Chapter 5: Sampling Distributions & Hypothesis Testing

Sampling Distributions
And
Hypothesis Testing
5.1 Introduction
Sampling is a statistical method of obtaining representative data (observations) from a
group. We have been using sampling concepts in our day to day lives knowingly or
unknowingly; for instance we take a handful of rice to check the rice quality of the full
lot. This is an example of random sampling from a large population.
Population (Universe): The group of
objects (individuals) under study is called
population or universe. Universe may be
finite or infinite.
Sample: A part containing objects
(individuals), selected from the
population is called a sample.
Random Sampling: The selection of
objects (individuals) from the universe in
such a way that each object (individual)
of the universe has the same chance of being selected is called random sampling. Lottery
system is the most common example of random sampling.
Simple Sampling: Simple sampling is a special case of random sampling in which each
event has same probability of success or failure.
Note: Every random sampling need not be simple. For example if balls are drawn
without replacement from a bag of balls containing different balls; the probability of
success changes in every trial. Thus the sampling though random is not simple.
Hypothesis: A hypothesis is an assumption based on insubstantial evidences that lends
itself to further testing and experimentation. For example a farmer claims significant
increase in crop production after using a particular fertilizer and after a season of
experimenting, his hypothesis may be proved true or false. Any hypothesis may be
accepted or rejected as per specific confidence levels and must be admissible to
refutation.
Null Hypothesis: A hypothesis which is tested for possible rejection under the
assumption of being true is known as null hypothesis. Usually the null hypothesis is
stated as ‘There is no relationship between two quantities’. It is denoted by .
Alternative Hypothesis: It is the opposite statement of null hypothesis and denoted
by .
Significance levels : The probability levels below which we reject a hypothesis are
called levels of significance. Most common significance levels employed in hypothesis
testing are , , in which critical (rejection) regions occupy
5% , 1% and 0.27% areas of normal curve respectively.
One Tailed and Two Tailed Tests: While testing statistical significance levels; one-
tailed test and a two-tailed test are used for accepting or rejecting a hypothesis. One-
tailed tests are used for asymmetric distributions (reference value is unidirectional) which
have a single tail; such as the chi-square distribution.
A two-tailed test is appropriate if the estimated value may lie on both sides of reference
value. Two-tailed tests are only applicable when the probability curve has two tails; such
as normal distribution.

One-tailed test; Two-tailed test; One-tailed test;

Reference values (two- tailed) for various levels are given in the table below:
Significance level Confidence level z value Testing at1% level of confidence means
0.27 % 99.73% 3.00 that assumption has 99% chance of
1.00% 99.00% 2.58 being true as rejection zone is just 1% of
4.55% 95.45% 2.00 the total area under the normal curve.
5.00% 95.00% 1.96 Similarly testing at 5% would provide
9.89% 90.11% 1.65 95% confidence level of acceptance.
Type I and Type II Errors: When we test a statistic at specified confidence level, there
are chances of taking wrong decisions due to small sample size or sampling fluctuations
etc.
Type I error is the incorrect rejection of a true null hypothesis, i.e. we reject , when it
is true.
Type II error is the incorrect acceptance of a false null hypothesis, i.e. we accept ,
when it is false.

Testing of Hypothesis:
Testing of statistical hypothesis is a procedure designed for accepting or rejecting a
hypothesis on the basis of some preset values.
Step1: Plant Null Hypothesis and alternate hypothesis (optional) where is the
hypothesis of no difference, i.e. presumes that there is no significant difference
between observed value and expected value.
Step2: Find the most befitting test statistic for the analysis.
Step3: Take a random sample and compute the test statistic.
Step4: is accepted if the value of test statistic lies in acceptance zone and rejected if it
falls in critical (rejection) region at the desired significance level.
5.2 Sampling Distributions
A sampling distribution is a distribution of all of the possible values of a statistic;
computed from randomly drawn samples of the same size from a population.
Some commonly used notations in sampling distributions are given below:
Population Sample

