The Chow-Test is a statistical test with which the coefficients of two linear regressions can be tested for equality. The test is named after its inventor, the economist Gregory Chow.
The CHOW test is used in economics to test time series on structural fractures. Another area of application is the programvaluation, here two different sub -groups (programs), such as two school types, are compared. In contrast to the time series analysis, the two subgroups cannot be assigned to any consecutive intervals, instead the division is made according to a qualitative aspect, such as the school type.
There is a data set
with
for
whose relationship through a linear function with a normal distributed error (
) with expectation value 0 (
) is described (multiple regression analysis), i.e. H. One has
-
for
.
However, one suspects that the data record is in two groups of the sizes
and
Division, which are better described by two different linear functions.
-
for
-
for
Here is
And it becomes the hypothesis
against
tested. If one refers to the sum of the square residues of the regression over the entire data set
and with the two sub -groups with
and
, then the test size defined below follows
an F distribution with the degrees of freedom
and
.
-
The following data record is given, the relationship of which is due to the linear function
should be modeled:
|
0.5 |
1.0 |
1.5 |
2.0 |
2.5 |
3.0 |
3.5 |
4.0 |
4.5 |
5.0 |
5.5 |
6.0 |
|
−0.043 |
0.435 |
0.149 |
0.252 |
0.571 |
0.555 |
0.678 |
3,119 |
2.715 |
3.671 |
3.928 |
3,962 |
The data plot sets a structural fracture
vicinity.
A data plot suggests that at
There is a structure of a structural fraction, so the data record is divided into 2 intervals
and
and runs separate regressions via this, in addition to regression over the entire data record. Then you test whether the two sub -regressions create the same linear function, i.e.
against
Regression on the entire data record:
|
|
|
|
|
|
Regression on
|
|
|
|
|
|
Data plot with regression line
Regression on
|
|
|
|
|
|
Calculation of the test size:
-
Because of
(Significance level
) is applicable
. So the null hypothesis can
be rejected. This means that the two regression straight on the partial intervals are not identical. So there is a structure of the structure and the partial regressions provide better modeling than regression over the entire data set.
Recent Comments