 Canonical analysis computes weights
for two sets of variables, A and B, such that it
maximizes the simple product moment correlation
between the weighted composite scores for set A and set
B.
 The canonical correlation is
interpreted exactly as a simple productmoment
correlation.
 The weights on both sides of the
equation (canonical variates) are analogous to
bcoefficients in linear regression.
 More than one canonical
correlation is produced for each analysis.
 The number of canonical
correlations (roots) equals the larger of the two
sets of variables.
 The first canonical correlation
is the largest; successive ones are progressively
smaller.
 Scores produced from canonical
variates are unrelated across roots.
 Whereas factor analysis can be used
appropriately as a tool for exploratory research,
canonical analysis needs to be guided by
theory.
 Canonical analysis is a tool for
testing theory; not developing measures.
 Alternative measures of the same
concept should not be included in a canonical
variate.
 Instead, combine the measures to
form a scale, compute its reliability, and use the
scale in analysis.
 Consider canonical analysis for
relating variables grouped under "umbrella
concepts."
 Umbrella concepts group together
variables that share some conceptual underpinnings but
that express distinct phenomena.
 Consequently, the concepts are not
closely relatedin the sense of
unidemensionality.
 Some examples of umbrella
concepts:
 International relations,
which embraces
 war and peace
 armed conflicts
 economic trade
 and so on.
 Congressional behavior,
which embraces
 party unity in congressional
voting
 introduction of
bills
 committee
participation
 Political party
organization
 centralization of
power
 structural
complexity
 Political party
performance
 winning
elections
 providing services for
members