Size

Mean

Variance

Standard Deviation

Suppose we take various samples each of size from a population. If and be the
probabilities of success and failure of each member of the sample, then the binomial
distribution given by provides the sampling distribution of the number of
successes in the sample with mean and variance .
 Mean (expected value) of number of successes
Standard deviation .
Probable occurrence range at 99.73% confidence level i.e. 0.27% significance
level is given by:
Probable occurrence range at 99% confidence level i.e. 1% significance level is
given by:
Probable occurrence range at 95% confidence level i.e. 5% significance level is
given by:
In case of proportion of successes, mean and standard deviation of proportion of
successes are obtained by dividing each statistic by .
Mean (expected value) of proportion of successes

Standard deviation
Probable occurrence range of the proportion at 99.73% confidence level i.e. 0.27%
significance level is given by:
Probable occurrence range of the proportion at 99% confidence level i.e. 1%
significance level is given by:
Probable occurrence range of the proportion at 95% confidence level i.e. 5%
significance level is given by:
Standard Error: The standard deviation of the sampling distribution of a statistic is
known as Standard Error (S.E.).
Precision: Reciprocal of standard error is known as precision.
Probable Error: It is taken as 0.67449 times the standard error and is used sometimes to
explain the concept of sampling errors to layman or unprofessional people.
5.3 Sampling of Attributes for large samples (n > 30)
Characteristics like language, religion, habits (traits) etc. cannot be measured in numbers
as they are attributes. Sampling of attributes means testing how many in a population
possess a particular attribute (trait) or whether the two populations share an attribute
(trait) in common and to how much confidence level.
When sample size ( ) is very large i.e. greater than 30 and neither nor are very small,
the binomial distribution tends to normal distribution and therefore we choose the
variate as test statistic.
Following procedure is adopted for testing the significance of large samples in terms of
attributes.
Step1: Postulate the null hypothesis ( ); if required.
Step2: If is the observed number of successes in a sample and is the standard normal
variate, then , i.e.

Step3: Accept or reject as per given values of parameter at different significance

levels:

Significance Level accepted rejected

0.27%

1%

5%

In case of absence of any specified significance level, we may consider 0.27% level, i.e.
take acceptance range as .
Example 1 A coin is tossed 400 times and turns up head 216 times. Discuss whether the
coin may be unbiased one.
Solution: Let : coin is unbiased
Here , if denotes probability of success; i.e. getting a head,

then , , Expected number of heads is

Again if denotes observed number of heads; then the test statistic

; i.e.

is acceptable at both 5% and 1% levels of significance.

Hence our assumption that coin is unbiased is true
Example2 A die is rolled 900 times and turns up five or six times. Discuss whether
the die may be unbiased one.
Solution: Let : die is unbiased
Here , if denotes probability of success; i.e. getting 5 or 6,

then , , Expected number of successes is

Again if denotes observed number of successes; then the test statistic

; i.e. 2.69

but and also

Hence is acceptable at 0.27% level of significance but fails at both 1% and 5%
significance levels.
Example3 A random sample of 500 oranges was taken from a large consignment and 65
were found to be bad. Show that the standard error of the proportion of bad ones in
a sample of this size is 0.015 and deduce that the percentage of bad oranges in the
consignment lies between 8.5 and 17.5.
Solution: Let denote the proportion of bad oranges in the given sample

Then 0.13 , 0.87

Standard Deviation (S.D.) 0.015

Standard Error (S.E.) = S.D.

Probable limit of bad oranges in the consignment is given by:

0.085 to 0.175
Probable percentage of bad oranges in the consignment is 8.5% to 17.5%
Example4 A random sample of 100 bolts was taken from the lot manufactured by a
machine and 10 were found to be defective. Find the 95% confidence limits for the
proportion of defective bolts produced by the machine.
Solution: Let denote the proportion of defective bolts in the given sample

Then 0.1, 0.9

Standard Deviation (S.D.)

Probable limit of defective bolts in the lot at 95% confidence level is given by:
 0.0412 to 0.1588

Probable percentage of proportion of defective bolts in the lot at 95% confidence level
is 4.12% to 15.88%
Example5 A sample of 900 days is taken from metrological records of a district and 100
of them are found to be foggy. What is the probable percentage of foggy days in
the district?
Solution: Let denote the probability of a foggy day in the district, then

Standard Deviation (S.D.)

