by

Prof. Mahmoud Gabr

Dr Nesrin Nabil Dr Asmaa El kady

Department of Mathematics and Computer Science

Faculty of Science - Alexandria University

-0-
 Contents
Subject Page
Review
Chapter 1
SAMPLING DISTRIBUTIONS 1
1.1 Introduction 1
1.2 Stochastic Convergence 2
1.3 law
The of large numbers
Distribution of X 3
1.4 The Chi-Squared
Central Distribution
Limit Theorem 6
1.5 The t- Distribution 10
1.6 The F- Distribution 13
EXERCISES 15
Chapter 2
POINT ESTIMATION 17
2.1 Introduction 17
2.2 Some Methods of Estimation 18
(I) The Method of Moments 18
(II) The Method of Maximum Likelihood 20
2.3 Properties of Estimators 24
a- Unbiased Estimators 24
b- Minimum variance unbiased Estimator (MVUE) 27
c- Efficiency 27
d- Consistency 28
EXERCISES 29
Chapter 3
INTERVAL ESTIMATION 46
3.1 Introduction 46
3.2 Confidence Interval for μ (case (i): σ2 is known) 47
Maximum Error and Sample Size 50
case (ii): σ2 is unknown and n < 30 51
3.3 Confidence Interval for the Difference of Means of Two 52
Populations
Case 1:(Use(μof the Normal Distribution)
1-μ2)
52
Case 2: (Use of the t-Distribution) 54
3.4 Confidence Interval For The Proportion p 55
EXERCISES 58

-1-
Chapter 4
TESTS OF HYPOTHESES 62
4.1 Basic Definitions 62
4.2 Type I and Type II Errors 62
4.3 One-Side and Two-sided Tests 63
4.4 Tests Concerning The population Mean μ(case (i): σ2 is known) 64
The P-value 67
case (ii): σ2 is unknown and n < 30 68
4.5 Tests Concerning Two Populations 70
Case 1:(Use of the Normal Distribution) 70
Case 2: (Use of the t-Distribution) 72
Paired Comparisons 73
4.6 Tests Concerning The Population Variance 75
4.7 Tests For The Equality of Two Variances 76
EXERCISES 77
Chapter 5
ANALYSIS OF VARIANCE (ANOVA) 89
5.1 One Way (One Factor) ANOVA 89
5.2 Two Way (Two Factor) ANOVA 94
EXERCISES 100
Chapter 6
SIMPLE LINEAR REGRESSION
6.1 Introduction 80
6.2 Least Squares Estimation of the Parameters 3
6.3 Point estimate of the mean response 4
6.4 Residuals 5
6.5 Estimation of σ2 8
6.7 Interval Estimation of the Mean Response 89
6.8 Prediction of New Observation 90
6.9 ANOVA Approach to Regression Analysis 92
6.10 Coefficient of Determination 94
6.11 Correlation Coefficient 95
Adjusted Coefficient of Determination 96
EXERCISES 97

-2-
 Chapter 1
SAMPLING DISTRIBUTIONS
1.1 Introduction
Statistics concerns itself mainly with conclusions and predictions resulting
from chance outcomes that occur carefully experiments or investigations. In the finite
case, these chance outcomes constitute a subset or sample of measurements or
observations from a larger set of values called the population. In the continuous
case they are usually values of i.i.d (independent identically distributed) random
variables, whose distribution we refer to as the population distribution, or the
infinite population sampled. The word “infinite” implies that there is, logically
speaking, no limit to the number of values we could observe.

Definition 1.1 Population

The totality of elements which are under discussion or investigation and about
which information is desired will be called the target population.

Definition 1.2 Random sample

If X1, X2, ..., Xn are i.i.d. r.v.’s, we say that they constitute a random sample
(abbreviated by R.S.) from the infinite population given by their common distribution
An important part of the definition of a R.S. is the meaning of the r.v.’s X1, X2,
..., Xn. The r.v. Xi is a representation for the numerical value that the ith item (or
element) sampled will be assumed. After the sample is observed, the actual values of
X1, X2, ..., Xn are known, we denote these observed values by x1, x2, ..., xn.

