Lecture
Of the laws of distribution of a system of two random variables, it makes sense to specifically consider the normal law as having the most widespread practice. Since a system of two random variables is depicted as a random point on a plane, the normal law for a system of two variables is often called the “normal” law on a plane.
In the general case, the density of the normal distribution of two random variables is expressed by the formula
. (9.1.1)
This law depends on five parameters: and . The meaning of these parameters is easy to establish. Prove that the parameters represent the mathematical expectation (centers of dispersion) values and ; - their standard deviations; - coefficient of correlation of quantities and .
In order to verify this, we first find the distribution density for each of the quantities in the system. According to the formula (8.4.2)
.
Calculate the integral
.
Put:
(9.1.2)
then
.
From integral calculus it is known that
. (9.1.3)
In our case
.
Substituting these values into the formula (9.1.3), we have:
,
from where
,
or, considering (9.1.2)
. (9.1.4)
Thus, the magnitude subject to normal law with a dispersion center and standard deviation . Similarly, we show that
, (9.1.5)
those. magnitude subject to normal law with a dispersion center and standard deviation .
It remains to prove that the parameter in the formula (9.1.1) is the correlation coefficient of the quantities and . To do this, we calculate the correlation moment:
,
Where - expected values and .
Substituting the expression in this formula , we get:
, (9.1.6)
Where
.
We make in the double integral (9.1.6) the change of variables, putting:
. (9.1.7)
The conversion jacobian is
,
Consequently,
Considering that
we have:
(9.1.8)
Thus, it is proved that in the formula (9.1.1) is the correlation coefficient of the quantities and .
Suppose now that random variables and , subordinate to the normal law on the plane, not correlated; we put in the formula (9.1.1) . We get:
. (9.1.9)
It is easy to make sure that random variables , subject to the law of distribution with density (9.1.9), are not only uncorrelated, but also independent. Really.
.
those. the distribution density of the system is equal to the product of the distribution densities of the individual quantities in the system, which means that random variables are independent.
Thus, for a system of random variables subject to the normal law, their independence also follows from the uncorrelated values. The terms "uncorrelated" and "independent" values for the case of normal distribution are equivalent.
With random variables are dependent. It is easy to verify, by calculating the conditional laws of distribution using formulas (8.4.6), that
Let's analyze one of these conditional distribution laws, for example . To do this, convert the density expression to the form:
.
Obviously, this is the density of a normal law with a center of dispersion
(9.1.10)
and standard deviation
. (9.1.11)
Formulas (9.1.10) and (9.1.11) show that in the conditional distribution law with a fixed value only the expectation depends on this value, but not the variance.
Magnitude called conditional expectation of magnitude at this . Dependence (9.1.10) can be represented on the plane , deferring conditional expectation y-axis. You’ll get a straight line called a regression line. on . Similarly straight
(9.1.12)
there is a regression line on .
Regression lines coincide only if there is a linear functional dependence from . With independent and regression lines are parallel to the coordinate axes.
Considering the expression (9.1.1) for the density of the normal distribution on the plane, we see that the normal law on the plane is completely determined by setting five parameters: the two coordinates of the center of dispersion two standard deviations and one correlation coefficient . In turn, the last three parameters and completely determined by the elements of the correlation matrix: dispersions and the correlation point . Thus, the minimum number of numerical characteristics of a system — mathematical expectations, variances, and the correlation moment — in the case when the system of subordination to the normal law determines the distribution law completely, i.e. forms a comprehensive system of characteristics.
Since in practice the normal law is quite common, it is often enough to set the minimum number — only five — of the numerical characteristics to fully characterize the distribution law of a system.
Comments
To leave a comment
Probability theory. Mathematical Statistics and Stochastic Analysis
Terms: Probability theory. Mathematical Statistics and Stochastic Analysis