Distributions of Functions of Normal Random Variables: The Unit (Or Standard) Normal

Distributions of Functions of
Normal Random Variables

The Unit (or Standard) Normal
The unit or standard normal random variable U is a normally distributed variable
with mean zero and variance one, i.e. U N(0, 1). Note that if x N(, 2 ) that
x
U N(0, 1)

Thus, to simulate a normal random variable with mean and variance 2 , we can
simply transform unit normals, as
x +U
If x N(, 2 ), then x N( , 2 /n ). Thus
x
U N(0, 1) (A5.1)
/ n
Central and Noncentral 2 Distributions

The 2 distribution arises from sumsPof squared, normally distributed, random
variables if xi N(0, 1), then u = i=1 x2i 2n , a central 2 distribution with
n
n degrees of freedom. It follows that the sum of two 2 random variables is also
2 distributed, so that if u 2n and v 2m , then
u + v 2(n+m) (A5.14a)
Two other useful results are that if xi N(0, 2 ), then
X
n
x2i 2 2n (A5.14b)
i=1
X
n
and for x = n1 xi ,
i=1
X
n
(xi x )2 2 2(n1) (A5.14c)
i=1
In this last case, subtraction of the mean causes the loss of one degree of freedom.
The 2 distribution is thus vital for both hypothesis testing and construction of
1
2 APPENDIX 5
confidence intervals of unknown variances, as Equation A5.12c implies that if

x N(0, o2 ) then for the sample variance
1 X
n
(n 1) S 2
S2 = ( xi x )2 , we have 2n1
n 1 i=1 o2
A noncentral 2 arises when the random variables P being considered have

nonzero means. In particular, if xi N(i , 1), then u = i=1 x2i follows a non-
n
central distribution with n degrees of freedom and noncentrality parameter

2
X
n
= 2i (A5.15a)
i=1
and we write u 2n, . As shown in Figure A5.3, increasing the noncentrality

parameter shifts the distribution to the right. This is also seen by considering
the mean and variance of u,
E(u) = n + and 2 (u) = 2(n + 2) (A5.15b)
=0
=1
=5
0 5 10 15 20 25
Figure A5.3 The probability distribution function for a noncentral 2 . As the

noncentrality parameter increases, the distribution is pulled to the right.
We plot here a 2 random variable with n = 5 degrees of freedom and non-
centrality parameters = 0 (a central 2 ), 1, and 5.
It follows directly from the definition that sums of noncentral 2 variables

also follows a noncentral 2 distribution, so that if u 2n, 1 and v 2m, 2 , then
(u + v) 2(n+m),(1 +2 ) (A5.15c)
Finally, Equations A5.14b,c can be generalized to noncentral 2 random variables
as follows. Suppose xi N(i , 2 ), then
X
n X
n
x2i 2 2n, where = 2i (A5.15d)
i=1 i=1
Distributions of Functions of Normals 3
and
X
n Xn
2i
( xi x )2 2 2(n1), where = (A5.15e)
i=1 i=1
2
Students t Distribution

If x N(, 2 ), then (Equation A5.1) ( x )/( / n ) N(0, 1), which allows
for both hypothesis testing and construction of confidence intervals when 2 is
know. When the variance is unknown, the above test statistic replaces the true
(but unknown) variance 2 with the sample variance S 2 ,
1 X
n
z
t= where S2 = ( zi z )2
S/ n n 1 i=1
Notice that
!
z 1 U
t= p =q
/ n S 2 / 2 n1 /(n 1)
2
Thus we define a t random variable with degrees of freedom by
U
t p
2 /
This distribution has mean zero and variance 2 (t ) = 1 + 2/( 2) (for > 2).
The coefficient of kurtosis is k4 = 6/( 4), implying that the t distribution has
heavier tails than a normal. The noncentral Students t- distribution is defined
as follows. If x N(o , 2 ), but we assume the correct mean in , then
z
t, =
S/ n
is distirbution an a noncentral t with degrees of freedom and noncentrality

parameter = ( o )/.
Central and Noncentral F Distributions

The ratio of two 2 -distributed variables leads to the F distribution. In particular,
if u 2n and v 2m , then the ratio of these two 2 variables divided by their
respective degrees of freedom follows a central F distribution with numerator
and denominator degrees of freedom n and m (respectively), i.e., (u/n)/(v/m)
Fn,m . Since
2
lim Fn,m n
m n
4 APPENDIX 5
the F distribution can be approximated by a 2n when the denominator degrees

of freedom is large. The F distribution was introduced by Snedecor. Notice that
the square of the t distribution is
U2 2 /1
t2 2
= 21 F1,
/ /
Thus the square of a t distribution with n degrees of freedom is distributed as an
F with 1 and n degrees of freedom.
The noncentral F distribution results when the numerator 2 variable is
noncentral. If u 2n, and v 2m , then F = (u/n)/(v/m) follows a noncentral
F distributed with noncentrality parameter , and we write F Fn,m, . As with
the noncentral 2 , increasing shifts the distribution further to the right. Again,
this is seen in the mean and variance, with

m 2
E(F ) = 1+ (A5.16a)
m2 n
m 2 (n + m)2 + (n + 2)(m 2)
2
(F ) = 2 (A5.16b)
n (m 2)2 (m 4)
Various mathematical and statistical packages provide routines for comput-
ing cumulative probabilities of noncentral 2 and F random variables, and a
number of approximations have been suggested (e.g., Patnaik 1949, Severo and
Zelen 1960, Tiku 1965). Winer et al. (1991) offer one such approximation based on
the unit normal U , with the probability that a noncentral F -distributed random
variable exceeds a value Fo being approximately
Pr( Fn,m, > Fo ) ' Pr( U > zo ) (A5.17a)
where
p p
(2m 1) B 2(n + ) A n + 2 n
zo = , A= , B= Fo (A5.17b)
A+B n+ m
From this general expression follow simplified approximations for the special
cases of central F and noncentral 2 variables. Setting = 0 gives an approxima-
tion for the central F distribution as
p
(2m 1) B 2n 1
Pr(Fn,m > Fo ) ' Pr( U > zeo ), with zeo = (A5.17c)
1+B
Likewise, taking the limit as m offers an approximation for the noncentral
2 , since 2n, n Fn,, . Taking the limit of Equation A5.17c as m gives
the probability that a noncentral 2 exceeds a value Co as approximately
p
2Co 2(n + ) A
Pr( n, > Co ) ' Pr( U > zeo ), with zeo =
2
(A5.17d)
A

Distributions of Functions of Normal Random Variables: The Unit (Or Standard) Normal

Uploaded by

Copyright:

Available Formats

Distributions of Functions of Normal Random Variables: The Unit (Or Standard) Normal

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Distributions of Functions of Normal Random Variables: The Unit (Or Standard) Normal

Uploaded by

Copyright:

Available Formats

Distributions of Functions of

Normal Random Variables

If x N(, 2 ), then x N( , 2 /n ). Thus

Central and Noncentral 2 Distributions

Two other useful results are that if xi N(0, 2 ), then

confidence intervals of unknown variances, as Equation A5.12c implies that if

A noncentral 2 arises when the random variables P being considered have

central distribution with n degrees of freedom and noncentrality parameter

and we write u 2n, . As shown in Figure A5.3, increasing the noncentrality

Figure A5.3 The probability distribution function for a noncentral 2 . As the

It follows directly from the definition that sums of noncentral 2 variables

Thus we define a t random variable with degrees of freedom by

is distirbution an a noncentral t with degrees of freedom and noncentrality

Central and Noncentral F Distributions

the F distribution can be approximated by a 2n when the denominator degrees

You might also like