One of central problems in statistics is the following:

If it is desired to study a population which has a known density function, but it contains
some unknown parameters. For example, suppose that we have a population, which has
the normal distribution, but the parameters μ and σ2 are unknown.
The procedure is to take a random sample (R.S.) X1, X2, ..., Xn of size n from this
population and let the value of some function represent or estimate the unknown
parameter. This function is called a statistic. Since many random samples are possible
from the same population, we would expect every statistic to vary somewhat from
sample to sample. Hence a statistic is a random variable, and as such it must have a

-3-
probability distribution.

Definition 1.3
If X1, X2, ..., Xn are i.i.d. r.v.'s, we say that they constitute a random sample from
the infinite population given by their common distribution.
If f (x1, x2, ..., xn ) is the joint p.m.f. (or p.d.f.) of such a set of r.v.'s , we can write
n
f( x1, x2, ..., xn ) = f(x1) f(x2) ... f(xn ) =  f (x )
i
i =1
where f(xi) is the common p.m.f. (or p.d.f.) of each Xi (or of the population sampled).

Definition 1.4
A statistic is a function of observable r.v.'s, which is itself an observable r.v. and
does not contain any unknown parameter.
For example, if X1, X2, ..., Xn is a r.s., then the sample mean
1 n
X =  Xi
n i=1
and the sample variance
1 n

2 2
S = ( Xi - X )
n - 1 i=1
are statistics.
Since statistics are r.v.'s, their values will vary from sample to sample, and it is
customary to refer to their distributions as sampling distributions. Note that;
1  n  1 n
 n  i=1  n i=1
1
E X = E   X i    E  X i   n  
n , and

 n  1 2
 
n
1 1
Var X =
n 2
Var   Xi   2
 i=1  n
 Var  Xi 
i=1 n2
n  2

n

1.2 Stochastic Convergence

Sometimes the distribution of a r.v. (perhaps a statistic) depends upon a +ve
integer n. Clearly, the distribution function (CDF) F of that r.v. will also depends upon
n and denote it by Fn. We now define a limiting distribution of a r.v. whose distribution
depends upon n.
Definition 1.5
Let the CDF Fn(x) of the r.v. Xn depends upon n. If
𝑙𝑖𝑚 𝐹𝑛 (𝑥) = 𝐹(𝑥)
𝑛→∞
for every point y at which F(y) is continuous where F(x) is a CDF, then the r.v. Xn is

-4-
said to have a limiting distribution with CDF F(x).

Theorem 1.1 (Weak law of large numbers)

Let (Xn), n=1,2,... be a sequence of independent r.v's such that all have the same
expectation, E(Xn) = μ and the same variance Var(Xn) = σ2 then we say that the r.v. Xn
converges stochastically (or in probability) to the constant μ iff, for every ε > 0 we
have
𝐥𝐢𝐦𝐏(|𝐗 𝐧 − 𝛍| < 𝛆) = 𝟏
𝐧→∞
and we may write
p
Xn  
We should like to point out a simple but useful fact. Clearly
𝐏(|𝐗 𝐧 − 𝛍| < 𝛆) + 𝐏(|𝐗 𝐧 − 𝛍| ≥ 𝛆) = 𝟏
Thus, the limit of 𝐏(|𝐗𝐧 − 𝛍| < 𝛆) is equal to 1 when and only when
𝐥𝐢𝐦𝐏(|𝐗 𝐧 − 𝛍| ≥ 𝛆) = 𝟎
𝐧→∞
That is, the last limit is also a necessary and sufficient condition for the stochastic
convergence of the r.v. Xn to μ.
In addition, a very important result on law of large numbers, is:
Let Xn be the mean of a random sample of size n, then;
p
Xn    lim P(| X n   | )  1
n 

1.3 The Distribution of 𝑿

Theorem 1.2
If X1, X2, ..., Xn constitute a R.S. from a normal population with mean μ and
X-
variance σ2, then X ~ N(μ, σ2/n) i.e. Z = ~ N ( 0 , 1 ).
/ n
The Proof is omitted.
From this theorem we also conclude that the mean and variance of X are given by

