University of Science and Technology

of Southern Philippines

The Principles of
Inferential Statistics

ROSIE G. TAN, Ph.D.

Chairperson, DMEd
 Outline

• Population and sample

• Research Questions and Hypotheses
• Type I and II Error
• Effect Size
• Statistical Power
• Sample size

University of Science & Technology of Southern Philippines 2

 Population vs. Sample
• For example, if you wish to study the effect of part-time
contracts on the job commitment of women in Philippines who
are in their twenties. What will constitute your sample?

Population Sample
Job commitment of all Job commitment some
women who are 20 to women who are 20 to
29 years old, living in 29 years old, living in
the Philippines, and in the Philippines, and in
part-time jobs. part-time jobs

3
University of Science & Technology of Southern Philippines 3
 Population vs. Sample

Understanding From a statistical

point of view, the job-
commitment scores

Samples collected here

represent a sample.
The population of
scores you really
want to know about
is the job-
commitment scores
of all Filipino women
in their twenties who
are in part-time work.

4
University of Science & Technology of Southern Philippines 4
 Population vs. Sample

• In conducting research we take a sample of

numbers drawn from a population of numbers
and, generally speaking, it is the population of
numbers, and not the sample, that you really
want to know about.

5 5
University of Science & Technology of Southern Philippines 5
 Population vs. Sample

What is it that you will want to know about the

population?

6
6 6
University of Science & Technology of Southern Philippines 6
 Population vs. Sample
ANSWER:

▪ For categorical variables

- it is the proportions of cases that fall into certain categories.
(e.g. the proportion of all Filipinos who are male and female)

▪ For continuous variable

- it is the central tendency and the dispersion of scores. (e.g. the
mean or median job-satisfaction scores of Filipino people, and the
standard deviation of the job-satisfaction scores of Filipino people).

7
7 7
University of Science & Technology of Southern Philippines 7
 Population vs. Sample

8
8 8
University of Science & Technology of Southern Philippines 8
 Population vs. Sample
Example

If you say that the mean is 3.44 and standard

deviation is 1.00, then you are describing the
sample. But, if you will use the mean to make
inference about the population, then you have
entered into the realm of inferential statistics.

Making inference about the population:

You can say that you are 95% confident sure that the population mean is
actually not lower than 1.85 and not higher than 5.03. 9
9 9
University of Science & Technology of Southern Philippines 9 9
 Population vs. Sample
The equation for a Confidence Interval for the population mean when the
population standard deviation is unknown and the sample size is large (over 30)
is
𝑠
𝑥ҧ ±𝑧
𝑛
Where:
𝑥ҧ =sample mean 𝑠= standard deviation
𝑠
𝑛=number of sample =standard error
𝑛
𝑧 = critical value

10
1010
University of Science & Technology of Southern Philippines 10
 Population vs. Sample
For example, sample D of 30 babies born in 2013 had a mean weight of 7.19lbs with a
standard deviation of 1.4, the 95% Confidence Interval for the population mean of all
babies is:
1.4
7.19 ± 1.96 = 7.19 ± 0.5 = (6.7, 7.7)
30

Therefore, we would expect the population mean to be between 6.7 lbs and 7.7 lbs, or
we can be 95% confident that mean weight of babies born in 2013 is between 6.7 lbs
and 7.7 lbs.

How this value was obtain? Look at the z table or study the link:
https://www.statisticshowto.com/tables/z-table/

11
1111
University of Science & Technology of Southern Philippines 11 11
 Population vs. Sample

Question to ponder:

• What will happen to the size of confidence interval if

the sample size gets bigger?

• What will happen to the size of confidence interval if

the standard deviation gets smaller/bigger?

12
1212
University of Science & Technology of Southern Philippines 12 12
 Population vs. Sample
Consider the job-commitment scores of people in two organization. Both
organizations have the same mean, but differ in standard deviation. What do you
observe with the size of the confidence interval?

