3/18/2024

2.0. Introduction
• Populations are described by their probability
distributions and parameters…
– For quantitative populations, the location and
Chapter 2: Statistical Estimation shape are described by m and s.
– For a binomial populations, the location and
By: Meseret A.
shape are determined by p.
Debre Markose University
College of Business and Economics
Department of Management • If the values of parameters are unknown, we make
December, 2016 inferences about them using sample information.
1 2

• Therefore, statistical inference is the process of

using sample results to draw conclusions about the 2.1. The Concept Of Estimation
parameters of a population. • Estimation is a process whereby we select a
• Generally, statistical inference has two basic tasks random sample from a population and use an
• statistical estimation estimator (a sample statistic) to estimate a
• Statistical hypothesis testing population parameter

Estimation: is estimating or predicting the value of the

population parameter, Important Terms
E.g: what is (are) the most likely values of m and P – Estimator: is a formula (rule) or a strategy that is used
to estimate a population parameter

Hypothesis testing: is deciding about the value of a

parameter based on some preconceived idea… – Estimate: is a number computed by using the data from
3 the sample 4

1
 3/18/2024

The Objective of Estimation Process of Estimation

• To determine the approximate value of a i. The relevant population is carefully defined,
population parameter ii. The parameter of interest is (are) identified,

• To determine size of a sample (n), which will be iii. A representative sample is selected,
enough for reliable estimation iv. Data are collected from all members of a sample,

• To determine some methods of statistical units v. Sample statistic (s) that serve (s) as estimate (s) is (are)
computed from the data collected
sampling from a population
vi. The population parameter (s) is (are)estimated using
sample statistic (s) (estimates)
5 6

e.g. point estimate of parameter θ generally is θ

2.2. Types of Estimation or Estθ = θ like Est µ is 𝒙 or Est P is 𝒑
• there are two types of estimations (estimators) Where,
– Point estimation (Estimator) θ is any population parameter
– Interval estimation (Estimator) θ is the corresponding sample statistic
N.B: in every case the appropriate point estimator of a
2.2.1. Point estimator (Estimation) parameter is simply the corresponding sample statistic:
 a point estimator draws inference about a population by
estimating the value of an unknown parameter using a
single value or a point…
 a point estimation deals with the task of selecting a specific
sample value as an estimate for population parameter…
a point estimate is a specific value of a sample statistic that
is used to estimate the unknown population parameter 7 8

2
 3/18/2024

Properties (qualities) of a best Estimators a) Unbiasedness

• These properties refers to the desirable qualities of • An unbiased estimator of a population parameter
estimators is an estimator whose expected value is equal to
the value of a parameter
i.e., E(θ) = θ
• Therefore, the best estimator satisfies (meets) • unbiased estimator of θ is sample characteristic,
following conditions: which satisfy condition :
a) Unbiasedness lim θ = θ
𝑛→∞
b) Consistency
b) Consistency
c) Efficiency • An unbiased estimator is said to be consistent if
d) Sufficiency the difference between the estimator and the
parameter grows smaller as the sample size grows
9
larger 10

c) Sufficiency • However, there are three drawbacks to using point

• An estimator is said to be sufficient for estimating a estimators :
population parameter, if it contains all the information a) It is virtually certain that the estimate will be wrong.
in the samples about the parameter. i.e., P(θ = θ)= 0
lim σθ = 0
𝑛→∞
b) No information about how close the estimator is to
the parameter…

d) Efficiency c) Point estimates don’t have a capacity to reflect the

• If we have two unbiased estimator of the same effects of large sample size…
parameter, the point estimator with small variance is
said to have greater Efficiency
 As a consequence, we use the second method of
• Let θ1 and θ2 be estimator of θ. Then, estimating a population parameter – Interval
Var(θ1) < var(θ2) , so θ1 is efficient estimator of θ Estimator …
11 12

3
 3/18/2024

2.3. Interval Estimator • The general form of an interval estimate is as follows:

Point estimate ± Margin of error
• An interval estimator draws inferences about a population by
estimating the value of an unknown parameter using an  Where, Margin of error (ME) is the maximum error of
interval. estimation at a given confidence level
• More generally, we can define confidence interval for any
• Thus, it is a rule for determining (based on sample information) a required probability content less than 1: Then, A and B are
range or interval in which the parameter is likely to fall…. such that
P(a<θ<b) = 1-α
N.B: an interval estimator (A.K.A confidence interval estimator): • Where,
• is stated in terms of Probability….  α is error probability-the risk of estimation taking any value
between 0 and 1; and to be equally distributed to both sides of
• gives information about Closeness to Unknown Population the normal curve…then α/2 for each side
Parameter  1- α is the probability content or level of confidence or
• Thus, the corresponding estimate is called an interval estimate … confidence coefficient for the interval- measure of the
confidence we have that the interval does indeed contain the
• The corresponding process is called an interval estimation parameter of interest.
which establish an interval consisting a lower limit and an upper N.B: a confidence interval calculated in this way is called a (1-α)
limit in which the population parameter is expected to fall 13 100% confidence interval for θ 14

