2.5 Multiple Linear Correlation
Uploaded by
ScribdTranslations2.5 Multiple Linear Correlation
Uploaded by
ScribdTranslations2.
5 MULTIPLE LINEAR CORRELATION
PARTIAL CORRELATION
Partial correlation is defined as the correlation between two variables if the other
variables do not vary, that is, the value of the other variables is fixed, the partial
correlation coefficient of p 123 is the correlation between variable 2 and 3 being
constant. the value of variable 1. The partial correlation r 123 would be the linear
correlation between variable 1 and 2 leaving variable 3 as constant, this means
that the correlation between variable 1 and 2 must be measured that is not a
reflection of their relationships with variable 3. , therefore a sample estimate r 123
can be obtained by calculating the deviation or residual e 13 of the regression of
variable 1 on variable 3, and the deviation or residual e 23 of the regression of
variable 2 on variable 3. And r 12 3 is the simple correlation coefficient between e 13
and e 23 . Partial correlation does not involve the notion of independent and
dependent variables, but is a measure of interdependence.
It can be shown that r 123 satisfies the following formula:
r 12−r 13 r 23
r 12.3= ❑
√¿¿¿
The typical error of this estimate is:
sr =
n−3
123
√
1−r 212.3
❑
So we can prove H 0 : r 12.3 =0 by t
r 123 ❑√ n−3
t=
√ 1−r
❑ 2
12.3
Partial correlation coefficients of a given order can be deduced from those of an
immediately lower order. Thus, a first-order coefficient can be obtained by applying
the already known relationship of the zero-order coefficients.
r 12−r 13 r 23
r 12.3= ❑
√¿¿¿
For a second order coefficient the relationship is with first order coefficients, this is:
r 12 .3−r 1 4.3 r 2 4.3
r 12.3 4 = ❑
√(1−r 2
14.3 )(1−r
2
24.3 )
And a partial correlation of kth order
r 12.34 … k =r 12.3 4. .(k−1) −r 1 k .34 …(k−1) r 2 k .34 … ¿¿ ¿
The test to check if this r is different from zero is
r 12. 3 4 … k ❑√ n−k
t=
√1−r
❑ 2
12.3 4.. k
MULTIPLE CORRELATION
Multiple regression analysis consists of finding the best linear combination of p
independent variables, X 1 , X 2 ,.., X P , to predict the dependent variable Y.
Multiple correlation is the simple correlation between Y and its values estimated by
the regression equation , Ŷ . Therefore, the objective in regression and multiple
correlation analyzes is to examine the relationship between several variables, X,
and one variable, Y.
Remember that the multiple regression analysis tried to find the linear combination
of p variables, X 1 , X 2 ,..., X P , that best predict the dependent variable Y. The
multiple correlation coefficient is the simple correlation between Y and its prediction
by the regression equation
The multiple correlation coefficient is applied to the situation in which a variable,
called “Y”, has been isolated to examine its relationship with the set of other
variables. This coefficient is determined by the following expression:
R1.23=❑ r
√ 1 2+¿
2
r 13−¿ 2r
1−r
2
2
r r23
12 13
23
¿¿
Being R 2 the coefficient of multiple determination. The value of a multiple
correlation coefficient, R, lies between zero and one. The closer it is to one, the
greater the degree of association between the variables, and the closer it is to
zero, the worse the linear relationship.
There is a relationship between the multiple correlation coefficient and the different
partial correlation coefficients, which can facilitate the calculation of the former, this
is:
1 – R 2 1.23 = (1 – r 2 12 ) (1 – r 2 13.2 )
1 – R 2 1.234 = (1 – r 2 12 ) (1 – r 2 13.2 ) (1 – r 2 14.23 )
A multiple correlation of kth order
1 – R 2 1.23…k = (1 – r 2 12 ) (1 – r 2 13.2 ) (1 – r 2 1k.23…(k-1) )
Just as in the fit test of a multiple regression, R 2 is the fraction of the sum of
squares of the deviations of Y from its mean, attributable to the regression, while
(1-R 2 ) is the unassociated fraction to regression; now, the test of the null
hypothesis, that the multiple correlation in the population is zero, is identical to the
F test of the null hypothesis that β1=β2=...=βk=0; And this is:
( n−k ) R2
F=
( K−1 ) (1−R 2)
Being R the coefficient of multiple determination. This F 0 is contrasted with F (K-1, nk;
a ).
CANONICAL CORRELATION
Canonical correlation analysis is applied to situations where the regression
technique is appropriate, but for more than one dependent variable. Although
another application of canonical correlation analysis is as a method to determine
the association between two groups of variables. It is a generalization of multiple
regression to the case of more than one dependent variable. This analysis is
closely related to canonical discriminant analysis and has certain properties
analogous to principal components analysis and factor analysis, in which instead of
trying to study the internal dependencies between the variables of the same group,
in the case of Canonical correlation What is studied is the relationship or
dependence between two groups of variables.
Canonical correlation analysis examines the linear relationship between a group of
variables, X, and a group, or more than one group, of variables Y. So the
difference is that now you have more than one Y variable. The technique consists of finding a
linear combination of the variables 2 and 2 +...+a q and q ) in such a way that the
correlation between U and V is maximum. Then find two other linear combinations
for each group of variables that have maximum correlation and so on, find a set of
linear combinations for each group of variables that have maximum correlation.
These linear combinations are called canonical variables, and the correlations
between the corresponding pairs of canonical variables are called canonical
correlations.
Canonical correlation analysis is applied to situations in which the regression
technique is appropriate, but there is more than one dependent variable. Another
useful application is to test the independence between the two groups of variables,
Y and X, as will be seen in a moment.
