimilarity, and on K-means clustering for
partitioning based on a standard objects by variables multivariate data matrix.
Class 9.
Chapter 9: Canonical Correlation.
Canonical correlation analysis (CCA) will be presented as a very general technique

for interrelating two (or more, in some generalizations) matrices of variables defined
on the same objects by finding linear combinations of each that have maximum
correlation. The method of computing the first set of canonical variates and their
canonical correlation is discussed and described, and then methods of computing a set
of ordered canonical variates, the Kth set of variates constrained to be orthogonal to the
first K-1 (in a certain sense that will be defined) will be covered.
Classes 10 and 11.

Chapter 11: Analysis of Variance, and generalizations.
A review of standard analysis of variance (ANOVA), plus an introduction of

ANCOVA (analysis of covariance), MANOVA (Multivariate ANOVA) and
MANCOVA (Multivariate ANCOVA) will all be covered. It will be shown that
ANOVA can be viewed as a special case of multiple linear regression analysis, with
“dummy” independent variables, while ANCOVA is a special case of partial
regression analysis. Both MANOVA and MANCOVA will be shown to be
formulatable as special cases of canonical correlation analysis.
Class 12.
Chapter 12: Multiple Discriminant Analysis.
Multiple Discriminant analysis (MDA) is used when it is desired to find linear

combinations of a set of given variables that best discriminate among two or more
different groups. MDA will be formulated, and shown to be a special case of
canonical correlation analysis, with the (usually continuous) given variables defining
one matrix and a set of dummy variables encoding the groups comprising the second
one.
Class 13.
Chapters 6: Confirmatory Factor Analysis and
Chapter 10: Structural Equations Models with Latent Variables.
OR,
Chapter 13: Logit Choice Models.
As time permits, this class will deal with a synopsis of various approaches to
confirmatory factor analysis (CFA) and Linear Structural Equation Modeling
(LISREL and related approaches) as compared to and contrasted with exploratory
factor analysis (EFA)
Alternatively, depending on preferences of class members, Class 13 will deal with

Chapter 13 on Logit Choice Models.
Final exam: Will be given in last class.
Grading: Grades will be based on midterm exam or project (40%), final exam (40%), as well as
class participation and homework (20%).

PHD Syllabus Multivariate Analysis

Uploaded by

Copyright:

Available Formats

PHD Syllabus Multivariate Analysis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PHD Syllabus Multivariate Analysis

Uploaded by

Copyright:

Available Formats

Syllabus for Multivariate Analysis

• PHONE: (973) 353-5814 (Office)

Teaching Assistant: Ulas Akkucuk

Review of Singular Value Decomposition (SVD) of general matrices, Eigenstructure

Overview of principal components analysis (PCA). Relationship to multivariate

Metric and nonmetric models and methods of “two-way” multidimensional scaling

Emphasis will be placed on hierarchical clustering, specifically single, complete and

Canonical correlation analysis (CCA) will be presented as a very general technique

Classes 10 and 11.

A review of standard analysis of variance (ANOVA), plus an introduction of

Multiple Discriminant analysis (MDA) is used when it is desired to find linear

Alternatively, depending on preferences of class members, Class 13 will deal with

Final exam: Will be given in last class.

You might also like