2
 X = E( X)   , and  2 = var ( X )  (1.1)
X n
Theorem 1. 3 (Central Limit Theorem)
If random samples of size n are drawn from any infinite population with mean μ
and variance σ2, the limiting distribution of
X-
Z=
/ n

-5-
 as n→∞ is the standard normal distribution N(0,1).
The Proof is omitted.
The Normal approximation in the central limit theorem will be good if n  30
regardless of the shape of the population. If the population variance σ2 is unknown, the
central limit theorem still valid when we replace σ2 by the sample variance S2, i.e. for
large n enough, we have
X-
Z= ~ N ( 0 ,1)
S/ n
It is interested to note that when the population we are sampling is normal, the
distribution of X is a normal distribution (see theorem 1.1) regardless of the size of n.

Example 1.1
Certain tubes manufactured by a company have a mean lifetime of 900 hrs and
standard deviation of 50 hrs. Find the probability that a random sample of 64 tubes
taken from the group will have a mean lifetime between 895 and 910 hrs.
Solution
Here we have μ = 900, σ = 50. Let X denotes the sample mean lifetime of the
tubes and since n = 64 is large enough, then by the central limit theorem
X-
Z= ~ N ( 0 ,1 )
/ n
Thus
895 - 900 X -  910 - 900
P (895 < X < 910) = P ( < < ) = P(-8.0 < Z < 1.6)
50 / 8 / n 50 / 8
= (1.6) - (-0.8) = (1.6) - 1 + (0.8) = 0.733

Case of two populations

If two independent random samples of sizes n1 and n2 are drawn from any two
populations with means μ1 and μ2, and variances 12 and  22 , respectively, the sampling
distribution of X1 - X 2 will be approximately distributed with mean and variance given
by
12  22
 X1  X2 =  X1 -  X2 = 1 -  2 , and X  X  X + X
2 2 2
= +
1 2 1 2 n1 n 2
Hence,

-6-
 ( X1  X 2 ) - (1 - 2 )
Z= ~ N(0,1)
 12  22
+
n1 n 2
If both n1 and n2 are greater than or equal to 30, the normal approximation for the
distribution of X1 - X 2 will be good regardless of the shapes of the two populations.
Similarly, if the variances 12 and  22 are unknown, the central limit theorem still valid
with using the sample variances S12 and S22 instead of 12 and  22 . Therefore
( - ) - ( 1 - 2 )
Z = X1 X22 ~ N ( 0 ,1)
S 1 S 22

n1 n 2
Example 1.2
The electric light bulbs of manufacturer A have a mean lifetime of 1400 hrs with
a standard deviation of 200 hrs, while those of manufacturer B have a mean lifetime of
1200 hours with a standard deviation of 100 hours. If random samples of 125 bulbs of
each brand are tested, what is the probability that the brand A bulbs will have a mean
lifetime which is at least 160 hours more than the brand B bulbs?
Solution
Let X and X denote the mean lifetimes of samples A and B respectively.
A B
Then the variable
( X A - XB ) - 
X A - XB ( X A - XB ) - 200
Z= = ~ N ( 0 ,1)
X - X 20
A B
The required probability is then, given by
 160 - 200 
P(XA  XB  160 ) = P  Z  = P(Z  - 2.0) = 1 - (-2.0 )  (2.0) = 0.977
 20 

1.4 The Chi-Squared Distribution

If Z1, Z2,...,Zv are independent r.v.'s having standard normal distribution N(0,1),
then the r.v.
U = 𝒁𝟐𝟏 + 𝒁𝟐𝟐 +. . . +𝒁𝟐𝝂 (1.2)
2
has the so called Chi-Squared Distribution (often denoted by χ distribution) with ν
degrees of freedom (d.f.) and it has the following properties;
1- The mean and variance of the   distribution are
2

μ = ν and σ2 = 2 ν

-7-
 df=2

df=7

df=12

Fig. 1.1𝝌𝟐𝝂 distribution curves for various values of ν

3- If U1, U2,...,Uk are independent r.v.'s having chi-squared distributions with v1,
v2,...,vk d.f., then
k
Y = Ui
i =1

has the chi-squared distribution with ν = ν1 + ν2 +...+ νk d.f.

