two sets

Report
Canonical Analysis
•
•
•
•
•
•
•
Introduction
Assumptions
Model representation
An output example
Conditions
Procedural steps
Mechanical steps
- with the use of attached file
1
•
The computer output can be obtained by
opening a syntax window and typing SPSS
commands. Click on File, click on New and
then click on SPSS Syntax, and type the
statements. The following box is the
statements of the example in Section 7.0.
MANOVA EFFICACY USAGE WITH SUPPORT EXP
/DISCRIM RAW STAN ESTIM CORR ALPHA (1)
/PRINT SIGNIF(MULT UNIV EIGN DIMENR )
/NOPRINTPARAM(ESTIM)
/METHOD=UNIQUE/DESIGN .
•
•
•
•
The expression ‘MANOVA’ in statement 1
requests a multivariate analysis of variance.
The dependent variables are computer selfefficacy and microcomputer usage. The
absence of the word ‘BY’ indicates that there
are no factors with lower measurement level
present. The word ‘WITH’ indicates that the
covariates with an interval measurement
level will follow; these are management
support and computer experience.
In the second statement, a canonical
correlation analysis is requested with
‘DISCRIM’. ‘STAN’ means that we would
like to have the standardized canonical
weights. With ‘CORR’ we request the
correlations between the original variables
and the canonical variables; these are the
structure correlations. ‘ALPHA’ means that
we establish the significance level of the
canonical variables. The default for this is
0.15. By setting the maximum at 1 (‘ALPHA
(1)’), we can make sure that all of the
canonical variables, even the insignificant
ones, will be included in the analysis.
2
3
Introduction
• is a multivariate statistical model which facilitates
the study of interrelationships among sets of
multiple criterion (dependent) variables and
multiple predictor (independent) variables
• is a powerful multivariate technique because it
can address a wide range of objectives.
4
Assumptions
• correlation is used only when
there are a set of multiple
dependent and independent
variables
• performed on correlation or
variance-covariance matrices
that reflect only linear
relationships
• conceptually linked sets of the
variables before applying
canonical correlation analysis
• all variables and all linear
combinations of variables are
normally distributed
• multicollinearity among either
variable set will isolate the
impact of any single variable
and this makes interpretation
less reliable
5
Procedural Steps
1. classify two sets of
variables or determine
the magnitude of the
relationships which may
exist among the two sets
2. derives a set of weights
for each set of
dependent and
independent variables
such that the linear
combinations
themselves are
maximally correlated
3. explain the nature of
whatever relationships
exist among the sets of
dependent and
independent variables,
generally by measuring
the relative contributions
of each variable to the
canonical functions
(relationships) that are
extracted
6
Data Analysis
Min canonical loading to be used for the analysis
Min R2 values in the result
References
7
8
Model Representation
9
Mechanical Steps
1: Objectives of Canonical
Correlation Analysis
2: Designing a Canonical
Correlation Analysis
3:Testing the Assumptions
4: Deriving the Canonical
Functions and Assessing
Overall Fit
5: Interpreting the
Canonical Variates
10
Interpretation of results
• Refer to the paper for discussion starting
from 122-123
11
Assessing Overall Fit
• Refer to p122
Valid one
12
Canonical Functions
• Refer to page 122
Here, it refers to only two significant functions are valid
13
Testing the Assumptions
• Refer to page 121
14
Designing a Canonical Correlation Analysis
• Define:
– What is individual variables from ethics literature
– What is seven ethical issues from ethics literature
15
Objectives of Canonical Correlation Analysis
• Refer to the page 117 of the attached file:
• The objective is as follows:
16

similar documents