You are on page 1of 78

HYPOTHESIS

&
Its Testing
Gaurav Singh
R.B.T.T.I., Bareilly
Hypothesis
 Hypo- Less Then
 Thesis- view or theory
 It is a tentative explanation or solution.
 It is declarative testable relationship between
two or more variables.
 A hypothesis is the assumption that we make
about the population parameter. This can be any
assumption about a population parameter not
necessarily based on statistical data.
Hypothesis:
Not Necessary in-
 Fact finding studies
 Exploratory researches
 In historical researches (theoretical)
 But in all Analytical and experimental
researches, hypothesis should be
formulated.
Characteristics

 Conceptually clear
 Testable
 Economic and parsimonious
 Related to existing body of theory and
facts.
 Logical unity and comprehensiveness
 General in scope
 Related to available scientific tools and
techniques
 Tentative answer to the proposed
problem
 Operational
 Specific but not trivial or inconsequential
Types of hypothesis
 Based on function:
 Descriptive
 Relational
 Based on approach:
 Working
 null
 Statistical
 Based on level of abstraction
 Common sense
 Complex
 Analytical
Descriptive

 It describes the characteristics of a


variable.
 The customer satisfaction level is
significantly good among the Hero Honda
owners.
 Public enterprises are more amendable for
centralized planning.
Relational

 It describe the relationship between two


variables.
 The Families with higher income spend more
for recreation.
 The lower the rate of job turnover in a work
group, the higher the work productivity.
Working hypothesis

 Initial hypothesis
 Not very specific
 These are subject to modification as the
investigation proceeds.
Null Hypothesis
 Null means no difference
 It is a non directional hypothesis.
 It is also known as Statistical Hypothesis
 It confirms the qualities of detachment and
objectivity.
 Null hypothesis are more exect.
 either it is accepted or rejected.
If rejected-
 Alternative hypothesis formulated.
 One tailed & Two tailed
Statistical hypothesis

 Statements about statistical population.


 These are derived from a sample.
 It may be hypothesis of difference or
hypothesis of association.
 In broad sense, all hypothesis might be
said statistical hypothesis in broad sense.
Common sense hypothesis
 These are common sense ideas.
 These are simple level hypothesis.
 Shop assistants in small shops lacks motivation.
 Shoulders from upper class are less adjusted in the
army then lower class men.
 it requires three tasks:
 Removal of value judgment
 Clarification of terms
 Application of validity tests
Complex hypothesis
 It aims at testing the existence of logically
derived relationships between empirical
uniformities.
 These are purposeful distortions of empirical
exactness.
 The function of such hypothesis is to create
tools and problems for further research in
otherwise very complex areas of investigation.
Analytical hypothesis

 These are concerns with the relationship


of analytical variables.
 These occur at highest level of
abstraction.
 e.g. there are two high fertility population
segments in India, viz. low income urban
Muslims and low income rural low caste
Hindus.
Sources of hypothesis
 Theory
 Observation
 Analogies
 Intuition AND personal experiences
 Findings of studies
 State of knowledge
 Culture
 Continuity of reserch
Functions of hypothesis
 It gives a definite point too the investigation.
 It guides the direction of study.
 It specified the source of data.
 It determines the data needs.
 It defines which fact is relevant.
 It determines most appropriate technique of
data analysis.
 It contributes to the development of theory.
Characteristics of a Good
Hypothesis
 Conceptual clarity
 Specificity
 Testability
 Availability of technique
 Theoretical relevance
 Consistency
 Objectivity
 Simplicity
Evaluating hypothesis
 Have the all aspects of the problem been considered in
the process of hypotheses formation?
 Do the hypotheses include all the pertinent possibilities
to answer the research question?
 Have the hypotheses been selected on the basis of
such possibilities?
 Have the hypotheses selected without fearing the
possibilities of tits sustainability?
 Have conditions at hypotheses selection, allowed the
researcher to reach the real solutions of problem?
 Are the concepts used in hypotheses specific?
 Is the posited relationship between the
variables verified?
 Is there any prior evidence as to truth or
falseness of posited relationship?
 Can an appropriate research design be
devised?
 Are the generalizations a part of a theoretical
system?
Process of hypothesis setting