Which has a good estimates of

the population parameters?

The larger the sample, and the smaller

the amount of variation in the scores
within the sample, the more confident
you can be that your sample statistic (the
thing you know about) is a good estimate
of the population parameter (the thing
you want to know about).
13
1313
University of Science & Technology of Southern Philippines 13
 Population vs. Sample
Confidence intervals for categorical data
• Suppose you wish to know the proportion of women in a large
organization who smoke. You take a sample of 20 women, and find
that eight of them smoke.

• This suggests that 40 per cent of women in the organization are

smokers.

• However, because you have only a sample of women, you can’t be

sure that the proportion of women in the organization as a whole who
smoke is exactly 40 per cent.

14
1414
University of Science & Technology of Southern Philippines 14
 Population vs. Sample
Confidence intervals for categorical data
• To obtain the confidence interval for this,

𝑝ො 1−𝑝ො
confidence interval for p = 𝑝Ƹ ± 𝑧 𝑛
Where:
𝑝Ƹ =sample proportion
𝑛=number of sample
𝑧 =z-value
𝑝ො 1−𝑝ො
=standard error
𝑛

15
1515
University of Science & Technology of Southern Philippines 15
 Population vs. Sample
Confidence intervals for categorical data
• If we want 95% confidence interval for the example above ( the
smokers), then
.40 1 − .40
. 40 ± 1.96 = (.19, .61) Is this a good estimates of
40 the population parameter?

Thus, we can be 95% confident that the actual number of

women who smoke in the organization is no less than 19 per cent
and no more than 61 per cent.

16
1616
University of Science & Technology of Southern Philippines 16
 Population vs. Sample

QUESTIONS ABOUT INFERENTIAL STATISTICS

1. How you can possibly know for sure that the mean of the sample
is the same as the mean of the population?

2. How big your statistical sample needs to be before you know

that it is a perfect estimate of the population?

17
1717
University of Science & Technology of Southern Philippines 17 17
 Population vs. Sample
• The answer to the first question is that you never know with
certainty.

• The best you can hope for is that you know the probability that
your estimate of a sample statistic (such as the mean of a
sample) is within a certain range of the mean of a population
parameter (such as the mean of the population), like the
example above.

18
1818
University of Science & Technology of Southern Philippines 18
 Population vs. Sample

• The second question raised is not a very good question to

ask since the conclusions drawn from inferential statistical
methods are not certainties and truths but estimates and
probabilities (Dewberry, 2004).

19
1919
University of Science & Technology of Southern Philippines 19
 Population vs. Sample
• However, other authors proposed a method for calculating the sample size.
𝑁
1) the sloven’s formula: 𝑛 = 2 1+𝑁𝑒
Where:
n refers to the number of samples or sample size
N refers to the population
e refers to margin of error
is used to determine the sample size

20
2020
University of Science & Technology of Southern Philippines 20
 Population vs. Sample
2) Cochran’s Sample Size Formula:
Where:
𝒆 is the desired level of precision (i.e. the margin of error),
𝒑 is the (estimated) proportion of the population which has
the attribute in question,
𝒒 is 𝟏 – 𝒑.
The z-value is found in a Z table.

• Cohen (1988) said that the sample size depends on the statistics, which will
be introduced later.

21
2121
University of Science & Technology of Southern Philippines 21 21
 Hypotheses

• When people use inferential statistics they are usually

concerned with testing hypotheses.

• That is, they want to know whether something that they are
interested in is true or not.

22
2222
University of Science & Technology of Southern Philippines 22
 Research Question and Hypotheses
Research questions can be defined as ‘questions in quantitative
or qualitative research that narrow the purpose statement to
specific questions that researchers seek to answer’.

Hypotheses are ‘statements in quantitative research in which

the investigator makes a conjecture about the outcome of a
relationship among attributes or characteristics’.
- It is a statement about the nature of the population, not
the nature of the sample.

