Hypothesis testing

In many statistical situations, we are more interested in testing statements about the
population rather than estimating one of the parameters. This procedure of statistically
testing claims about populations or universes is known as hypothesis testing.
Hypothesis testing begins with an assumption called a Hypothesis that is made about
the population. Data are then collected from a representative sample after which it is
analyzed using a suitable sample statistic. The information obtained from the sample
is used to decide how likely it is that the hypothesized population parameter is correct.

What is a Hypothesis?
When a statistician sets out to make some investigation about a given problem, he has to
make prior guesses or predictions about the eventual findings. This is usually done basing
on certain theories or practices in as far as the population of interest is concerned.

Definition:

A statistical hypothesis is an assumption or guess or a prediction, which may or may

not be true, that is made about a population.
The procedure involves taking a sample and on the basis of the results from the sample
data, we evaluate the accuracy of the claims concerning the population at some specified
level of significance (  − level ) .This procedure based on sample evidence and probability
theory enable us to determine whether the hypothesis is a reasonable statement.
Hypothesis testing starts with a statement or assumption about a population parameter
such as the population mean.
Example
A hypothesis might be that the mean monthly household income of Gisozi Sector in
Gasabo district is:  = 20,000 frw. We cannot count all the people in the sector to ascertain
that the mean is in fact 20,000.frw.
To test the validity of the assumption (μ=2000frw), we must select a sample from the
population of all people the sector, calculate the sample statistic(sample mean) and based
on certain decision rules accept or reject the hypothesis.
The following steps are involved in the testing hypothesis.
1. State the hypothesis (known as the null hypothesis (H 0 ) that will be tested and its
alternative hypothesis H1.
2. Select and compute the appropriate sample statistic (chi-square, student-t-statistic, z-
statistic etc.)
3. Using a specified level of significance normally 5 % (  = 0.05) or 1% (  = 0.01)
determine the acceptance and rejection regions.
4. Using the sample results compute the appropriate value based on the chosen test
statistic (chi-square value, student-t-vale, z-vale etc.)
5. Compare the computed/obtained value (  ob 2
, tob , Zob ) with the tabled/critical value (
 cv2 , tcv , Z cv ) at the specified level of significance. If computed value falls within the
acceptance region then the null hypothesis is accepted and if it falls within the rejection
area/region (critical region) then the null hypothesis is rejected.

The diagram below showing the rejection and acceptance regions based on the z-values
at  = 0.5 ,two-tailed.

(Rejection region) (Rejection region)

 =0.025  =0.025
2 2
0.475 0.475

=1
Z=+1.96 Z=0 Z = - 1.96

In the diagram above, the area where the null hypothesis is not rejected is:
-1.96 < Z < +1.96.
The value 1.96 is called the critical value.
The decision rule is formulated by fining the critical value of the statistic being used.
Statistical Hypotheses
The Null Hypothesis (H0 or HN)
A hypothesis that predicts no difference or relationship between given variables is
called a Null Hypothesis. It is normally formulated for the sole purpose of rejecting it.
There is usually a “not” or a “no” term in the null hypothesis meaning that there is “no
change.”
For example the Null hypothesis may be:
The mean number of life hours of a light bulb produced by NOVA Electric Company is not
different from 60 000 hours. H = 60,000
The Null hypothesis is written as: H0 :  = H or H0 :  = 60,000
It should be noted that we often begin the Null hypothesis by stating:
“There is no significant difference between……….,” or
“The mean number of life time hours of light bulbs produced by NOVA Electric Company
is not significantly different from 60,000.”
Such a hypothesis that predicts no difference or relationship between
given variables is called a Null Hypothesis.
Examples:
• There is no difference in Academic performance between male and female students in
statistics.
• There is no relationship between smoking and lung cancer.
• There is no difference in racial attitudes between children who grow up in multi- racial
communities and those who grow up in mono racial ones.
• There is no relationship between the number of road accidents in Rwanda and the
number of left-hand drive vehicles. Etc
Alternative Hypothesis H1 or HA
This Hypothesis states a difference or relationship between variables. It is the opposite of
the Null hypothesis. It may be directional or non directional. When it is directional, it
states the direction of the difference (bigger than , smaller than) etc .
The Alternative hypothesis is written as:
H A :   H or H A :   60,000
This is the null hypothesis which has to be tested.
Examples.
• Boys perform academically better than Girls in Maths
• Children who grow up in multi –racial communities exhibit fewer racial attitudes
than those who grow up in mono-racial ones.
• There is a difference in Academic Performance between students who are taught
using method A and those taught using method B.
Types of Errors committed in Testing Hypotheses

When testing the hypothesis, there one two possible errors.