Probable limit of foggy days at 99.73% confidence level is given by:

; i.e. 71.71 to 128.29

Probable percentage is to i.e. 7.97% to 14.25%

5.3.1 Comparing Proportions of Large Samples from Two Different

Populations in Terms of Attributes
For an attribute (trait), let proportions and be given from two large samples of
sizes and respectively from two different populations. We may want to test whether
two populations are similar regarding the specified attribute (trait).
Working methodology:
1. Set up the hypothesis : The two populations are similar regarding the specified
attribute (trait)
2. Find common proportion of two populations for the specified attribute as:
,

3. Compute combined standard error of two populations as:

4. Compute the statistic parameter

5. Accept or reject as per given values of parameter at different significance
levels:
Significance Level accepted rejected
0.27%
1%
 5%
Example 6 In a sample of 600 men from a certain city, 450 are found to be smokers. In
another sample of 700 men from another city, 450 are found smokers. Do the data
indicate that the cities are significantly different with respect to the habit of smoking
among men?
Solution: Let : The two populations are similar regarding smoking habits among men.

Here ,

, is rejected, i.e. the difference between two populations is highly

significant in terms of smoking habits among men both at 1% and 5% levels of
significance.
Example 7 In a large city , 20% of a random sample of 900 school boys had defective
eye-sight. In another large city , 15.5% of a random sample of 1600 school boys
had defective eye-sight. Is the difference between two proportions significant?
Solution: Let : The two populations are similar regarding school boys having defective
eye sight.
Here , 0.2

, 0.155

0.1712
 2.8699

, is accepted at 0.27% level of significance, i.e. The difference between

two populations is not significant in terms of defective eye-sights.
Remark: should be rejected at 1% or 5% significance levels.
Example 8 Following data gives proportion of dark coloured people in two cities.
City Sample size Percentage of dark coloured people
250 42
450 48
Can the difference between two percentages taken as sampling fluctuations?
Solution: Let : The two populations are similar regarding smoking habits among men.
Here ,
,

1.53

, is accepted, i.e. The difference between two populations is not

significant in terms of dark coloured people both at 1% and 5% levels of significance.
5.4 Sampling Distributions of Sample means
The sampling distribution of the mean refers to the pattern of sample means, observed by
different samples drawn from the population at large.
Result : If all possible samples of size are drawn without replacement from a finite
population of size and if and denote population mean and
standard deviation respectively;

and denote mean and standard deviation respectively of sampling distribution,

then and
Result : If all possible samples of size are drawn with replacement from a finite
population of size and if and denote population mean and standard deviation
respectively;

and denote mean and standard deviation respectively of sampling distribution,

then and

Example 9 Suppose a population consists of five numbers: 1,3,5,7,9. Find sampling

distribution of sample means. Also calculate mean and standard deviation of the sample
means, if random samples of two numbers are drawn
without replacement with replacement
Verify Results and by comparing these statistics with population mean and
population standard deviation.

Solution: Population mean ( )

Population variance ( )

Total number of possible samples each of size two without replacement is 5

Sample Sample mean ( )
1,3 2
1,5 3
1,7 4
1,9 5
3,5 4
3,7 5
3,9 6
5,7 6
5,9 7
7,9 8
Sampling distribution of Sample mean ( )
2 3 4 5 6 7 8
Frequency 1 1 2 2 2 1 1
 Mean of sample means ( )

Variance of sample means

Standard deviation of sample means ( )

, also

Hence Result is verified.

Total number of possible samples each of size two with replacement is
Samples of 2 numbers with replacement are shown below; individual sample mean is
given under each sample.
(1,1) (1,3) (1,5) (1,7) (1,9)
1 2 3 4 5
(3,1) (3,3) (3,5) (3,7) (3,9)
2 3 4 5 6
(5,1) (5,3) (5,5) (5,7) (5,9)
3 4 5 6 7
(7,1) (7,3) (7,5) (7,7) (7,9)
4 5 6 7 8
(9,1) (9,3) (9,5) (9,7) (9,9)
5 6 7 8 9
Mean of sample means is given by:

Variance of sample means is given by:

Standard deviation of sample means ( )

and , hence Result is verified.

