Biostatistics

Lecture 5
Sampling Distribution of the Mean
2022-2 Fall Semester

Instructor: Min Jin Ha

Department of Health Informatics and Biostatistics
Graduate School of Public Health
Yonsei University
Reading
• Pagano and Gauvreau, Chapter 8
Overview
• While the topic may seem a little theoretical, the sampling
distribution of the mean plays a very important role in statistical
inference
• We need these points in order to make statistical inference and draw
conclusions about the results we see
• The basic idea, called the Central Limit Theorem, is that for any
distribution with well-defined mean and variance, the distribution of
the means computed for sample of size n is approximately normal.
• This is extremely important result for hypothesis testing and
construction of confidence intervals later in the course
Statistical Inference
• We examined several theoretical distributions, such as the binomial
distribution and the normal distribution
• In all cases, the relevant population parameters were assume to be known;
this allowed us to describe the distributions completely and to calculate the
probabilities associated with various outcomes
• However in most practical applications, we are not given the values of these
parameters
• Instead, we must attempt to describe or estimate some characteristic of a
population- such as mean and standard deviation- using information
contained in a sample of observations
• The process of drawing conclusion about an entire population based on the
information in a sample is known as statistical inference
Forms of Statistical Inference
• Point Estimation: estimating an unknown parameter using a single
number calculated from the sample data

• Interval Estimation: estimating an unknown parameter using an

interval or range of values that is likely to cover the true population
value

• Hypothesis Testing: checking whether sample data provide evidence

against some claim made about the population
Parameters and statistics
• It is imperative to understand the distribution between parameters
and statistics
 Point Estimation for Mean and SD
We’re interested in estimating mean height of women ages 18-74 in the US
• Target population: women ages 18-74 in the US
• Quantity of interest: mean height (𝜇) and SD (𝜎)
• Randomly sample 𝑛 women from the population
• The observed heights of the 𝑛 samples are denoted by
𝑥1 , 𝑥2 , … , 𝑥𝑛
𝟏
• The statistic for mean 𝜇 (sample mean) is 𝒙
ഥ= σ𝒏𝒊=𝟏 𝒙𝒊
𝒏
𝟏
• The statistic for SD 𝜎 is 𝒔 = √ σ𝒏𝒊=𝟏 ഥ
𝒙𝒊 − 𝒙 𝟐
𝒏−𝟏
• Point estimation is single guess about the value of a parameter in the population
 Sampling Distribution of the Mean
Suppose we can list all the women of the papulation. We then will do the
following:
• Take a random sample of size 𝑛  Sample 1
• Compute the sample mean 𝑥ҧ1
• Put the sample back, and take a second random sample also of size 𝑛 Sample
2
• Compute the sample mean 𝑥ҧ2
• Repeat this many times, we obtain many sample means 𝑥ҧ1 , 𝑥ҧ2 , 𝑥ҧ3 , …
• What is the ‘Theoretical’ distribution of the statistics 𝑥ҧ1 , 𝑥ҧ2 , 𝑥ҧ3 , … ?
• Check simulations in workshop!
 Central Limit Theorem
Let 𝒙𝟏 , 𝒙𝟐 , … , 𝒙𝒏 be the random sample 𝒏 from a population
distribution with mean 𝝁 and standard deviation 𝝈. Then the
sampling distribution of 𝒙 ഥ approaches to normal distribution with
mean 𝝁 and standard deviation 𝝈/ 𝒏 as 𝒏 → ∞.
• The the sampling distribution of the mean is identical to the population mean
• The standard deviation of the distribution of the sample mean is the
population standard deviation divided by square root of the sample size
• For the sample size large enough, the shape of the sampling distribution is
approximately normal  holds for any population distributions with finite
variances
Example
• The distribution of IQ in the general population is
normal with mean 100 and sd 15.
• Suppose we draw samples of size n=10 from this
population
• Then from the CLT, we know that the distribution of the
sample averages will also be normal with mean 100 and
standard deviation 4.7.
• If our data on IQ are exactly normal, then the
distribution of the sample averages will also be exactly
normal
• If our data on IQ are not exactly normal, then the rule of
thumb is that at a minimum of 30 observations are
need for CLT to kick in
Back to proportions
• Going back to our example, the Central Limit Theorem (CLT) tells us
that the distribution of sample averages 𝜋ො 𝑖 should follow a normal
𝜋(1−𝜋)
distribution with mean 𝜋 and standard deviation .
𝑛
• The CLT tells us that the sample averages are normally distributed if
we have enough data. This results holds even if our original variables
(here, a binary variable) are not normally distributed
• We will check this in the workshop
Applications of CLT
• Consider the distribution of serum cholesterol levels for all 20- to 74-years old males
living in South Korea. The mean of this population is 𝜇 = 211𝑚𝑔/100𝑚𝑙, and the
standard deviation is 𝜎 = 46𝑚𝑔/100𝑚𝑙. If we select repeated samples of size 25 from
the population, what proportion of the samples will have a mean value of 230mg/100ml
or above?
• By CLT, we know that 𝑋ത ∼ 𝑁(211,462 /25) and want to compute 𝑃(𝑋ത > 230)

Solution 1: find the right tail probability from 𝑁(211,462 /25)

pnorm(230,lower.tail=F,mean=211,sd=46/5)
ത
𝑋−211 230−211 230−211
Solution 2: Based on 𝑃( 46 > 46 )=𝑃(𝑍 > 46 ), find right tail probability from the
5 5 5
standard normal distribution
 pnorm((230-211)/(46/5),lower.tail=F)

Interpretation:
• About 1.9% of the samples will have a mean greater than 230mg/100ml
• If we select a single sample of size 25 from the population of 20-74years old males, the probability that the
mean serum cholesterol level for this sample is 230mg/100ml or higher is 0.019
Applications of CLT
• What mean value of serum cholesterol level cuts off the lower 10% of
the sampling distribution of means?
ത
𝑋−211 𝑀−211 𝑀−211
• 𝑃 𝑋ത < 𝑀 = 𝑃 46 < 46 =𝑃 𝑍< 46 = 0.1
5 5 5
𝑀−211 46
• Find 𝑧0.1 = qnorm(0.1) = 46  𝑧0.1 + 211 = 199.2
5
5
• Or qnorm(0.1,mean=211,sd=46/5)
• Interpretation: approximately 10% of the samples of size 25 have
means that are less than or equal to 199.2mg/100ml
Applications of CLT
• Let us now calculate the upper and lower limits that enclose 95% of the means of
samples of size 25 drawn from the population
• Consider a symmetric interval that is the shortest intervals
• 𝑃(𝑧0.025 < 𝑍 < 𝑧0.975 ) = 0.95
𝑋ത − 211
⇔ 𝑃(−1.96 < < 1.96) = 0.95
46
5
46 46
ത
⇔ 𝑃(211 − 1.96 < 𝑋 < 211 + 1.96 ) = 0.95
5 5
⇔ 𝑃(193 < 𝑋ത < 229) = 0.95
• Interpretation: Approximately 95% of the means of samples of sizes 25 lie
between 193mg/100ml and 229.0mg/100ml
• Calculate the same when we select samples of sizes 5,15,25,50,100,1000?
Applications of CLT
• Finally, let’s consider a more complicated question. How large would
our random samples need to be for 95% of their averages to lie within
+/-10 of the population mean 𝜇?
• To solve this, find sample size 𝑛 for which
𝑃(𝜇 − 10 < 𝑋ത < 𝜇 + 10) = 0.95

• Interpretation: Samples of size xxx would be required for 95% of the

sample means lie within +/-10mg/100ml of the population mean 𝜇