Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Xi
*
*
* Ci1
* * *
O
*
*
*
*
*
j 1
j
i 1
i
j 1
j
Similarly, approaching the problem from the variable spaces, at the k step
(phase), the correlation between the new Ck component and the previously
determined ones is:
m
1 (Ck ) t XX t Ck (Ck ) t k Ck
j 1
R (Ck ,X j )
2
n (Ck ) Ck t
t
(Ck ) Ck
k .
1. Kaiser criterion:
1. Cattell criterion:
2. Cattell criterion:
Some researchers simply use the rule of keeping enough factors to account
for 90% (sometimes 80%) of the variation.
Where the researcher's goal emphasizes parsimony (explaining variance
with as few factors as possible), the criterion could be as low as 50%.
Scores
ik
c 2
k 1
k 1
The quality of representing the i observation on aj axis, is determined by
cij2
the ratio: m
ik
c 2
k 1
The value of the ratio is equal with square cosine of the angle between the
vector associated to point i and aj vector.
h R X j , Ck
s
2 2
k 1
R X j , Ck
s
2
For s = m, h
2
k 1
becomes equal to 1, meaning that those m principal components explain
entirely the information from the initial data table X.
Correlation coefficients
Correlation coefficients
Having given the hypothesis that the initial variables are only centered, but
not normalized.
The initial variables variance is no longer 1.
The analysis is conducted on covariance matrix, since:
1 t
X X
n
is the covariance matrix of the observation tables.
In the observation space, the optimum criterion at any given phase remains
the same, but it applies to a different cloud of points.
The vectors a k , k 1, m , are eigenvectors of the covariance matrix.
Maxim R
2
(Ck , X j )
Ck j 1
becomes:
m
Maxim
2
Cov (Ck , X j )
Ck j 1
Maxim i ik
2
p c
i 1