Chapter 3
Sections 3.3-3.4
Principles of Inference
Confidence Intervals
Definition. If 𝜃ˆ1 and 𝜃ˆ2 are values of the random variables Θ̂1 and Θ̂2 such that
𝑃(Θ̂1 < 𝜃 < Θ̂2 ) = 1 − 𝛼
for some specified probability 1 − 𝛼, we refer to the interval (𝜃ˆ1 , 𝜃ˆ2 ) as a (1 − 𝛼)100% confidence
interval for 𝜃. The probability 1 − 𝛼 is called the confidence level, or confidence
coefficient.
Remark:
A confidence interval consists of a range of values together with a percentage that specifies how
confident we are that the parameter lies in the interval.
Example: (C.I. for the mean of normal distribution)
We construct (1 − 𝛼)100% C.I for 𝜇 based on a random sample 𝑋1 , … , 𝑋𝑛 from 𝑁(𝜇, 𝜎 2 ), when 𝜎
is assumed to be known. We know that
𝑋‾ − 𝜇
𝑍=
𝜎/√𝑛
Then,
𝑃(−𝑧𝛼/2 < 𝑍 < 𝑧𝛼/2 ) = 1 − 𝛼
where 𝑧𝛼 is such that 𝑃(𝑍 > 𝑧𝛼 ) 𝛼.
𝑋‾−𝜇
Replacing 𝑍 by 𝜎/ 𝑛, we have
√
𝑋‾ − 𝜇
𝑃 (−𝑧𝛼/2 < < 𝑧𝛼/2 ) = 1 − 𝛼
𝜎/√𝑛
𝜎 𝜎
𝑃 (−𝑧𝛼/2 < 𝑋‾ − 𝜇 < −𝑧𝛼/2 ) = 1 − 𝛼.
√𝑛 √𝑛
𝜎 𝜎
𝑃 (𝑋‾ − 𝑧𝛼/2 < 𝜇 < 𝑋‾ + 𝑧𝛼/2 ) =1−𝛼
√𝑛 √𝑛
𝜎 𝜎
Thus (𝑋‾ − 𝑧𝛼/2 , 𝑋‾ + 𝑧𝛼/2 ) is (1 − 𝛼)100% C.I for 𝜇.
√𝑛 √𝑛
𝜎 𝜎
The term 𝑧𝛼/2 is called the Margin of Error, usually denoted by 𝐸. That is, 𝐸 = 𝑧𝛼/2 .
√𝑛 √𝑛
Notice that the width of the C.I., for this case, is 2𝐸.
Relation between Testing of Hypothesis and C.I.
Suppose we are testing 𝐻0 : 𝜇 = 𝜇0 𝑣𝑒𝑟𝑠𝑢𝑠 𝐻1 : 𝜇 ≠ 𝜇0 .
Accept (Fail to Reject) the hypothesis at 𝛼 level if 𝜇0 lies in the (1 − 𝛼)100% C.I for 𝜇. Otherwise,
reject the hypothesis.
Example: Suppose 𝐻0 : 𝜇 = 62 𝑣𝑒𝑟𝑠𝑢𝑠 𝐻1 : 𝜇 ≠ 62 and a 90% C.I. for 𝜇 is the interval (58,64).
Since 62 is within this interval, we accept the hypothesis at 𝛼 = 0.1.
Sample Size
For a given margin of error, we can determine the required sample size to construct (1 − 𝛼)
confidence interval for 𝜇.
𝜎 𝜎 2
𝐸 = 𝑧𝛼/2 ⇒ 𝑛 = (𝑧𝛼/2 ) .
√𝑛 𝐸
Example: Construct a 95% C.I. of width not exceeding 3 units for the mean 𝜇 of a normal
population with 𝜎 =7.5.
7.5 2
Solution: 𝑧𝛼/2 = 𝑧0.025 = 1.96, and 2𝐸 = 3 ⇒ 𝐸 = 1.5. Thus, 𝑛 ≥ (1.96 1.5) = 96.04.
So, 𝑛 ≥ 97.
Sample Size for Testing a hypothesis
We might be interested in determining how big a sample we need to have reasonable power
against a specified value of 𝜇, say 𝜇a , in the hypothesis
𝐻0 : 𝜇 = 𝜇0 vs 𝐻1 : 𝜇 > 𝜇0
Several prespecified criteria must be considered:
1. the required level of significance ( 𝛼 ),
2. the difference, called 𝛿 (delta), between the hypothesized value and the specified value (
𝛿 = 𝜇a − 𝜇0 ), and
3. the probability of a type II error (𝛽) when the real mean is at this specified value (or one
larger than the specified value).
The value of 𝑛 that satisfies these criteria can be obtained using the formula
𝜎 2 (𝑧𝛼 + 𝑧𝛽 )
𝑛= .
𝛿2
Example: In a study of the effect of a certain drug on the behavior of laboratory animals,
a research psychologist needed to determine the appropriate sample size. The study was to
estimate the time necessary for the animal to travel through a maze under the influence of this
drug. Suppose we wanted to test the following set of hypotheses:
𝐻0 : 𝜇 = 35 vs. 𝐻1 : 𝜇 > 35
Use a level of significance 𝛼 = 0.05, 𝛽 = 0.10 if the actual mean time is 37 s, and assume 𝜎 =
11.25.
Solution:
(1.645 + 1.28)2
2
𝑛 = (11.25) = 270.7.
(37 − 35)2
Therefore, if we take a sample of size 𝑛 = 271 we can expect to reject the hypothesis that 𝜇 =
35 if the real mean value is 37 or higher with probability 0.90 .