Result : If a population is normally distributed with mean and variance and
respectively; also denotes sample mean for a random sample ( , , , ); then
sampling distribution of is also normally distributed with mean and variance , i.e.
If then

Proof: Since the population is normally distributed, for any object of the population;
and

Also

Standard deviation of sampling distribution ( )

Standard Error: The standard deviation of the sampling distribution is called the
standard error Standard error of sampling distribution is

Remark: We have assumed here that is known. However if is not known, we

take to be equal to the standard deviation of the sample.
Properties of the sampling distribution of sample means:
 A sample drawn from a normally distributed population follows normal distribution,
Z-value for the distribution of is given by

 The mean of the sample means will be the same as population mean from which
the samples were drawn, i.e.
 The variance of the sampling distribution of will be equal to the variance of the
population divided by the sample size i.e.
Result : Central Limit Theorem: As the sample size gets large enough (30 or
higher); the sampling distribution becomes approximately normal regardless of shape of
population.
Remark: For large samples ( ), probability distribution is taken as normal for
computational purposes.
Example 10 A population has mean 0.1 and standard deviation 2.1. Find the probability
that the mean of a random sample of size 900 will be negative.
Solution: Given that population mean and standard deviation
Since the sample size is large enough, sampling distribution is approximately normal
with mean 0.1 and standard deviation 0.07

i.e. and

Example 11 Suppose a population has mean 10 and variance 4. What is the probability
that the sample of size 36 has mean lying between 9.8 and 10.2?
Solution: Given that population mean and standard deviation . Since
the sample size is large enough, sampling distribution is approximately normal with mean
10 and standard deviation 0.33

i.e. and
Example 12 A firm produces electric bulbs that have normally distributed mean burning
life 800 hours with a standard deviation of 40 hours. Find the probability that a random
sample of 16 bulbs will have average burning life of less than 775 hours.
Solution: Given distribution is normal with population mean and standard
deviation , sample size

By symmetry of standard normal curve

Result : Test of significance of sample mean for a large sample

Here we test the hypotheses about the sample mean of a population in the case when
sample size is at least 30 (for large samples, distribution is approximately normal as per
central limit theorem). Let and be mean and variance of a population. If be the
mean of a simple large sample of size from an unknown population, we may want to
test whether the sample belongs to given population.
Working rule:
1. Set up the hypothesis : the sample belongs to given population
2. Compute the statistic parameter
3. Accept or reject as per given values of parameter at different significance
levels:
Significance Level accepted rejected

0.27%

1%

5%
Remark: The statistic parameter can also be used to check whether the sample
taken from the given population is random or not.
Example 13 A sample of size 900 is having mean 3.6 mm; could it be reasonably
regarded as a random sample from large population whose mean is 3.35 mm and standard
deviation 2.6 mm at 1% significance level.
Solution: Let : sample belongs to the given population
Here population mean mm, population standard deviation mm,
also sample size and sample mean mm

Now

As , is rejected, i.e. difference between sample mean ( ) and population

mean ( ) is significant at 1% level and the sample may not be considered from the same
population.
Example 14 A random sample of 400 canes of vegetable oil with labeled net weight as
5kg has a mean net weight of 4.98 kg with a standard deviation of 0.22 kg. Can we accept
the hypothesis of net weight 5 kg per cane on the basis of given sample, at 5 % level of
significance?
Solution: Let : Mean net weight of each tin is 5 kg.
Here population mean kg, sample size and sample mean kg,
sample standard deviation kg

Now

As , is accepted at 5% significance level, i.e. Mean net weight of each tin is

5 kg.
Note: Here population standard deviation is not known, sample standard deviation is
taken as population standard deviation.
Example 15 A pharmaceutical company fills its best-selling 8 ounce jars of liquid
medicine by an automatic dispensing machine. The machine is set to dispense a mean
of 8.1 ounces per jar with a standard deviation of 0.22 ounce. Uncontrollable factors in
the process can shift the mean away from 8.1 and cause either under-fill or overfill, both
of which are undesirable. In such a case the dispensing machine is stopped and
recalibrated. A quality control engineer routinely selects 30 jars from the assembly line to
check the amounts filled. On one occasion, the sample mean is 8.2 ounces with the
standard deviation of 0.25 ounce. Determine if there is sufficient evidence in the sample
to indicate, that the machine should be recalibrated, at the 5% level of significance.
Solution: Let : machine need not be recalibrated
Here population mean ounce, population standard deviation
ounce.
Also sample size and sample mean ounce

