Continuous Random Variables and Probability Density Functions

A random variable X is said to be continuous if its set of possible values is an entire interval of
numbers – that is, if for some A < B, any number x between A and B is possible. Let X be a
continuous random variable. Then a probability density function (pdf) of X is a function f(x) such
that for any two numbers a and b with a  b ,
b
(a    b)   f ( x)dx .
a

That is, the probability that X takes on a value in the interval [a, b] is the area above this interval
and under the graph of the density function f(x) which is referred to as the density curve.

f(x)

x
a b

Fig. 1: (a    b)  the area under the density curve between a and b.

The two conditions that must be satisfied for f(x) to be a legitimate pdf:

(i) f ( x)  0 for all x.


(ii) 

f ( x ) dx = area under the entire graph of f(x)

=1
If X is a continuous random variable, then for any number c, P(X = c) = 0. Furthermore, for any
two numbers a and b with a < b,
(a    b)  (a    b)  (a    b)  (a    b)

The cumulative distribution function F(x) for a continuous random variable X is defined for
every number x by
x
F ( x)  (   x)   f ( y )dy

For each x, F(x) is the area under the density curve to the left of x. This is illustrated in Figure 2,
where F(x) increases smoothly as x increases.

f(x) F(x)
1
F(8)
F(8)
.5

x x
5 8 10 5 8 10
Fig. 2: A pdf and associated cdf

Let X be continuous random variable with pdf f(x) and cdf F(x). Then for any number a,
P(X > a) = 1 - F(a)

and for any two numbers a and b with a < b,

(a    b)  F (b)  F (a)

The desired probability is the shaded area under the density curve between a and b, and it equals
the difference between the two shaded cumulative areas.

f(x)

a b b a
Fig 3: Computing (a    b) from cumulative probabilities

Example 1

Suppose the pdf of the magnitude X of a dynamic load on a bridge (in Newton’s) is given by

1 3
  x 0 x2
f ( x)   8 8
0 otherwise


For any number x between 0 and 2,

x x1 3  x 3
F ( x)   f ( y )dy     y dy   x 2


0 8 8  8 16
Thus,

 0 x0

 x 3
F ( x)    x 2 0  x  2
 8 16
 1 2 x


The probability that the load is between 1 and 1.5 is

(1    1.5)  F (1.5)  F (1)

1 3  1 3 
  (1.5)  (1.5) 2    (1)  (1) 2 
8 16  8 16 
19
  .297
64
The probability that the load exceeds 1 is

1 3 
(   1)  1  (   1)  1  F (1)  1   (1)  (1) 2 
8 16 

11
  .688
64
If X is a continuous random variable with pdf f(x) and cdf F(x), then at every x which the
derivative F (x) exists, F ( x)  f ( x) .

Continuous Probability Functions: The conditional expectation of X given that even A has
occurred is

    xf (x )dx .




If X is a discrete, then integral will replace with summation as

     xi   (xi ) .
i

The probability density function (pdf), if it exists, is given by

dF ( x)
f ( x)  ,
dx

where F ( x)  F ( x) , since we are dealing with only a single random variable.

Properties of pdf
If f(x) exists then:

(a) 

f ( )d  F ()  F ( )  1

x
(b) F ( x)   f ( )d  (   x)


x2 x1
(c) F ( x 2 )  F ( x1 )   f ( )d   f ( )d
 

x2
  f ( )d  ( x1    x2 ) .
x1

Interpretation of f(x)
( x    x  x)  F ( x  x)  F ( x)

If F(x) is continuous in its first derivative then, for sufficiently small x ,

x  x
F ( x  x)  F ( x)   f ( )d  f ( x)x
x

Observe that if f(x) exists, then F(x) is continuous and P(X = x) = 0.