23
2323
University of Science & Technology of Southern Philippines 23
 Research Question and Hypotheses

Two types of hypotheses:

❑ Research/Alternative hypothesis (𝑯𝟏 )
❑ Null hypothesis (𝑯𝟎 )

24
2424
University of Science & Technology of Southern Philippines 24
 Research Question and Hypotheses
Research hypothesis (𝑯𝟏 )
- It is a tentative prediction about how change in a
variable(s) will explain or cause changes in another
variable(s).

Null hypothesis (𝑯𝟎 )

- It is the antithesis to the research hypothesis and
postulates that there is no difference or relationship
between the variables under study.
-It is tested with inferential statistics

25
2525 25
University of Science & Technology of Southern Philippines 25
 Research Question and Hypotheses
Examples of 𝑯𝟏 and 𝑯𝟎 :
𝑯𝟏 : There will be a statistically significant difference in high school
dropout rates of students who use drugs and students who do not
use drugs.

𝑯𝟎 : There will be no statistically significant difference in high school

dropout rates of students who use drugs and students who do not
use drugs

26262626
University of Science & Technology of Southern Philippines 26
26 26
 Research Question and Hypotheses

A well formulated research question, according to

Bartos (1992):

• be stated clearly and in the form of a question

• be testable (i.e. possible to collect data to answer the question)
• not pose an ethical or moral problem for implementation
• be specific and restricted in scope (your aim is not to solve the
world’s problems)
• identify exactly what is to be solved.

University of Science & Technology of Southern Philippines 27

 Research Question and Hypotheses

Example of a Poorly Formulated Research Question:

What effect does social media have on people’s minds?

Example of a Well Formulated Research Questions:

What effect does daily use of Twitter have on the

attention span of under-16s?
It is not specific enough: what type of
social media? Which people? What kind
It is specific defines its concepts more clearly. It is
of effects?
researchable through qualitative and quantitative
data collection.

University of Science & Technology of Southern Philippines 28

 Research Question and Hypotheses

Example of a Poorly Formulated Research Question:

What is the effectiveness of parent education for parents
with problem children? Not specific

Example of a Well Formulated Research Questions:

What is the effect of the STEP parenting program on the
ability of a parents to use natural, logical consequences (as
opposed to punishment) with their child who has been
diagnosed with Bipolar disorder?
specific
Testable through experiment
University of Science & Technology of Southern Philippines 29
 Research Question and Hypotheses

Inferential Questions
• examine relationships between different variables;
• try to predict scores on one variable from the scores on one or more
others;
• compare central tendency of a variable for two or group of people or.

Broadly speaking, Inferential Questions focused on two sorts of

relationships: associations and differences.

University of Science & Technology of Southern Philippines 30

 Research Question and Hypotheses
Which of the following research questions needs
hypothesis testing?

RQ 1. Does critical thinking ability relate to student

achievement?

RQ 2. How do the students rate on critical thinking skills?

University of Science & Technology of Southern Philippines 31

 Research Question and Hypotheses

Answer: RQ 1
(An inferential question relating two variables)

University of Science & Technology of Southern Philippines 32

 Research Question and Hypotheses

Which research questions below needs hypothesis?

RQ 3. What are the student’s achievement level in science

classes?

RQ 4 . Do students perform better on the analytical portion of

the USTP admission test than on the verbal portion?

University of Science & Technology of Southern Philippines 33

 Research Question and Hypotheses

Answer: RQ 4
(An inferential question comparing students performance in two set of test )

University of Science & Technology of Southern Philippines 34

 Research Question and Hypotheses

State the research and null hypotheses of the

following:
RQ 1. Does critical thinking ability relate to student
achievement?

RQ 4 . Do students perform better on the analytical portion of

the USTP admission test than on the verbal portion?

RQ 5. Is there a significant difference between male and female

math achievement?