 For each objective, search the possible


answers.
 It requires searching, delving, trying,
failing, and trying again and coming to a
conclusion.
 Then write it in appropriate style.
 Evaluate these tentative hypothesis and
refine them in logical and testable way.
Rules of hypothesis
testing
 Search the variable measurement with most
quantitative characteristics.
 Make variable scales in mutually exclusive and
totally inclusive categories.
 Describe the meaning of terms operationally in
testable form.
 Always consider alternative operations.
 Analyze variables through relationship.
 Link two or more formal propositions through a
shared independent or independent varaibles.
Testing the hypothesis
Procedure of Hypothesis
Testing
 First, state 2 hypotheses,
 the null hypothesis (“H0”) and
 the alternative hypothesis (“HA”)
 Select an appropriate statistical test
 Select the desirable level of significance
 Compute the appropriate statistic from the
sample data
 Compute the significant test value
 Obtain the critical test value
 Make the decision
Identification of hypotheses

 The null hypothesis always represents


the status quo, i.e. the hypothesis that
requires no change in current behavior.
 The alternative hypothesis is the
conclusion that the researcher is trying to
make.
 In statistics, we always assume the null
hypothesis is true.
 Then, make a decision based on the
available evidence.
 If there is sufficient evidence (“beyond a
reasonable doubt”), reject the null
hypothesis.
 If there is not enough evidence, do not
reject the null hypothesis.
Hypothesis Testing

We generally designate values of the


parameter, say , under the null
hypothesis as 0.
Hypothesis Testing

Example 1. An ambulance service is


considering replacing its ambulances with
new equipment. If 0 is the average
weekly maintenance cost of one of the old
ambulances and  is the average weekly
maintenance cost it can expect for one of
the new ones, it wants to test the null
hypothesis  = 0.
Hypothesis Testing

(a) What alternative hypothesis should it


use if it wants to buy the new ambulances
only if it can be shown that this will reduce
the average weekly maintenance cost?
Answer:  < 0. This is called a one-sided
alternative.
Hypothesis Testing

(b) What alternative hypothesis should it


use if it is anxious to buy the new
ambulances (which have some other nice
features) even if they are more costly but
has no idea if they are or not.
Answer:   0. This is called a two-sided
alternative.
Hypothesis Testing

Nulls and alternatives can take the


following forms
Null Possible Alternatives
 = 0   0  < 0  > 0
  0  < 0
  0  > 0
Hypothesis Testing

Now we are going to either reject the null


hypothesis or not. In doing so it is
important to realize that we can make two
types of errors in rejecting the null
hypothesis.
Hypothesis Testing

Type I error is rejecting the null


hypothesis when it is true.

Type II error is not rejecting the null


hypothesis when it is false.
Hypothesis Testing

Accept H0 Reject H0

H0 is true Correct Type I


Decision Error

H0 is false Type II Correct


Error Decision
Hypothesis Testing

We would commit a Type I error


if we rejected the device when it was
indeed effective. We would commit a
Type II error if we failed to reject the
device when it was ineffective.
Hypothesis Testing

We call the probability of type I error, or


the probability of rejecting the null
hypothesis when it is true, . We call the
probability of type II error, or the
probability of not rejecting the null
hypothesis when it is false, .
Hypothesis Testing

, the probability of type II error, is a


difficult concept in the theory. So for this
elementary presentation we concentrate
on , the probability of type I error – the
probability of rejecting the null hypothesis
when it is true.
Hypothesis Testing

The first approach to hypothesis testing is


traditional.

1. We formulate a null hypothesis


and an appropriate alternative hypothesis
from the language of the problem.
Hypothesis Testing

2. We specify a probability of type I error by


convention. Generally people choose  = .01
or .05. That is to say, we decide that we are
willing to tolerate a probability of .01 or .05 of
making a type I error, i.e., of rejecting the
null hypothesis when it is true.  is called the
significance level of the test.
Hypothesis Testing

3. Based on the sampling distribution of


an appropriate statistic, we construct a
criterion for testing the null hypothesis against
the chosen alternative hypothesis at the
specified level of significance. We use a two-
sided criterion for a two-sided alternative and a
one-sided criterion for a one-sided alternative.
Hypothesis Testing

4. We calculate the value of the statistic


on which the decision is to be made.

5. We decide whether or not to reject the


null hypothesis. Essentially we reject the
null hypothesis if what we observe is too
far from it.
Hypothesis Testing

We call the test one- or two- tailed


depending on whether the criterion (or
alternative) is one-sided or two- sided.
Hypothesis Testing

Example 2. It has been claimed that on


the average 2.6 workers are absent from
an assembly line. If an efficiency expert
is asked to put this to a test,
Hypothesis Testing

(a) What null hypothesis and what


alternative hypothesis should she use?