Some examples of Continuous probability functions are:

The univariate normal (Gaussian) pdf: The pdf is given by

1  x
2

1 2  
f ( x)  e  

2 2

There are two independent parameters:  , the standard deviation (  2 is the variance) and  ,
the mean. If a random variable X obeys the normal probability law with mean  and standard
deviation  , we write in symbol as  :  (  ,  2 ) . The normal pdf is widely encountered in all
branches of science, engineering, social and demographic studies. For example, the masses of
lecturers in a university, the intelligent quotient of children, the heights of a growing child, the
yields of agricultural produce in a farm, the noise voltage produced by a thermally agitated
resistor, all are postulated to be approximately normal over a large range of values.
Properties of the Normal Distribution

(a) The normal distribution function depends on the mean  and the standard deviation  .
(b) The normal distribution curve is bell-shaped.
(c) The curve is asymptotic to the x-axis.

(d) The function is continuous from   to   .

(e) The curve is symmetrical about the vertical line through the mean.

Since a normal distribution function is a probability function, the total area under its curve is 1.

Statisticians have found it rather convenient to choose a normal curve with mean 0 and standard
deviation 1. Such a normal distribution curve is called a standard normal distribution curve or a
standard normal curve.
p(x)

Standard normal (z) curve

 0 z
Fig. 4: A Standard Normal Curve,
where p(x) axis becomes an axis of symmetry. If the probability distribution has mean 0 and
standard deviation 1, we say that the distribution has been standardized and z is called
standardized score or z-score.

x
z .

A table of standard normal distribution is available in most statistical tables.

When  ~ (,  2 ) , probabilities involving X are computed by “standardizing.” The

standardized variable is (   ) /  . Subtracting  shifts the mean from  to zero, and then
dividing by  scales the variable so that the standard deviation is 1 rather than  .


If X has a normal distribution with mean  and standard deviation  , then Z 

has a standard normal distribution. Thus

a b
 ( a    b )   Z 
   
 b  a
     
     

a b 
(   a )    (   b)  1   
     

The standardized normal table can be used. The proposition can be proved by writing the cdf of
Z  (    ) /  as
z  
( Z  z )  (   z   )   f ( x;  ,  ) dx


Using a result from calculus, this integral can be differentiated with respect to z to yield the
desired pdf f(z: 0, 1).
N(0, 1)

( , 2 )
=

 x 0
(x  ) / 
Fig 5: Equality of nonstandard and standard normal curve areas
Example 2

The life-length of an electronic device manufactured by Company A is normally distributed with

mean 45 and standard deviation 8, while that of a similar electronic device manufactured by
Company B has mean life-length 48 and standard deviation 4, all measurements being in hours.
Which of the electronic devices is to be preferred if it is required for:

(i) a 48 hour period;

(ii) a 52 hour period.

giving reasons for your answer.

Solution

Let  1 be a random variable which is a measure of the life-length of the electronic device from
Company A and let  2 be a random variable which is a measure of the life-length of the
electronic device from Company B.
We shall compare:
 (i) ( 1  48) and (  2  48 )

    48   
(1  48)   1  
   
 48  45 
  Z  
 8 
 ( Z  0.375)  1  ( Z  0.375)
= 1 – 0.6461
= 0.3539
    48   
(  2  48)   2  
  4 
 ( Z 2  0)  1  ( Z 2  0)
 1  0.5000
 0.5
Comment: The electronic device of Company B is to be preferred to the electronic device of
Company A because it has a greater chance of lasting more than 48 hours than the electronic
device of Company A.

   45 52  45 
(ii) ( 1  52)   1  
 8 8 
 ( Z1  0.875 )  1  ( Z  0.875 )
= 1 – 0.8092
= 0.1908
   48 52  48 
(  2  52)   2  
 4 4 
 ( Z 2  1)  1  ( Z 2  1)
= 1 – 0.8413
= 0.1587
Comment: The electronic device of Company A is to be preferred to the electronic device of
Company A because it has a greater chance of lasting more than 52 hours than the electronic
device of Company B.
Example 3
The yields, in kilogrammes, of tomatoes from 10 identical plots on a farm are:

30, 36, 39, 48, 27, 42, 39, 48, 51, 30

Calculate the (a) mean and (b) standard deviation of the yields.

