66

XII. SAMPLING DISTRIBUTION

In this chapter we will learn very important properties that first led to the development of statistical
inference. The main thought is about the probability distribution of a statistic that we can compute from a sample.

Let us consider all possible samples of size n which can be drawn from a given population with size N. For
each sample we can compute a statistic, such as the mean or the variance among others, and the value of the
statistic will vary from sample to sample. Hence, the statistic that we compute from sample to sample can be thought
of as a random variable. Being a random variable, it has a probability distribution. The probability distribution
of a statistic is called a sampling distribution. If the statistic that we compute is the mean of each possible sample,
then we have a sampling distribution of the mean. If we are interested in the variance of each sample, then we
come up with a sampling distribution of the variance.

Example 12.1. Suppose we have a population of size 4 whose values are 0, 1, 2 and 3. If we draw a random sample
of 2 elements (without replacement) the possible results and the corresponding mean of the sample are as follows:

Sample Sample mean,

x
0,1 0.5
0,2 1
0,3 1.5
1,2 1.5
1,3 2
2,3 2.5
The sampling distribution of the sample mean therefore is given by:

x P( X  x )
0.5 1/6
1 1/6
1.5 2/6
2 1/6
2.5 1/6
If we graph this, the result is symmetric close to a bell-shaped form.

For each sampling distribution we can compute the overall mean of the statistic or the variance of the
statistic, etc. For example, we can compute the mean of all the sample means, or we can compute the variance of
the sample means.

Example 12.2. The mean of the sample means in the previous example is equal to
0.5  1  1.5  1.5  2  2.5
x   1.5
6
The variance of the sample means is
k

 x   
2
i x
 x2  i 1

k
 (0.5  1.5)2  (1  1.5)2  (1.5  1.5)2  (1.5  1.5)2  (2  1.5)2  (2.5  1.5)2   6
 (1  .25  0  0  .25  1)  6  2.5  6  0.416667

Now let us also compute the mean and variance of the population from which we took these samples. Recall that
the population consists of the values 0, 1, 2 and 3.

0 1 2  3 6 (0  1.5)2  (1  1.5)2  (2  1.5)2  (3  1.5)2

   1.5 2 
4 4 4
2.25  0.25  0.25  2.25 5
   1.25
4 4
From the results we see that
x   (i.e. the mean of the sample means = mean of the pop'n)
Stat 32/ Math 31, First Semester, A.Y. 2013 – 2014 Mathematics Dep’t, MSU, Marawi City
 67

 N n   4  2  1.25  1.25
2
and  x2        .416667
 N  1  n  4  1  2  3

These relationships hold true for any population and for any size of samples. We now state these properties
formally. These are taken from the very popular theorem in Statistics called the Central Limit Theorem.

Properties: If all random samples of size n are drawn from a population with mean  and variance 2, then

a.) mean of the sample means is equal to the mean of the population, i.e.,  x  
 N - n 
2
b.) the variance of the sample means,  x2    if N is finite
 N -1  n
2
 x2  if N is infinite or N is very large
n

Equivalently, the standard deviation of the sample means is

N-n   
x    if N is finite
N -1  n

x  if N is infinite or very large
n
c.) if the sample size n is large enough (at least 30) the sampling distribution of the sample mean X is
approximately normally distributed
x  x x-μ
Hence, z or equivalently, z = is a value of the standard normal variable z.
x σ/ n

Remarks:
 N n
1. In letter b, the factor   is called the finite population correction factor. When N is very large, the value
 N 1 
of this factor is close to 1 and hence this expression is eliminated when N is very large to make it simpler in form.

2. The normal distribution of the sample mean X in letter c is found to be true when the sample size of the
samples is large or n 30. If the sample size is small, the sampling distribution of the sample mean X also follows a
normal distribution as long as the population from which the samples were taken is normally distributed.

Now let us consider some examples on how we can use these properties in projecting the result of a sample which
we will draw from a population.

Example 12.3. Suppose that the distribution of weights of a population is known to have a mean of 50 kilograms
with a standard deviation of 5 kilograms. What is the probability that a random sample of 36 elements drawn
from this population will have a mean weight between 49.5 to 50 kilograms?