• Type I error (the alpha error).
• Type II error (the beta error).
Type I error is committed when a null hypothesis is rejected when it should have been
retained. Type I error is the incorrect rejection of a true null hypothesis (false positive).
Stopping a production line to recalibrate a piece of equipment when the operation is
within tolerable limits is an example of type 1 error.
Type II error is the error committed when a null hypothesis is retained when it should
have been rejected. Type II error is incorrectly retaining a false null hypothesis (false
negative).Not stopping a production line to recalibrate a piece of equipment when the
operation is outside of tolerable limits is an example of type 2 error.
As the probability of a Type I error increases, the probability of a Type II error
decreases.
The reverse is also true, as the probability of a Type I error decreases, the probability of
a Type II error increases.
P (Type I) + P (Type II) = 1.0
To minimize Type I error, a very small α – level should be used. The smaller the α – level,
the smaller the critical region and the greater the mean μ of the sampling distribution and
the beginning of the critical region. With a small α – level, it will be hard to reject the null
hypothesis.
However as the critical region decreases in size, the non-critical region becomes larger.
This now increases the probability of committing Type II error.
This shows that although the probability of committing type I error decreases with a lower
α – level, the probability of committing Type II error increases with a lower α – level.
Thus the two errors are inversely related and it is not possible to minimize both errors in
the same test.
What all this means finally is that the selection of an α – level must be in such a way as to
balance the two sources of errors. The α – level of .05 tries to balance b the two types of
errors.
A good way of minimizing both types of errors is to increase the sample size and to select
a non-biased representative sample.

The Table below is a recap of both types of error and how they are defined.

Actual Situation in the Population

Ho is true Ho is False
Accept Ho (DNR) Correct Decision Type II Error
Action Taken Beta Error
Reject H o Type I Correct Decision
Alpha Error

Level of Significance (α- level)

The level of significance is the maximum probability with which the researcher would be
willing to reject a Null Hypothesis when it should have been retained. It is stated in terms
of areas above a critical value.
The level of significance is the probability of rejecting the null hypothesis
when it is true (type I error).It is usually set at or below 5%.
As an example, consider a situation where a researcher is interested in finding out
whether the mean monthly family income of people in Gasabo District has changed or
not. Suppose that according to a prior investigation in the year 2015 the mean monthly
family income was 3600Frw.Now in the year 2016 the researcher wishes to find out
whether there has been any change as a result of the introduction of small scale industries
in the District.
Two Hypotheses are formulated:
Ho : The mean monthly family income has not changed : μ = 3600 or μ = μH
H1 : The mean monthly family income has changed : μ ≠ 3600 or μ ≠μH
The researcher then takes a random sample and on the basis of this sample, the mean is
calculated. Now as long as the mean does not deviate too much from 3600 Frw; we may
say that the mean monthly family income has not changed. If the deviation is too large,
Ho is rejected and we conclude that the mean income has changed.
Now a question arises:
What does too much or too large mean? Where should we draw a line
between small and insignificant deviations and large significant ones that
may have another cause other than mere sampling error?
The too much or too large is determined by setting a level of significance (α-level).

The test procedure divides the sample space into two mutually exclusive parts.

• The part containing the Rejection region; the region within which Ho will be
rejected and
• The part containing the Acceptance region within which Ho will not be rejected
(retained)
In this particular case, the rejection region will consist of two parts as shown in the
diagram below.

Acceptance region

R R

C1 μ = 3600 C2
In the diagram above, the shaded parts show the Rejection regions R
To decide whether the mean family income has changed, the researcher has to determine
two values c1 and c2 that set limits on the amount of sampling variation he feels is
consistent with the Null hypothesis. If the mean income lies between C1 and C2 , Ho is not
rejected. The values C1 and C2 ar determined by the  − level .
Rejection region: The rejection region is a range of values such that if the test statistics
falls into that range, we decide to reject the null hypothesis in favor of the alternative
hypothesis.
One –Tailed and Two Tailed -Tests of Hypothesis
There are three forms of hypothesis testing.
• Form I is a left-tailed rejection region.
• Form II is a right-tailed rejection region.
• Form III is a two-tailed rejection region.
Two Tailed Test of Hypothesis:
Definition:
A Two-Tailed test is a test in which the researcher wants to determine whether a
population parameter has changed regardless of the direction of change.
It is a test in which the critical region/area of a distribution is two sided and tests
whether a sample statistic is greater than or less than a certain critical value.
In the example just considered above, the test of the Null hypothesis was non-directional
because the alternative hypothesis did not specify the direction of the change of the mean
monthly family income.
It did not specify whether the mean income had decreased or increased.
HO: μ =μH : There is no change in the mean monthly family income .
H1: μ ≠μH: The mean monthly family income has changed
In a two- tailed test, the critical region falls within both tails of the distribution curve. The
diagram below shows such a test at α = .05