University of Science & Technology of Southern Philippines 35

 Research Question and Hypotheses
Answers:
RQ 1. Does critical thinking ability relate to student achievement?
(Directional/one-tailed)

𝐻1 : Critical thinking ability does relate to student achievement.

𝐻0 : Critical thinking ability does not relate to student achievement

University of Science & Technology of Southern Philippines 36

 Research Question and Hypotheses
Answers:
RQ 4 . Do students perform better on the analytical portion of the
USTP admission test than on the verbal portion?
(Directional/one-tailed)

𝐻1 : Students do score statistically significantly better on the analytical

portion of the USTP admission test than on the verbal portion.

𝐻0 : Students do not score statistically significantly better on the

analytical portion of the USTP admission test than on the verbal
portion.
University of Science & Technology of Southern Philippines 37
 Research Question and Hypotheses
Answers:
RQ 5. Is there a significant difference between male and female math
achievement? (Non directional/ two-tailed)

𝐻1 : There is statistically significant difference between male and

female math achievement. .

𝐻0 : There is no statistically significant difference between male and

female math achievement. .

University of Science & Technology of Southern Philippines 38

 Research Question and Hypotheses
There are four steps to hypothesis testing:
Information derived from Rovai, et al. (2012)

Step 1. State the null and research hypothesis as just discussed.

Step 2. Choose a statistical significance level or alpha, say .05, .04, etc.

An alpha is always set ahead of time.

University of Science & Technology of Southern Philippines 39

 Research Question and Hypotheses
Step 3. Choose and carry out the appropriate statistical test, which will discussed in
the next section.
Step 4. Make a decision regarding hypothesis (i.e. reject or fail to reject the null
hypothesis).

Decision Guide:
▪ If the sig.(p-value) for the analysis is lower than the significance level or
alpha, then the results is statistically significant, and so reject the null
hypothesis.
▪ If the p-value for the analysis is equal or more than the significance level or
alpha, then the results is not statistically significant, and so there is no enough
evidence to reject null hypothesis.

University of Science & Technology of Southern Philippines 40

 Research Question and Hypotheses
Example 1:
Suppose you set alpha level equals .05. If the sig(p-value)
is .02, then what is your decision for the null hypothesis?

For this example, we can consider the results to be

statistically significant, and so the null hypothesis is rejected.

University of Science & Technology of Southern Philippines 41

 Research Question and Hypotheses
Example 2:
Suppose you set alpha level equals .05. If the sig(p-value)
is .17, then then what is your decision for the null hypothesis?

For this example we can consider the results to be not

statistically significant, and so the null hypothesis is not rejected.

University of Science & Technology of Southern Philippines 42

 Type I and II Error

University of Science & Technology of Southern Philippines 43

 Type I and II Error

• Note that in step two, we set a significance level, say .05. What does it
mean?

✓ It means that if we reject the null hypothesis, we are willing to

accept that there is no more than 5% chance that we made
the wrong conclusion.

✓ In other words, with a .05 significance level, we want to be at least

95% confident that if we reject the null hypothesis we have made
the correct decision.

* This means that our decision is accompanied

by a certain degree of error, Type I or Type II.

University of Science & Technology of Southern Philippines 44

 Type I and II Error

University of Science & Technology of Southern Philippines 45

 Type I and II Error

• The choice of .05 as the criterion for an acceptable risk of Type I

error (𝜶), that is the danger of rejecting the null hypothesis when it
should be accepted, is a tradition or convention, based on
suggestions made by Sir Ronald Fisher..

• If we reduce the level required for statistical significance from .05 to,
say, .001, the likelihood of making a Type I error is reduced (Why?).
The trouble is that if we do this the probability of making a Type II
error (𝜷), that is accepting the null hypothesis when it is actually
false, increases (Why?).

University of Science & Technology of Southern Philippines 46

 Type I and II Error

• Therefore, the choice of .05 level of significance represents a

reasonable trade-off between

(1) the danger of making a Type I error;and

(2) the danger of making a Type II error.

