Links relevant for the session

▪ Confidence Interval calculator – Link

▪ Introduction to t-distribution – Link
▪ t-distribution and t-test in depth - Link
Topics Covered
▪ Problems - Interval Estimation for n >30
▪ Interval Estimation for n < 30
o t distributions
o t-tests
Motivation : Why are we learning what we are learning ?

▪ Important from the point of view of working on samples and making inferences which are backed up by
Statistical Analysis
▪ Important from Hypothesis testing perspective
▪ Examples of problems that can be solved :
➢ Measuring whether Marketing Campaigns have been actually effective
➢ Compare the effectiveness of Two or more Drugs on Patient Health
➢ Examine the statistical significance of relationship between Two variables, e.g., routine exercise and health
Problems on Interval Estimation
Quick Recap on Confidence Intervals (CI)

𝜎
ഥ
𝑿 𝝈𝑿ഥ =
𝑛
Sample Mean Std. Dev of the
sampling distribution Interval Estimates
Confidence Intervals : Example Question 01
Example : A large Automotive parts wholesaler needs an estimate of the mean life it can expect from windshield wiper blades
under typical driving conditions. We know that the standard deviation of the population life is 6 months. We select a random
sample of 100 wiper blades, collect data on their usage and obtain results, where their average life is found to be 21 months.
The wholesaler wants to know the interval estimate with a confidence level of 95 Percent.

𝜎 𝜎
𝑥ҧ − 𝑧 , 𝑥ҧ + 𝑧
ഥ = 21
𝒙 𝑛 𝑛

6 𝟏𝟗. 𝟖𝟐 , 𝟐𝟐. 𝟏𝟖
𝝈𝒙ഥ =
100

C𝒐𝒏𝒇𝒊𝒅𝒆𝒏𝒄𝒆 = 95% 𝜎
𝑳𝒆𝒗𝒆𝒍 𝑥ҧ − 𝑧 = 21 − 1.96 (0.6)
𝑛
𝒛=?
𝜎
𝑥ҧ + 𝑧 = 21 + 1.96 (0.6)
𝑛
𝒛 = 1.96
Confidence Intervals Intuition
▪ For any point estimate, there is an associated uncertainty which can cause its value to fluctuate between a range of values
▪ We can define this range (interval of estimates) using 𝜎𝑥ҧ , i.e. Standard Error of Sampling Distribution (𝑠/ 𝑛) and the z-score
o 𝑥ҧ ± 𝑧𝜎𝑥ҧ
▪ Thus, Point estimate will lie in the interval 𝑥ҧ − 𝑧𝜎𝑥ҧ , 𝑥ҧ + 𝑧𝜎𝑥ҧ

Sampling Distributions
of Sample Means

Lower ഥ
𝒙 Upper
Limit Limit

If we construct intervals around means of 1,000 randomly selected samples, then

• 680 of intervals created with + 1 Standard Errors will include the population mean
• 954 of intervals created with + 2 Standard Errors will include the population mean
• 999 of intervals created with + 3 Standard Errors will include the population mean
Earlier problems were of the type : Given a Normal Distribution 𝑿 ~ 𝑵(𝝁, 𝝈)

Find the probabilities

𝐒𝐭𝐚𝐧𝐝𝐚𝐫𝐝 𝐍𝐨𝐫𝐦𝐚𝐥 𝐟𝐨𝐫𝐦 𝒂−𝝁

𝑷 𝑿<𝒂 ? 𝑷 𝒁< ?
𝑷 𝑿>𝒃 ? 𝝈
𝒃−𝝁
𝑷 𝒁> ?
𝝈

𝒃 𝒂

𝒂−𝝁 𝒃−𝝁
𝑷 𝒂<𝑿<𝒃 ? 𝑷 <𝒁< ?
𝝈 𝝈

𝒂 𝒃
Quick Example : 𝑮𝒊𝒗𝒆𝒏 𝑿 ~ 𝑵(𝟓𝟎, 𝟏𝟎)
𝑭𝒊𝒏𝒅 𝑷(𝑿 > 𝟕𝟓) ?

𝑷(𝑿 > 𝟕𝟓)  0.0062

𝑷(𝒁 > ? )  0.0062

𝝁 = 𝟓𝟎 𝟕𝟓
Now the problems are based on a “Single Sample” of size n
Given a sample : ഥ = 𝟓𝟎 , 𝒔 = 𝟏𝟎
𝒏 = 𝟔𝟒, 𝒙 𝑭𝒊𝒏𝒅 𝑷(𝑳𝑳 < 𝝁 < 𝑼𝑳) for CL of 95%

LL ഥ
𝒙 UL
(Lower Limit) (Upper Limit)
Relation – Confidence Intervals and Confidence Levels

▪ Probability associated with an interval estimate is the confidence level

▪ Large Standard Error cause the sample statistic to be widely dispersed resulting in wider confidence intervals.
The confidence level of population parameter lying in such intervals is also high

Example
Intervals Confidence Levels
Getting delivery of a 1 Year High Confidence ~ 100%
Newly launched
Sports Car Model, 1 Month Medium Confidence ~ 50%
after booking
1 Day Low Confidence ~ 10%

1 Hour Very Low Confidence ~ 1%

Student t Distribution
Student t-Distribution : Properties
▪ In CLT, we saw for larger values of n, the distributions become Normal with smaller standard deviations (or std. error)
▪ For smaller values of n, the distribution is flatter than a Standard Normal distribution
▪ Thus, the tails go farther (on both extremes), to cover for the shortfall of area in the centre
▪ These flatter distributions are Student t-distributions (applicable to small sample sizes) ~ n < 30
Introduction to Student t-Distribution
Earlier we used the below expression for non-normal Distributions

𝑥ҧ − 𝜇𝑥ҧ 𝜎
𝑍= where 𝜎𝑥ҧ =
𝜎𝑥ҧ 𝑛

Almost in all practical situations, 𝜎 will be unknown, thus we need to use the sample statistic ‘s’ instead

Thus, the expression becomes

This statistic follows a Estimating the parameters
𝑥ҧ − 𝜇𝑥ҧ distribution different with the statistic ‘s’ results
than a Standard in an added
𝑠Τ 𝑛 Normal Distribution uncertainty and thus
greater variability (this is
reflected in t-distribution shape)
Student t-Distributions for different sample sizes

df – Degree of Freedom
t-Distribution Table