Acceptance region = 95%

or 0.95

α/2= 0.025 α/2= 0.025

C1=-1.96 μ = 36000 C2 =+1.96

C2
One –Tailed Test of Hypothesis:
Definition: A one-Tailed Test of Hypothesis is a test in which the researcher wants to
determine whether a Population parameter is greater or less than a hypothesized value.
It is called one tailed because the Null hypothesis will be rejected by observing a statistic
in only one tail of the appropriate sampling distribution. This tail may be either the lower
or the upper.
Example:
Suppose that in the past under carefully observed customer care conditions, the number
of business people registering their Businesses in RDB per month on the average has been
30 people or less.

An expert in Customer Care has now devised a new method of registration and now claims
that the number of people registering their businesses per month is greater than 30.A
researcher decides to test this claim by selecting a random sample of 50 businesses. Two
hypotheses are now formulated as follows:

HO: μ ≤30: The mean number of registered businesses per month is less than or equal
to 30.
H1: μ > 30: The mean number of registered businesses per month is greater than 30.
In this case the Null hypothesis would be rejected if the mean ( X ) of the sample is more
than some appropriate value C and not to reject the Hypothesis of the mean is equal to or
less than C.
The diagram below shows such a test at α =.05

95% = 0.95

0.500 0.450

α=0.05

μ = 30 C

The test described above is a one tailed test.

Here, X must fall below C for the null hypothesis to be rejected.
Conclusions of a Test of Hypothesis
1. If we reject the null hypothesis, we conclude that there is enough statistical evidence to
infer that the alternative hypothesis is true.
2. If we do not reject the null hypothesis, we conclude that there is not enough statistical
evidence to infer that the alternative hypothesis is true.

Degrees Of Freedom

The number of degrees of freedom (df) for any statistic (mean, variance, χ2, t-value etc) is
the number of observations / values or scores in its calculation that are free to vary. This
degree of freedom refers to the degree of a score to have any possible value.

Suppose you are asked to pick two numbers without a restriction. Both numbers would
be free to vary (take any value) and in this case, there are two degrees of freedom.
If a restriction is imposed – say that Σ xi = 0 , (i = 1,2), then one degree of freedom is lost
because of the restriction. Only one of the numbers would be free to vary.
For example, if you choose 3 for the first number, the second number must be -3 because
of the restriction that Σ xi = 0.
In a similar way, if you were to pick 5 numbers with a restriction that Σ xi = 19, or any
other number, you would have only 4 degrees of freedom. Once the 4 numbers are chosen
say as 2, 3, 4, and 7, than the 5th number would strictly be 3.
Degrees of Freedom for the mean X

The sample x has N degrees of freedom if the sample size is N. to show this; let us examine

the formula for the mean. X =

 Xi
n

Now when calculating a mean for any specified sample of scores, the N does not vary
because it is a fixed value for the sample. In contrast, the scores x i which are used in the
calculation of Σ xi do not take on different values; they are free to vary. Knowing the values
of all but one score does not tell us the value of the last score.
For example if there are five scores and only 4 of them (5, 3, 2, 7, X) are known, there is
no way the value of X can be determined. This we must know all the N scores in order to
calculate the mean.
It is in this sense that all the N scores are free to vary making the number of degrees of
freedom for the mean to be equal to N.

Number of degrees of freedom for the variance (S2)

The variance is given by the formula: S 2

=
( X i − X )2
n −1
Neither the sample size N, nor the mean x is free to vary. For any specified sample, N and
x are constants.
However, the deviations between each score and the mean ( Xi - X ) take on different
values depending upon the scores in the distribution. Now how many of these N
deviations are free to vary?
The answer is that all but one (N – 1) of these deviations are free to vary due to the fact
that the sum of the deviations about the mean is always zero.
That is Σ ( Xi - X ) = 0.
Knowing N – 1 of these deviations, you can always determine the last one.
Thus the number of degrees of freedom for the sample variance (and standard deviation)
is N – 1.