(b) Should she use a one-tailed test or a


two-tailed test if she is going to base her
decision on the mean of a random
sample?
Hypothesis Testing

Solution.
(a) The null hypothesis is 0 = 2.6. If
management worries only that more workers
are absent then she should use the alternative
 > 2.6.
(b) If her alternative is  > 2.6 then she should
use a one-tailed test because the alternative
hypothesis is one-sided.
Hypothesis Testing

We illustrate the traditional procedure with


an example. In the example, we are
dealing with one sample,  is known, and
n is > 30.
Hypothesis Testing

Example 3. In a study of new sources of food,


it is reported that a pound of a certain kind of
fish yields on the average 3.52 ounces of FPC
(fish-protein concentrate) used to enrich various
food products, with standard deviation 
= 0.07 ounces.
Hypothesis Testing

To check whether  = 3.52 is correct,


a dietician decides to use the alternative
hypothesis   3.52 ounces, a random
sample of size n = 32, and the .05 level
of significance. What will she conclude if
she gets a sample mean of 3.55 ounces
of FPC (per pound of fish)?
Hypothesis Testing

Solution. The null hypothesis is 0 = 3.52


ounces,  = .07, n = 32 and _
X = 3.55. _
If the null hypothesis is true then E(X) =  =
0 = 3.52. Also from the CLT we know that
_
Z = (X - 0)/(/sqrt(n)) has approximately a
standard normal distribution.
Hypothesis Testing

Also, the dietician chose  = .05 and a


two-sided alternative. That means that in
this case “far away” is defined as a total of
.05 probability in the two tails. From the
board drawing you can see that z.025 =
1.96 and - z.025 = -1.96 leave .025 in each
tail respectively so a total of .05 in the two tails.
Hypothesis Testing

So the statistician rejects the null


hypothesis if the observed value of Z
under the null hypothesis is in either tail,
i.e., > 1.96 or < -1.96. But the observed value
of Z under the null hypothesis is
_
(X - 0)/(/sqrt(n)) =
(3.55 – 3.52)/(.07/sqrt(32))  2.42.
Hypothesis Testing

Since 2.42 is in the right tail, i.e., > 1.96


the observed mean 3.55 is too far away
from the null hypothesized mean 3.52 and
so the dietician must reject the null
hypothesis.
Hypothesis Testing

In the previous example the dietician


chose  = .05 for her level of significance
for her two-tailed test so that the rejection
region was the right tail > 1.96 and the left
tail < -1.96.
Hypothesis Testing

Had she chosen  = .01 then the rejection


region would have been the right tail
beyond Z.005 = 2.575 and the left tail
below - Z.005 = -2.575. Since the value of
her observed statistic was Z  2.42 for
this choice of  she would have failed to
reject the null hypothesis while for  = .05 she
rejected it.
Hypothesis Testing

So rejection or not of the null hypothesis


depends crucially on choice of , the
probability of rejection given that the
hypothesis is true that one is willing to
tolerate. And the choice of  is
essentially by convention.
Hypothesis Testing

The conventional (some would say


arbitrary) nature of  choice and the
dependency of rejection or not on that
choice have led to a slightly different
approach to hypothesis testing. The first
two of the five steps are the same as
before.
Hypothesis Testing

The second approach to hypothesis


testing is called the p-value approach.
The first two of the five steps are the same as
before.

1. We formulate a null hypothesis


and an appropriate alternative hypothesis
from the language of the problem.
Hypothesis Testing

2. We specify a probability of type I error by


convention. Generally people choose  = .01
or .05. That is to say, we decide that we are
willing to tolerate a probability of .01 or .05 of
making a type I error, i.e., of rejecting the
null hypothesis when it is true.  is called the
significance level of the test.
Hypothesis Testing
Now we have steps that differ from those in the
traditional approach.

3’. We specify the test statistic.

4’. We calculate the value of the specified


test statistic from the data and then find
the tail probability value in the table that
corresponds to that value of the test
statistic. We call that value the p-value.
Hypothesis Testing

5’. We compare the p-value obtained in


step 4’ with the level of significance
specified in step 2. If the p-value is less
than or equal to the level of significance,
the null hypothesis must be rejected.
Hypothesis Testing

Example 4. Use the p-value approach to redo


example 3.
Solution. The first two steps are the same.
3’. Again, the test statistic is_
Z = (X - 0)/(/sq rt(n)).
4’. Calculating the value of that statistic gives
Z = (3.55 – 3.52)/(.07/sqrt(32))  2.42.
Hypothesis Testing

Now we go to the Z table and see what


probability values are left in the two tails
by the points Z = -2.42 and Z = 2.42. We
find that .5 - .4922 = .0078 is left in the
right tail by Z = 2.42 and the same
amount is left in the left tail by Z = -2.42.
So the p-value of the result is .0078 + .0078 =
.0156.
Hypothesis Testing