• Thus, the interval estimate of θ at 1-α CL is: • And it will have the general form given by:
P(a≤θ≤b)= 1-α θ ± Margin of Error
Where;
where: a and b are the lower and upper limits of interval – Margin of error is maximum error of estimation or the
difference between Ӫ and θ
α is the risk of estimation
• And Margin of error (ME)=
f(θ)
– Where;
• δӪ is standard error of Ӫ
• Zα/2 is the critical value of Z or the z value that cuts off a right-tail
α/2 α/2 area of α/2 under standard normal curve-at 1-α confidence level
• Since Ӫ is a point estimator of θ, the confidence
a b
interval of θ at 1-α CL is:
Ӫ±
15 16

4
 3/18/2024

Process of constructing CI
i. Compute sample statistic θ v. Multiply standard error of the statistic by tabled
value ( compute margin of error)
ii. Compute the standard error of the statistic δ θ
iii. Make a decision about level of confidence desired
vi. Form interval by adding and subtracting the
(usually, 90%, 95%, 98%, and 99%) calculated value (margin of error) to and from the
iv. Find the table value of the desired CL statistic
1-𝛼 Commonly
𝛼 used CL and
𝛼/2 𝐳 Zα/2 𝛂/𝟐
Ӫ±
0.90 0.1 0.05 𝐳𝟎.𝟎𝟓 = 𝟏. 𝟔𝟒𝟓
• Then, 1 - a of all the values of obtained in
repeated sampling from a distribution, construct an
0.95 0.05 0.025 𝐳𝟎.𝟎𝟐𝟓 = 1.96
interval:
0.98 0.02 0.01 𝐳𝟎.𝟎𝟏 = 𝟐. 𝟑𝟑
σ σ
0.99 0.01 0.005 𝐳𝟎.𝟎𝟎𝟓 = 𝟐. 𝟓𝟕𝟓 θ − 𝑍 α 2 𝑛 , θ + 𝑍 α 2 𝑛 that includes (covers)
17
the expected value of the population parameter. 18

2.3.1. interval estimator of m when δ is known [Link] Interval Estimator of m when δ is unknown
• Assume • If δ is unknown, and n<30, we have to use t-distribution
the sampling distribution of the mean 𝑥 is normal or with (n-1) degrees of freedom given by:
δ is known t(n-1)=
the observed sample mean is 𝑥 , then (1-α)100% CI for m is
given by
s s N.B: df is number of observations that are free to vary after
P( x  z a 2  m  x  za 2 )  1 a sample mean has been calculated…
n n • Assume:
N.B: in this form µ is in the center of the interval created by The sampling distribution 𝑥 has normal distribution
adding and subtracting Zα/2 standard errors to and from 𝑥 .
If δ is unknown, replace δ with the sample standard
This equation says that, with repeated sampling from this deviation s,
population, the proportion of values of 𝑥 for which the
interval includes μ is equal to 1 − α. • Then (1-α)100% CI for m when δ is unknown is given by:
This form of probability statement is very useful because it is 𝑠 𝑠
the confidence interval estimator of μ. 𝑥 − 𝑡 (𝑛 − 1 ) α 2 < µ < 𝑥 + 𝑡 (𝑛 − 1) α 2
19
𝑛 𝑛 20

5
 3/18/2024

• However, for large degrees of freedom, the t-

distribution is approximated by Z-distribution given • How to Interpreting the Confidence Interval
Estimate?
by:  to interpret the confidence interval estimate properly,
𝒙−µ remember that the confidence interval estimator was
Z= 𝒔 √𝒏 derived from the sampling distribution of the sample
mean.
• Where,  the sampling distribution was used to make probability
s is sample standard deviation statements about the sample mean
 although the form has changed, the confidence interval
𝒔 √𝒏 is estimated standard error of 𝒙 estimator is also a probability statement about the sample
mean. It states that there is 1 − α probability that the
• Therefore, a larger sample (1-α)100% CI for μ is sample mean will be𝒔 equal to a value𝒔such that the
given by: interval 𝒙 − 𝒁 𝜶 𝟐 𝒕𝒐 𝒙 + 𝒁 𝜶 𝟐 will include the
𝒏 𝒏
𝑠 𝑠 population mean.
𝑥 −𝑍α 2 < µ < 𝑥+𝑍α 2  once the sample mean is computed, the interval acts as
𝑛 𝑛 the lower and upper limits of the interval estimate of the
population mean
21 22

2.3.2. Interval Estimator of Population Proportion 2.3.3 Confidence Interval for (μ1 – μ2) (for large
(Large Sample Size) samples and known variances)
• Assume • Let:
𝒑 is the observed proportion of “success” in random – Ẍ1 represents the observed sample mean of random
sample of size n sample size of n1 with population mean of μ1
Sampling distribution of 𝒑 is normal – Ẍ2 represents the observed sample mean of random
sample of size n2 with population mean of μ2
• Then (1-α)100% CI for population proportion P is
– the sample sizes are large and population variances are known
given by:
• Then, (1-α)100% confidence interval for (μ1 – μ2) is
𝒑(𝟏 − 𝒑) 𝒑(𝟏 − 𝒑) given by:
𝒑−𝑍α 2 <p<𝒑+𝑍α 2
𝒏 𝒏