Assuming that the yields are normally distributed with these values of the mean and standard
deviation, find the yield W such that the probability of the yield from a plot being greater than W
kilogrammes is 5%.
Solution

(a) Let the mean of the distribution be x ; and let the standard deviation be S.

x xx (x  x)2
30 -9 81
36 -3 9
39 0 0
48 9 81
27 -12 144
42 3 9
39 0 0
48 9 81
51 12 144
50 -9 81
390 630

x
 x  390  39
 10

S
 (x  x) 2


630

10
 7.937

(b) Let X be a random variable which is a measure of the yields from the plot. We are required to
 find P(X > W) = 0.05.

 0 W

It follows that (  W )  1  0.05  0.95

To find a Z whose probability correspond to 0.95, we use the standard normal distribution table
inverse. From the Table, this is 1.645

Z 
 1.645

W  39
 1.645
7.937
W  39  1.645  7.937
 W  39  13.06

= 52.06kg

Example 5
In a very large collection of plants, it is found that 20% have heights greater than 36.3 cm and
67% have heights greater than 29.9 cm. Assuming that height x cm is normally distributed in this
collection, find the mean and standard deviation of the heights of the plants in the collection.

What is the probability that the height x cm of a plant will exceed 33.0 cm?
Solution

Let X be a random variable which is a measure of the height of a plant.

(   36 .3)  0.2
(   36 .3)  0.8

    36.3   
    0.8
   
 36.3   
  0  Z    0.3
  
From the table,

36.3  
 0.845

 36 .3    0.845  (1)

(   29 .9)  0.67
(   29 .9)  0.33

    29.9   
    0.33
   
 29.9   
  Z    0.33
  

From the table,

29.9  
 0.44

 29 .9    0.44 (2)

Solving Equation (1) and (2) simultaneously, we have

6.4  1.285
   4.981

From (1), we have

36.3    0.845
   36.3  (0.845  4.981)
 36.3  4.2089

= 32.0911cm

 32.1 cm

    33 .0  32 .1 
(b) (   33 .0)  1    
  4.981 

 1  (Z  0.1807)

= 1 – 0.5714

= 0.4286
The Gamma Distribution and Its Relatives: The graph of any normal pdf is bell-shaped and thus
symmetric. There are many practical situations in which the variable of interest to the
experimenter might have a skewed distribution. A family of pdf’s that yields a wide variety of
skewed distributional shapes is the gamma family. For   0 , the gamma function ( ) is
defined by

( )   x  1e  x dx . (3)
0

The properties of the gamma function are:

(i) For any   1, ( )  (  1)  (  1) [via integration by parts]

(ii) For any positive integer, n, (n)  (n  1)!

(iii)  12   

By Equation (3), if we let

 x  1e  x
 x0
f ( x;  )   ( ) (4)
0 otherwise



then f ( x; )  0 and 0
f ( x;  )dx  ( ) / ( )  1 , so f ( x;  ) satisfies the two basic

properties of a pdf.

The family of Gamma Distributions: A continuous random variable X is said to have a gamma
distribution if the pdf of X is

 1  1  x / 
   ( ) x e x0
f ( x;  ,  )   (5)
0 otherwise


where the parameters  and  satisfy   0,   0 . The standard gamma distribution has
  1 , so the pdf of a standard gamma random variable is given by (4).
f ( x; ,  ) f ( x;  )

1.0   2,   1
3 1.0  1

  1,   1   .6
0.5   2,   1 0.5  2

  2,   2  5
0 x 0 x
1 2 3 4 5 6 7 1 2 3 4 5
Fig. 6: (a) Gamma density curves (b) Standard gamma density curves

Figure 6(a) illustrates the graphs of the gamma pdf f ( x; ,  ) (2.14) for several ( ,  ) pairs,
whereas (b) presents graphs of the standard gamma pdf. For the standard pdf, when   1 , f ( x;  )
is strictly decreasing as x increases from 0; when   1, f ( x;  ) rises from 0 at x = 0 to a
maximum and then decreases. The parameter  in (2.14) is called the scale parameter because
values other than 1 either stretch or compress the pdf in the x direction.