4- The percentage points of the   distribution have been extensively tabulated.
2

Define   , as the percentage point or value of the chi-square r.v. U with ν d.f.
2

such that

2
P(U    ,)=  f2 (u) du = 
2
 ,
This probability is shown as the shaded area in Fig.1.2
Note that if X1, X2,...,Xn constitute a R.S. from a normal population with mean μ
X -
and variance σ2, then Z i = i ~ N ( 0 , 1 ) and therefore the variable

 X -
2
n n
U   Zi =   i
2
 .
i1 i 1   
Has the χ2 distribution with ν = n d.f.

-8-
 αα

0
2 , 

Fig. 1.2 Percentages point of the chi-squared distribution

Using MINITAB
Suppose we want to find  .05,12 , then its CDF is F(  .05,12 ) = 1- α = 0.95. Now
2 2

press Calc → Probabilty Distributions → Chi-Square then click on "inverse

cumulative distribution" and write 12 for the degrees of freedom and 0.95 for "input
constant" as in the following Figures.

Click on " "ok" we obtain:

-9-
Chi-Square with 12 DF

P( X <= x ) x
0.95 21.0261

i.e.  .05,12 = 21.02.

2

The Distribution of S2
Theorem 1.4
If X1 , X2 , ..., Xn are a random sample from N(μ,σ2), then
1- X and the terms X i  X ; i=1,..., n are independent,
2- X and S2 are independent.
The proof is omitted.

Theorem 1.5
If X and S2 are the mean and variance of a r.s. of size n from a population
having N(μ,σ2), then
1- X and S2 are independent;
( n - 1 ) S2
2- the r.v. U = has the chi-squared distribution with ν = n-1 d.f.

2

Proof
First note that by adding and subtracting X and then expanding, we obtain the
relationship

U=
( n - 1 ) S2 n X i -  

2

=
n
Xi - X
2


n X -  
2

(6.3)
   
2 2 2 2
i =1 i =1
= U1 + U2
Since
 
2
Xi - 
2
n n
 X   n
U =  i   Zi ~ n
 2 2

i =1 
2
i 1    i 1

and
X- (X - )2 n (X - )2
X ~ N(μ, σ /n)  Z =
2
~N(0,1)  Z =
2
  U 2 ~ 2
  2
 2 1

n n
Thus,
U ~ 2n and U 2 ~ 2
1
Therefore by property (3) of the chi-square distribution we have
U1  U  U2 ~ 2n1
Corollary
Since the mean and variance of the  n1 are respectively, (n-1) and 2(n-1), it
2

follows that

- 10 -
  ( n - 1 ) S2 
E   n  1  ES 2   2
  
2

and
 ( n - 1 ) S2  2 4
Var 

2   2(n  1)   
var S 2 
n 1
 

1.5 The t- Distribution

We know that if X1, X2,...,Xn are a R.S. from a normal population with mean μ
and variance σ2, then X ~ N(μ, σ2/n) i.e.
X-
Z= ~ N( 0, 1 )
/ n
Most of the time we are not fortunate enough to know the variance of the population
from which we select our random samples. For samples of size n < 30, a good estimate
of σ2 is provided by calculating S2. What then happens to the distribution of the Z
values in the central limit theorem if we replace σ2 by S2? As long as S2 is a good
estimate of σ2 and does not vary much from sample to sample, which is usually the case
for n  30, the values
X-
S/ n
are still approximately distributed as a standard normal variable, and central limit
theorem is valid. If the sample size is small (n<30), the values of S2 fluctuate
considerably from sample to sample and the distribution of the values ( X - )/(S / n )
is no longer a standard normal distribution.
Thus the theory which follows leads to the exact distribution of
X-
T=
S/ n
for r.s.'s from normal populations.