Solution:
Since n=36 is greater than 30, then by the Central Limit Theorem, the sample means would be normally
distributed. Then the probability on X can be computed using the normal curve. Assuming that the size of the
population is very large, then

 
 49.5  50 50  50 
P(49.5  X  50)  P  Z   P(0.60  Z  0.00)
 5 5 
 36 36 

 P(Z  0.00)  P(Z  0.60)  0.5000  0.2743  0.2257

Example 12.4. Consider the population in the previous example. If suppose a random sample of 36 elements has a
mean of 52 kilograms, will you still believe that the mean of the population is 50 kilograms?
Stat 32/ Math 31, First Semester, A.Y. 2013 – 2014 Mathematics Dep’t, MSU, Marawi City
 68

Solution: We answer this problem by taking the P( X  52) . We include the values greater than 52 because if a
random sample that has a mean of 52 is far from 50, so much more are the values greater than 52. And so we take
their cumulative probability to see how likely this could happen.

P( X  52)  P( X  52) since X is continuous

 
 52  50 
 PZ    P( Z  2.40)  1.0  P(Z  2.40) by the rule on complementary events
 5 
 36 
 1.0  0.9918  0.0092

The probability of getting a random sample whose mean is at least 52 kg is so small that we can say it is very
unlikely to happen and yet it happened. If the mean is 52 kg it implies that almost all of the samples drawn have
weights much higher than 50 kg. But since the sample was drawn randomly, it is very unlikely that you will get
almost all of the weights higher than 50. Because it is very improbable that this thing can happen, then we are apt
to believe that the mean of the population is higher than 50 kg which can explain why we got a sample of 36
elements whose mean is 52 kg. This is basically how inference is made in research.

The use of the Central Limit theorem requires that the population standard deviation be known. However,
in many inferential problems it is not known. The theorem below states that we can use the Student t-distribution
instead of the z-distribution when the population standard deviation is unknown, n is small, i.e., n < 30, and the
population is normally distributed.

Theorem. (Student’s t - distribution)

If x and s2 are the mean and variance, respectively, of a random sample of size n taken from a population
x-μ
that is normally distributed with mean  and unknown variance 2, then t= is the value of the
s/ n
random variable T having the t distribution with v = n – 1 degrees of freedom.

Remarks:

1.) The t-distribution can be used only when the population from which the sample is taken is normally distributed
especially when the sample size is small.

2.) When the sample size n is large, the sample variance will approximate closely the value of the population
variance. Hence the t-distribution and the z distribution will be almost identical.

XIII. STATISTICAL INFERENCE

Statistical Inference is a procedure whereby inferences or conclusions about a population are made on the
basis of the results obtained from a sample drawn from that population.

Two major areas of Statistical Inference

1. Estimation
In estimation, the concern is on estimating unknown population characteristics (or parameters) like the
mean, proportion, or variance. These parameters will be estimated using the results obtained from the sample.

2. Hypothesis testing
The concern is whether to accept or reject a statement regarding a population characteristic. A parameter
is designated or hypothesized before the collection of sample data is made. To make valid decisions, most of
these statistical procedures assume that the population of interest is normally distributed.

Stat 32/ Math 31, First Semester, A.Y. 2013 – 2014 Mathematics Dep’t, MSU, Marawi City
 69

Statistical Inference I: Estimation

Suppose we want to estimate the true mean IQ of the population of MSU students. To achieve this we get a
random sample of MSU students and compute their mean IQ. If the mean IQ of the sample is 98 then we estimate
the mean IQ of the population to be around 98. But just stating a single value of the estimate without indicating
the amount of error is a dangerous habit because it can easily mislead readers to believe in something that is not
reliable.

How do we indicate the amount of error in our estimate? That depends on the estimator we are using. For
the population mean, we can use the sample mean, sample median or sample mode as our estimator. If we use the
sample mean as our estimator, then the amount of error (or the standard deviation or standard error) according to
the Central Limit Theorem assuming that the population is very large, is given by  (or s if σ is
n n
unknown and the sample size is at least 30). This means that we estimate the population mean to be within the
sample mean ± the standard error.

Therefore, if the sample size in the example above is 100 and the standard deviation of the sample is 5, then
the true mean IQ of the population of MSU students is estimated to be within 98 ± 5 . Simplifying this
100
would result to 97.5 to 98.5. Such an interval estimate for the mean is so narrow indicating a high accuracy of our
estimate.