() and (  2 ) can be obtained from a reasonably straightforward integration, and then
V ( )  (  2 )  [( )] 2 . The mean and variance of a random variable X having the gamma
distribution f ( x; ,  ) are

()     V ( )   2   2

Exponential (   0 ): The exponential pdf is a special case of the general gamma pdf in which
  1 and  has been replaced by 1 /  . X is said to have an exponential distribution with
parameter  (  0) if the pdf of X is

e  x x0

f ( x;  )   .
0 otherwise


The mean and variance of X are

 1 1
     2   2 
 2
1
Both the mean and standard deviation of the exponential distribution equal . The exponential

pdf can easily be integrated to give the cdf of X as:
0 x0

F ( x;  )  
 x
1  e x0

The exponential law occurs in waiting-time problems, lifetime of machinery and in describing
the intensity variations of incoherent light. Thus, the exponential distribution is frequently used
as a model for the distribution of times between occurrence of successive events, such as
customers arriving at a service facility or calls coming in to a switchboard.

Suppose that the number of events occurring in any time interval of length t has a Poisson
distribution with parameter t (where  , the rate of event process, is the expected number of
events occurring in 1 unit of time) and that numbers of occurrences in non-overlapping intervals
are independent of one another. Then the distribution of elapsed time between the occurrence of
two successive events is exponential with parameter    . For example, for the time  1 until
the first event occurs is given as:
(  1  t )  1  (  1  t )  1  [( no events in (0, t )]

e t  (t ) 0
 1  1  e t
0!
which is exactly the cdf of the exponential distribution.

Uniform ( b  a ): A continuous random variable X is said to have a uniform distribution on the

interval [a, b] if the pdf of X is

 1
 a xb
f ( x)   b  a
0 otherwise


The uniform pdf is used in communication theory, in queuing models and in situation where we
have no a priori knowledge favouring the distribution of outcomes except for the end points; that
is we don’t know when a business call will come but it must come
Rayleigh (   0 ):

x / 2 2
f ( x)  e x
2
u ( x) .
 2

where the function u (x) is the unit step, that is, u ( x)  1, x  0, u ( x)  0, x  0 . Thus,
f ( x)  0 for x < 0. Examples of where the Rayleigh pdf shows up are in rocket-landing errors,
random fluctuations in the envelope of certain waveforms and radial distribution of misses
around the bull’s eye at a rifle range.

1 Exponential


0.606


Rayleigh Uniform
1
ba

0  a b

Fig. 2.6: The Rayleigh, exponential and uniform probability density functions (pdf’s)

The Chi-Squared Distribution: The chi-squared distribution is important because it is the basis
for a number of procedures in statistical inference. Let v be a positive integer. Then a random
variable X is said to have a chi-squared distribution with parameter v if the pdf of X is the
gamma density with   v / 2 and   2 . The pdf of the chi-squared random variable is

 1 ( v / 2 ) 1  x / 2
 2 v / 2 (v / 2) x e x0
f ( x; v)  
 0 x0


The parameter v is called the number of degrees of freedom of X. The symbol of “chi-squared” is
2.

There are other families of continuous distributions used by Statisticians to tackle many practical
situations in which no member of normal, gamma, exponential and uniform families of
distributions (that provides a wide variety of probability models for continuous variables) fits a
set of observed data very well. They are:

The Weibull Distribution: A random variable X is said to have a Weibull distribution with
parameters  and  (  0,   0) if the pdf of X is

   1 ( x /  )
  x e x0
f ( x;  ,  )   
 0 x0


When   1, the pdf reduces to the exponential distribution (with   1 /  ) , so the exponential
distribution is a special case of both the gamma and Weibull distributions. However, there are
gamma distributions that are not Weibull distributions and vice versa, so one family is not a
subset of the other.