Theorem 1.6
Let Z and U be two r.v.'s with
1- Z ~ N(0,1),
2- U ~  r
2

3- Z and U are independent.

Then the distribution of

- 11 -
 Z
T=
U /r
is called the t-distribution with r degrees of freedom and its p.d.f. is given by
 r 
  + 1 -
r +1

 2 
  1 + x  2
2

f (x )= - <x< 
 r   r 
r   
 2
The t-distribution is also known as the student-t distribution. The t-distribution is
similar to the N(0,1) distribution in that they both are symmetric about a mean of zero.
Both distributions are bell shaped but the t-distribution is more variable. The areas
under the curve have been tabulated in sufficient detail to meet the requirements of
most problems. The distribution of t is similar to the distribution of Z, in that they both
are symmetric about a mean of zero. Both distributions are bell shaped but the t
distribution is more variable. The distribution of t differs from that of Z in that the t-
distribution depends on the degrees of freedom r and is always greater than 1. Only
when r   (or r large > 30) will the two distributions become the same. In Figure 1,
we show the relationship between a standard normal distribution (r = ), and t
distribution with 4 and 8 degrees of freedom

 =  (Normal)

 =10

 =4

Fig. 1.3 t distribution curves for ν = 4, 10 and .

Theorem 1.7
If X and S2 are the mean and variance, respectively, of a random sample of size
n taken from a population that is normally distributed with mean μ and unknown
variance σ2. Then the variable

- 12 -
 X-
T=
S/ n
has a t-distribution with ν = n-1 degrees of freedom.
Proof
If X1 , X2 , ..., Xn are a random sample from N(μ,σ2), then
X-
1- X ~ N(μ, σ2/n) i.e. Z = ~ N(0, 1)
/ n
( n - 1 ) S2
2- U = ~ 2n1

2

3- X and S2 are independent, thus also Z and U are independent.

Therefore by theorem 1.7, we have
X - 
Z / n X-
T= = = ~t
U /r ( n - 1 ) S2 S/ n n1
/ (n - 1)

2

For a t-distribution with (n-1) degrees of freedom the symbol tα denotes the t-value
leaving area of α to the right. tα is the upper α- point of the t-distribution with (n-1)
degrees of freedom (see Fig. 1.4). The t-table is arranged to give the values tα for
several frequently used values of α and different values of ν = (n-1).
Since the t-distribution is symmetrical about the value t = 0, the lower points can be
obtained form the, upper points. The relationship between the lower and upper points is
t1 = - t 

Fig. 1.4 α- point of the t-distribution with (n-1) d.f.

For example; if ν = n-1 = 5, then from the t-table we have

t0.025 = 2.57 therefore t0.975 = -2.57
Using MINITAB

- 13 -
 Suppose we want to find t0.025, 5 , then its CDF is F(t0.025, 5) = 1- α/2 = 0.975. Now
press Calc → Probabilty Distributions → t then click on "inverse cumulative
distribution" and write 5 for the degrees of freedom and 0.975 for "input constant".
Click on "ok" we obtain:
Student's t distribution with 5 DF

P( X <= x ) x
0.975 2.57058
i.e. t0.025, 5 = 2.57058.

1.6 The F- Distribution

Another derived distribution of great importance in statistics is called the F
distribution.
Theorem 1.8
Let U1 and U2 be two r.v.'s with
1- U1 ~  r ,
2
1

2- U2 ~ 
2
r2

3- U1 and U2 are independent.