But there is still a problem with our interval estimate because it does not indicate how confident we are that
the interval contains the true population mean. This can be solved by incorporating the sampling distribution of
the sample mean into our estimation process which, according to the Central Limit Theorem, the sample means
follow a bell-shaped distribution when the sample size is sufficiently large.

This interval estimate is now called confidence interval estimate. This degree of certainty of our confidence
interval is called the level of confidence or confidence coefficient, denoted by 1- (where α is any number between 0
to 1).

When computing a 100(1- )% confidence interval, where 1- is the confidence coefficient, we may select
any level of confidence we wish by varying the values of . Commonly used confidence coefficients are 90%, 95%
and 99% where the corresponding level of is 0.1, 0.05 and 0.01, respectively. A practical interpretation of a 95%
confidence interval for a parameter, say population mean, means that we are 95% confident that the interval
computed from a random sample of size n contains the population mean .

Confidence Interval for the Population Mean

To construct a 100(1-)% confidence interval for the population mean , consider the following notations:

x = sample mean
z/2 = standard normal deviate whose area above it is /2
t/2 = standard t deviate whose area above it is /2
 = population standard deviation
s = sample standard deviation
n = sample size
v = degrees of freedom

The computing formulas are:

σ σ
Case 1. When  is known and n  30: x - z/2 <  < x + z/2
n n
s s
Case 2. When  is unknown and n  30: x - z/2 <  < x + z/2
n n
Case 3. When  is unknown, the population is normally distributed, and n is small:
s s
x - t/2, v <  < x + t/2, v with v = n – 1degrees of freedom
n n

Stat 32/ Math 31, First Semester, A.Y. 2013 – 2014 Mathematics Dep’t, MSU, Marawi City
 70

Exercise 13.1.
1. Suppose the average IQ of a random sample of 50 MSU students is found to be 105 with a standard deviation of
9. Find a 95% and a 99% confidence interval estimate for the true average IQ of MSU students. Compare
your results.

2. The data below represents the length of life in minutes, measured to the nearest tenth, of a random sample of
10 black flies subjected to a new spray in a controlled laboratory experiment: 0.7, 1.3, 1.7, 1.9, 2.1, 2.5, 2.8, 3.4,
3.9, 6.3. Assuming that the length of life is normally distributed, construct a 99% confidence interval for the
true length of life of black flies subjected to the new spray.

3. A random sample of 8 batteries for pocket calculators is found to have a mean life of 58 hours of continuous
use with a standard deviation of 1.5 hours.
(a) Assuming that the lifetime of batteries is normally distributed, find a 95% confidence interval for the true
mean life of this battery.
(b) The manufacturer of this battery claims that the average life of this battery is 60 hours. Is the
manufacturer justified by his claim using the result in (a)?

Confidence Interval for Population Proportion

When sample size is sufficiently large (n > 20) and population proportion is known to be not so close to 0 or
1, then a 100(1-)% confidence interval for the true population proportion is given by
p̂q̂ p̂q̂
p̂ z α 2 < p < p̂ + z α 2 , where p̂ and q̂ are the proportion of success and failure, respectively, in a
n n
random sample of size n.

Exercise 13.2. In a random sample of 500 people eating lunch at a hospital cafeteria on various Fridays, it was
found that 160 preferred seafood. Find a 95% confidence interval for the actual proportion of
people who eat seafood on Fridays at this cafeteria.

Sample size determination

If we want to be 100(1-)% confident that our error in estimating the population mean will not exceed an

 z 2  
2

amount e, then our sample size will be at least n  . If the population variance is not known a pilot
 e 
 
study with n ≥ 30 may be conducted to estimate this value. The size of e depends on the amount of error in
estimating .

If the value to be estimated is the population proportion then the sample size should be at least

z2 pˆ qˆ
n 2
2
. The size of e depends on the amount of error in estimating p.
e

Exercise 13.3. A soft-drink machine is regulated so that the amount of drink dispensed is approximately normally
distributed with a standard deviation equal to 1.5 deciliters. How large a sample is needed if we
wish to be 95% confident that our sample mean will be within .852 deciliter of the true mean.

Exercise 13.4. In exercise 12.2, how large a sample is required if we want to be 95% confident that our estimate of
p is within 0.02?

Stat 32/ Math 31, First Semester, A.Y. 2013 – 2014 Mathematics Dep’t, MSU, Marawi City