As , is rejected, i.e. difference between sample mean ( ) and population

mean ( ) is significant at 5% level and the machine needs to be recalibrated.
Result : Test of significance of difference between means of two large samples

(a) If samples are from two different populations

Let and be sample means of two populations of sizes and from two different
normally distributed populations having means ; and variances ; respectively;

i.e. and

Then

Also

Standard Error ( )

-value for the distribution of is given by

(b) If samples are from same population

If and be two sample means of sizes and taken from a normally distributed
population having mean and variance

Then

Also
 Standard Error ( )

z-value for the distribution of is given by

Example 16 A random sample of 150 villages was taken from a district A having
standard deviation 32 and average population per village was found to be 440. Another
random sample of 250 villages from district B with a standard deviation of 56 gave an
average population of 480 per village. Is the difference between the averages of two
populations significant? Give reasons.
Solution: Let : The differences between averages of two populations is not significant,
i.e.

Here , ,
, ,

, under the assumption

, thus differences between two averages is highly significant.

is rejected and the averages of two populations cannot be taken as same.
Example 17 Random samples of 500 and 400 are having means 11.5 and 10.9
respectively. Can the two samples be regarded as drawn from the population of
standard deviation 5?
Solution: Let : The two samples are drawn from the same population with standard
deviation
Here ,
,
 , is accepted at 5% level of significance, i.e. the two samples can be
considered to be drawn from same population.

Example18 50 new entrants in a class are found to have an average height of

135cm and 30 old one have an average height of 140cm with a class standard
deviation of 8cm. Does this indicate that mean height of old students is greater than
that of new entrants?
Solution: Let : Average height of new entrants is same as old students
Here standard deviation of the class is
,
, ,

, is accepted at 0.27% significance level, i.e. average height of new

entrants is same as old students.
Remark: will have to be rejected at 1% and 5% levels of significance.
Exercise 5

1. A coin is tossed 400 times and head turns up 225 times. Discuss whether the coin
is biased or unbiased at 5% level of significance.
2. A random sample of 600 oranges was taken from a large consignment and 60 were
found to be rotten. Show that the standard error of the proportion of bad ones in a
sample of this size is 0.1 and deduce that the percentage of bad oranges in the
consignment almost lies between 6.3 and 13.7
3. In a city 20% of a random sample of 900 school children wore spectacles and in
another city 18.5% of a random sample of 1600 school children used to wear
spectacles. Is the difference between the proportions significant?
4. In a sample of 500 people from a state 280 take tea and rest take coffee. Can we
assume that tea and coffee are equally popular in the state?
5. A sample of 900 members is found to have a mean of 3.4cm. Can it be reasonably
regarded as truly random sample from a large population with mean 3.25cm and
S.D. 1.61cm.
6. A sample of 100 electric bulbs produced by a manufacturer showed a mean life
time 1190 hours with a standard deviation of 90 hours. Another sample of 75
bulbs produced by manufacturer showed a mean life time 1230 hours with a
standard deviation of 120 hours. Is there a difference between the mean life times
of two brands at 5% level of significance?
7. The means of two large samples of 1000 and 2000 members are 168.75 cm and
170cm respectively. Can these be regarded as drawn from the same population of
standard deviation 6.25 cm.
8. A stenographer states that he can take dictation at the rate of 120 words per
minute. Can we accept his claim on the basis of 100 trials in which he showed a
mean of 116 words with standard deviation of 15 words ?
9. A sample of height of 6400 soldiers has a mean of 67.85 inches and a standard
deviation of 2.56 inches, while a random sample of heights of 1600 sailors has a
mean of 68.55 inches and a standard deviation of 2.52 inches. Does this indicate
that the sailors are on average taller than the soldiers?
10. A random sample of 400 students has an average weight of 55 kg. Can we say that
the sample comes from a population with mean 58 kg. with a variance of 9 kg. ?
11. In a big city two samples of people are drawn. First sample of size 100, the
average daily income of people is 210$ with a standard deviation 10$ and in the
second sample of size 150 persons, average daily income is 220$ with a standard
deviation of 11$. Test if there is any significant difference in average incomes.
Answers

1. Biased
3. the difference is not significant
4. the difference is highly significant
5. it cannot be regarded as a random sample.
6. Yes
7. No
8. The claim is not acceptable
9. Yes
10. No
11. the difference is highly significant.