Since .0156 is < .05 the null hypothesis is


rejected at the .05 level but since .0156 is
> .01 the null hypothesis is not rejected at
the .01 level.
Thus the p-value is actually the lowest
level of significance at which the null
hypothesis would be rejected.
Hypothesis Testing

Researchers nowadays prefer the p-value


approach because they can publish the p-value
and let individual readers decide what
significance level they feel comfortable with and
so whether they want to reject the null
hypothesis or not. The p-value approach allows
researchers to avoid choosing a conventional
significance level for others.
Hypothesis Testing

Example 5. A horticulturist knows from


experience that the honeybees visiting her
orchard weigh .87 gram on the average.
Feeling that this year’s honeybees look bigger,
she decides to weigh a random sample of n =
50 of the bees all together and she gets an
average weight of .91 grams per bee with s =
.15 gram.
Hypothesis Testing

Using the .01 level of significance, what


can she conclude about her impressions
that this year’s bees are larger?
Hypothesis Testing

Solution. We use the p-value approach.


1. H0:  = .87
HA:  > .87
2.  = .01 _
3. Z = (X - 0)/(s/sqrt(n))
4. Z = (.91 - .87)/(.15/sqrt(50)) = 1.89.
Hypothesis Testing

Since we use a one-sided alternative we


use a one-tailed test. From the Z table
when Z = 1.89, there is .0294 in the right
tail so that is the p-value.
5. Since p = .0294 > .01 we cannot reject
the null hypothesis.
Hypothesis Testing

Example 6. A random sample of n = 12


graduates of a secretarial school typed on
the average _
X = 78.2 words per
minute with a standard deviation of s =
7.9 words per minute.
Hypothesis Testing

Assuming that such data can be looked


upon as a random sample from a normal
population, use the one-sample t test to
test the null hypothesis  = 80 words per
minute against the alternative hypothesis 
< 80 words per minute of this secretarial
school. Use the .05 level of significance.
Hypothesis Testing

Solution. We use the p-value approach.


1. H0:  = 80
HA:  < 80
2.  = .05 _
3. t= (X - 0)/(s/sqrt(n))
4. t =(78.2 - 80)/(7.9/sqrt(12)) = -.79.
Hypothesis Testing

Degrees of freedom are 12 – 1 = 11. So


from the t table we see that for t = -.79
with degrees of freedom 11 there is
probability > .10 in the left tail (see
drawing on board). So the p-value is > .10.
5. So since p >.10 > .05 we
can’t reject the null hypothesis.
Hypothesis Testing

So far we have tested hypotheses about


population means using either the Z or t
tables, depending on what information we
had. If n is large, it is also possible to test
hypotheses about population proportions
using the Z table.
Hypothesis Testing

Example 7. In a random sample of 600


cars making right turns at a certain
intersection, 157 pulled into the wrong
lane. Test the claim that 30% of all
drivers make this mistake, using
(a) .05 level of significance
(b) .01 level of significance
Hypothesis Testing

Solution. (a) We use the traditional approach


1. H0: p = .3 HA: p  .3
2.  = .05 
3. Z = (p - p0)/sqrt[p0 (1 – p0)/n]. Reject H0 if
z < -1.96 or z > 1.96
4. Z = [(157/600 - .3)]/sqrt[(.3)(.7)/600]  -2.05
5. Since –2.05 < -1.96 reject null hypothesis
Hypothesis Testing

Solution. (b) We use the traditional approach.


1. H0: p = .3 HA: p  .3
2.  = .01 
3. Z = (p - p0)/sqrt[p0 (1 – p0)/n]. Reject H0 if
z < -2.575 or z > 2.575
4. Z = [(157/600 - .3)]/sqrt[(.3)(.7)/600]  -2.05
5. Since –2.05 > -2.575 can’t reject null
hypothesis
Hypothesis Testing
Note. The standard normal distribution used in
the previous example, and in others of its type,
will give an accurate result only if
p0  3sqrt[p0(1 - p0)/n] does not include 0 or 1.
In that example n was 600 and p0 was .3 so
p0  3sqrt[p0(1 - p0)/n] = .3  3sqrt[(.3)(.7)/600] =
.3  .017 which does not include 0 or 1 so the
result reported in the example should be
accurate.
General Idea of
Hypothesis Testing
 Make an initial assumption.
 Collect evidence (data).
 Based on the available evidence, decide
whether or not the initial assumption is
reasonable.
Generally used tests for
hypotheses testing
 Parametric tests:
 Z scores
 t- test
 F test
 ANOVA
 Non parametric test
 Chi-square test
 Yate’s correction
 The median test
 The Mann Whitney test
 The Sign test

You might also like