23 24

6
 3/18/2024

2.3.3 Confidence Interval for (μ1 – μ2) (for small • It therefore follows that random variable
samples and unknown variances) 𝒁=
Ẍ𝟏−Ẍ𝟐 −(𝛍𝟏 – 𝛍𝟐)
has a standard normal
• Suppose again that we 𝐧𝟏+𝐧𝟐
𝜹𝟐 ( 𝒏𝟏𝒏𝟐 )
– have independent random samples n1 and n2 distribution……
observations from normal populations with means μ1 • Since this variance is common to the two populations,
and μ2, the two sets of sample information can be pooled
– populations have common unknown variance (δ2), together to estimate it. The estimator used is:
n1−1 S12 +(n2−1)S22
– Inference about population means is based on the S=
(𝑛1+ 𝑛2−2)
difference Ẍ1 − Ẍ2 between the two sample means. 2
• Replacing the unknown 𝛿 gives the random variable
• Therefore, This random variable has a normal Ẍ𝟏 − Ẍ𝟐 − (𝛍𝟏 – 𝛍𝟐)
distribution with mean (μ1 – μ2) and variance 𝒕=
𝒗𝒂𝒓 Ẍ𝟏 − Ẍ𝟐 = 𝐯𝐚𝐫 Ẍ𝟏 + 𝐯𝐚𝐫 Ẍ𝟐 𝐧𝟏 + 𝐧𝟐
𝑺 ( )
𝒏𝟏𝒏𝟐
𝜹𝟏𝟐 𝜹𝟐𝟐 𝟏 𝟏 𝐧 + 𝐧𝟐
= + =𝜹𝟐 (𝒏𝟏 + 𝒏𝟐) = 𝜹𝟐 ( 𝐧𝟏 ) N.B: the random variable obeys the student’s t
𝐧𝟏 𝐧𝟐 𝟏 𝐧𝟐 distribution with (n1+n2-2) degrees of freedom
25 26

• Then, confidence intervals for (1-α)100% 2.2.4 Confidence Interval for (P1 – P2) (for large
confidence interval for the difference between the samples)
population means (μ1 – μ2) can be obtained • Let:
through: – þ1 is observed proportion of success in random sample
𝐧𝟏 + 𝐧𝟐 of n1 and with population proportion of P1
Ẍ𝟏 − Ẍ𝟐 − 𝒕𝒏𝟏 + 𝒏𝟐 − 𝟐 𝛼 2 𝑺 ( ) – þ2 is observed proportion of success in random sample
𝒏𝟏𝒏𝟐 of n2 and with population proportion of P2
< 𝛍𝟏 – 𝛍𝟐 – the sample sizes are large
𝜶 𝐧𝟏 + 𝐧𝟐 • Then, (1-α)100% confidence interval for (P1 – P2) is
< Ẍ𝟏 − Ẍ𝟐 + 𝒕𝐧𝟏 + 𝐧𝟐 − 𝟐 𝑺 ( ) given by:
𝟐 𝒏𝟏𝒏𝟐

27 28

7
 3/18/2024

Information and the Width of the Interval Sample-Size Determination

• Interval estimation is designed to convert data into
information….. • Before determining the necessary sample size,
three questions must be answered:
• The width of the confidence interval estimate is a function of How close do you want your sample estimate to be to
the population standard deviation, the confidence level, and
the sample size… the unknown parameter? (What is the desired bound,
– If there is a great deal of variation in the random variable B?)
(measured by a large standard deviation), it is more difficult to
accurately estimate the population mean. That difficulty is
translated into a wider interval. What do you want the desired confidence level (1-α) to
– decreasing the confidence level narrows the interval; increasing it
be so that the distance between your estimate and the
widens the interval. However, a large confidence level is generally parameter is less than or equal to B?
desirable because that means a larger proportion of confidence
interval estimates that will be correct in the long run.
What is your estimate of the variance (or standard
– A larger sample size provides more potential information. The deviation) of the population in question?
increased amount of information is reflected in a narrower
29 30
interval.

Error of Estimation • which can also be expressed as

• sampling error was the difference between the
sample and the population exists because of the
observations that happened to be selected for the • This tells us that the difference between Ẍ and μ lies
sample. between and with probability 1 α.
• Now we can define the sampling error as the
difference between an estimator and a parameter
• Expressed another way, we have with probability
• We can also define this difference as the error of
estimation (Ẍ-μ) 1 α,
• In our derivation of the confidence interval
estimator of μ • We interpret this to mean that is the
maximum error of estimation that we are willing
31 to tolerate. 32

8
 3/18/2024

• We label this value B, which stands for the bound

on the error of estimation

Determining the Sample Size

• Solving for n, we produce Minimum required
sample size in estimating population mean, μ as:

• With the same sort of transformation and

calculation the Minimum required sample size in
estimating population proportion is given by:
33