Integrating to obtain E(X) and (  2 ) yields

 1 
  2   1  
2

  1      1    1   
2 2

          
 

The computation of  and  2 thus necessitates using the gamma function. The integration
x
0
f ( y;  ,  )dy is easily carried out to obtain the cdf of X. The cdf of a Weibull random variable

parameters  and  is

 0 x0

F ( x;  ,  )  
 ( x /  )
1  e x0


The Lognormal Distribution: A nonnegative random variable is said to have a lognormal

distribution if the random variable Y = ln(X) has a normal distribution. The resulting pdf of a
lognormal random variable when ln(X) is normally distributed with parameter  and  is

 1 [ln( x )   ]2 /( 2 2 )
 e x0
f ( x;  ,  )   2 x
 0 x0


The mean and variance of X is

 ( )  e   V ( )  e 2    (e  1)
2 2 2
/2

A lognormal curve has a positive skew. Because ln(X) has a normal distribution, the cdf of X can
be expressed in terms of the cdf (z) of a standard normal random variable Z. For x  0 ,

F ( x;  ,  )  (  x)  [ln()  ln( x)]

 ln( x)     ln( x)   
  Z     
     

Example 6

The article “Reliability of Wood Joist Floor Systems with Creep” suggests that the lognormal
distribution with   0.375 and   0.25 is a plausible model for X = the modulus of elasticity
(MOE, in 106 psi) of wood joist floor systems constructed from #2 grade hem-fir. Find

(a) The expected value of MOE.

(b) The variance of MOE.

(c) The probability that MOE is between 1 and 2.

(d) What value c is such that only 1% of all systems have an MOE exceeding c?

Solution
.375  (.25) 2 / 2
(a) The mean value of MOE is ()  e  e.40625  1.50

(b) The variance, ( )  e .8125 (e .0625  1)  0.1453

(c) (1    2)  ln(1)  ln( )  ln( 2) 

 (0  ln( )  .693)

 0  .375 .693  .375 
  Z 
 .25 .25 
 (1.27)  (1.50)  0.8312

 ln( c)  .375 
(d) .99  (   c)   Z  
 .25 

from which (ln(c) - .375)/.25 = 2.33 and c = 2.605

Thus, 2.605 is the 99th percentile of the MOE distribution.
The Beta Distribution: All families of continuous distributions listed except for the uniform
distribution have positive density over an infinite interval (though typically the density function
decreases rapidly to zero beyond a few standard deviations from the mean). The beta distribution
provides positive density only for X in the interval of finite length.

A random variable X is said to have a beta distribution with parameters  ,  (both positive), A
and B if the pdf of X is
 1  1
 1 (   )  x       x 
      x
f ( x;  ,  , , )      ( )  (  )          
 0 otherwise


The case A = 9, B = 1 gives the standard beta distribution. The mean variance of X are

 (  ) 2 
    (  )  2 
  (   ) 2 (    1)

Example 7
Suppose that in constructing a single-family house, the time X (in days) necessary for laying the
foundation has a beta distribution with A = 2, B = 5,   2 and   3 . Then  /(   )  .4 , so
E(X) = 2 + (3)(.4) = 3.2. For these values of  and  , the pdf of X is a simple polynomial
function. The probability that it takes at most 3 days to lay the foundation is
3 1 4!  x  2  5  x 
(   3)      dx
2 3 1!2!  3  3 

4 3 4 11 11

27 2
( x  2)(5  x) 2 dx   
27 4 27
 0.407

The standard beta distribution is commonly used to model variation in the proportion or
percentage of a quantity occurring in different samples, such as the proportion of a 24-hour day
that an individual is asleep or the proportion of a certain element in a chemical compound.