Then the distribution of the r.v.
U /r
F= 1 1
U 2 / r2
is called the F distribution with ( r1 , r2 ) degrees of freedom and its p.d.f. is given by
 r1 + r2 
   r1 / 2 r1
1
-
r1 + r 2

f (x )=  2   r 1 
 x2

 1 + r 1 x


2
- <x< 
   r  r
  r1    2  2   
r2
2   2 

Corollary
Let X1, X2,...,Xn and Y1, Y2,...,Yn be independent random samples from
populations with respective distributions Xi ~ N( μ1, σ12 ) and Yj ~ N( μ2, σ22 ). If r1 =
n1-1 and r2 = n2-1, then
( n1 - 1 ) S12 ( n 2 - 1 ) S 22
U1 = ~  n 1
2
and U2 = ~ 2n 1
12
1 22
2

so that
U1 / (n1  1) S1 2
2 2
F= = 2 2 ~ Fn 1 , n 1
U2 / (n 2  1) S2 1 1 2

<+><+><+><+><+><+><+><+><+>

- 14 -
 EXERCISES

[1] Certain tubes manufactured by a company have a mean lifetime of 800 hours and a
standard deviation of 40 hours. Find the probability that a random sample of 36
tubes taken from the group will have a mean lifetime.
a- Between 790 and 810 hours, b- More than 815 hours.

[2] A and B manufacture two types of cables, having mean breaking strengths of 4000
and 4500 pounds and standard deviations of 300 and 200 pounds respectively. If
100 cables of brand A and 50 cables of brand B are tested, what is the probability
that the mean breaking strength of B will be
a- At least 600 pounds more than A, b- At least 450 pounds more than
A.

[3] Find a- P(-t0.005 < t <t0.01) b- Find P(t >-t0.025).

[4] Given a random sample of size 24 from a normal distribution, find, K such that
a- P(-2.069 < t < K) = 0.965 b- P(K < t < 2.807) = 0.095.
c- P(-K < t < K ) = 0.90.

[5] Consider the four independent random variables X, Y, U and V such that X ~
N(0,16), Y ~ N(5,4) , U ~ χ2(4) and V ~ χ2(16).
State the distribution of each of the following variables
2 2
X ( Y - 5) 𝐗 𝟒𝐔
a- + b- c- d- X+2Y e- 2X-Y
16 4 √𝐕 𝐕

[6] If X1, X2, ..., Xn are i.i.d. N(0,σ2), state the distribution of each of the following
variables:
n
a- U = 3 X1 - 5 X2 + 8 b- V =  X i
i =1

 
2
n
2 X12  Xi
c- W = X
i =1
i
n 2
d- Y = 2
X2 + X3
2
e- Y =
 X2i

[7] If X1, X2, ..., Xn are i.i.d. N(0,σ2), state the distribution of each of the following
variables:
2 X12  Xi
a- Y = 5 X1 -7 X2 +2 b- Y = 2 c- Y =
 Xi2
2
X2 + X3
[8] Suppose that X1, X2, X3 and X4 are i.i.d. N(0,σ2), then the distribution of the
random variable Y = X1  X2 is
X 23  X42
2
a. χ (2) b. t(2) c. F(2,2) d. None of the above.

- 15 -
[9] Consider the three independent random variables X, U and V such that X~
N(0,1), U ~ χ2(4) and V ~ χ2(16). Find the distribution of W = X2 +U+V.

[10] Let 𝐗 and 𝐘 be sample means of two independent random samples of sizes 10
and 20 from N(4,9) and N(5,16) respectively. Find mean, variance and
distribution of Z = X - 2Y + 3 .

[11] Show that if X has a t distribution with v d.f., then Y=X2 has an F distribution
with ν 1=1 and ν 2 = ν d.f.

[12] Circle the best answer from each of the following multiple-choice questions:
Let X ~ N(1,16), Y ~ N(0,4) and U ~ χ2 (15) be three independent r.v's.
a- The distribution of 2X-3Y+5 is
i. N(7,28) ii. N(7,100) iii. N(2,105) iv. None of the above.

b- One of the following r.v.'s has F(16,1)

U + Y2 / 4 U + Y2 / 4 ( U + Y2 / 4 ) / 16
i. 2
ii. 2
iii. 2
iv. None of the above.
( X - 1 ) / 16 ( X-1) ( X-1)

X1
c- The distribution of is
2
Z +U
i. t(3) ii. t(15) iii. t(16) iv. None of the above.

<+><+><+><+><+><+><+><+><+>

- 16 -