The pdf of all continuous distributions share the properties:

(a) f ( x)  0

(b) 

f ( x)dx  1 .
When F(x) is not continuous, its finite derivative does not exist and –in the classical sense-the
pdf does not exist. If F(x) is continuous for every x and its derivative exists everywhere except at
a countable points, then we say that X is a continuous random variable. The expected value, if it
exists, of a real random variable X with pdf f  (x) is defined as

(  )   x f  ( x)dx .


The Normal Approximation to the Binomial Distribution: Let X be a binomial random variable
based on n trials with success probability p. then if the binomial probability histogram is not too
skewed, X has approximately a normal distribution with   np and   npq . In particular,
for x = a possible value of X,
 area under the normal curve
(   x)  ( x; n, p)   

 to the left of x  .5 
 x  .5  np 
  
 
 npq 
In practice, the approximate is adequate provided that both np  10 and nq  10 . If either
np  10 or nq  10 , the binomial distribution is too skewed for the (symmetric) normal curve to
give accurate approximation. For example, a binomial probability histogram for the binomial
distribution with n = 20, p = .6 [so   20(.6)  12 and   20(.6)(.4)  2.19] . A normal curve
with mean value and standard deviation equal to the corresponding values for the binomial
distribution has been superimposed on the probability histogram.

.25 normal curve,   12 ,   2.19

.20
.15

.10
.05

0 2 4 6 8 10 14 16 18 20
Fig. 2.7: Binomial probability histogram for n = 20, p = .6 with normal approximation curve
superimposed.
Example 8
Suppose that 10% of all steel shafts produced by a certain process are nonconforming but can be
reworked (rather than having to be scraped). Consider a random sample of 200 shafts, and let X
denote the number among these that are non-conforming and can be reworked. What is the
(approximate) probability that
(a) At most 30?
(b) Less than 30?
(c) Between 15 and 25 (inclusive)?
Solution
From the data: p = .10,   np  200(.10)  20 and   200(.10)(.90)  4.24 . Since
np  20  10 and nq  200(.9)  180  10 , the approximation can be applied.
 30  .5  20 
(a) (   30 )  (30; 200 , .10 )   
 4.24 
 (2.48) = 0.9934
 29  .5  20 
(b) (   30 )  (   29 )  (29; 200 , .10 )   
 4.24 
 (2.24) = 0.9875
(c) (15    25 )  (25; 200 , .10 )  (14; 200 , .10 )

 25 .5  20   14 .5  20 
     
 4.24   4.24 
 (1.30)  (1.30)
= 0.9032 – 0.0968
= 0.8064
The probability (15    25) is being approximated by the area under the normal curve
between 14.5 and 25.5 – the continuity correction is added in both the upper and lower limits.

When the objective of our investigation is to make an inference about a population proportion p,
interest will focus on the sample proportion of successes X/n rather than on X itself. Because this
proportion is just X multiplied by the constant 1/n, it will also have approximately a normal
distribution (mean   p and standard deviation   pq / n provided that both np  10 and
nq  10 . This normal approximation is the basis for several inferential procedures.
ASSIGNMENT
1. A theoretical justification based on a certain material failure mechanism underlies the
assumption that ductile strength X of a material has a lognormal distribution. Suppose the
parameter are   5 and   .1 .
(a) Compute E(X) and V(X).
(b) Compute (  125) .
(c) Compute (110    125) .
(d) What is the value of median ductile strength?
(e) If ten different samples of an alloy steel of this type were subjected to a strength test,
how
many would you expect to have strength of at least 125?
(f) If the smallest 5% of strength values were unacceptable, what would the minimum
acceptable strength be?

2. Suppose only 40% of all drivers in a certain city regularly wear a seat belt. A random
sample of 500 drivers is selected. What is the probability that
(a) Between 180 and 230 (inclusive) of the drivers in the sample regularly wear a seat belt?
(b) Fewer than 175 of those in the sample regularly wear a seat belt? Fewer than 150?