University of Science & Technology of Southern Philippines 47

 Type I and II Error
REMEMBER:
Type I Error : If the researcher says that there is statistically
significant difference and rejects the null hypothesis
when the null hypothesis is true.

Type II Error: If the researcher says that there is no statistically

significant difference and fails to reject the null
hypothesis when the null hypothesis is false.

University of Science & Technology of Southern Philippines 48

 Type I and II Error
The null hypothesis is rejected

Example of Type I Error:

• A researcher concludes that the STEP parenting program is more
effective than the parent support group.
• Based on these results, a counseling center revises their parenting
program and spends a significant amount of money for the STEP
program. After the expenditure of time and money, the parents do not
appear to show improvement in their implementation of logical, natural
consequences.
• Subsequent research does not reveal results demonstrated in the
original study. Although the ultimate truth about the null hypothesis is not
known, evidence supports that a Type I error likely occurred in the
original study.
University of Science & Technology of Southern Philippines 49
 EFFECT SIZE

University of Science & Technology of Southern Philippines 50

 EFFECT SIZE
A large study compared two weight loss methods with 13,000
participants in a control intervention group and 13,000 participants in an
experimental intervention group. The control intervention used scientifically
backed methods for weight loss, while the experimental intervention group
used a new app-based method. After six months, the mean weight loss
(kg) for the experimental intervention group (M = 10.6, SD = 6.7) was
marginally higher than the mean weight loss for the control intervention
group (M = 10.5, SD = 6.8). These results were statistically significant (p =
.01).

Will you buy the new weight loss product?

University of Science & Technology of Southern Philippines 51

 EFFECT SIZE
A study compared two methods of teaching. The control
group used the conventional method, while the experimental
intervention group used a new teaching method. After two
months, the achievement mean score for the experimental
group (M = 25, SD = 8) was statistically significantly higher
(p = .04) than the achievement mean score for the control
group (M = 24, SD = 7).

Will you adopt the new teaching method?

University of Science & Technology of Southern Philippines 52

 EFFECT SIZE

• An effect size is simply a measure of how big a

difference or an association is.

• Effect sizes look at the size of differences in central

tendency, or the strength of the associations.

• Effect size looks at practical significance

University of Science & Technology of Southern Philippines 53

 EFFECT SIZE
There are dozens of measures for effect sizes.The most common are:
▪ Partial eta squared indicates the proportion of variance in the DV
that is explained by the IV. The values range from 0 to 1 and are
interpreted as the chart below indicates.
▪ Cohen’s 𝒅 refers to the difference in groups based on standard
deviations. It is obtained by taking the difference between two
means and dividing by the standard deviation of the scores on which
they are based.
▪ Pearson’s 𝒓 refers to the strength and direction of the relationship
between variables. The correlation coefficient is an excellent
measure of effect size.

Cohen (1988) sets forth guidelines for interpreting effect size as shown in
Table below.
University of Science & Technology of Southern Philippines 54
 EFFECT SIZE

From practical point of

view, insubstantial and
uninteresting

University of Science & Technology of Southern Philippines 55

 Reporting the size of an effect

• For example, if the mean performance for women is 22.7, the

mean performance for men is 13.4, and the standard deviation of
the performance of all of the men and women is 14.5, the effect
size of this difference between the means is:

𝑑 = (𝑥1 - 𝑥ҧ2 )/ 𝑆𝐷 𝑝𝑜𝑜𝑙𝑒𝑑

Why 𝑑? = (22.7 − 13.4)/14.5 = 0.64

• So, in this case, the difference between the performance of men

and women has a medium to large effect size.

University of Science & Technology of Southern Philippines 56

 EFFECT SIZE

Can you obtain statistical significance if the

effect size you have found is small?

University of Science & Technology of Southern Philippines 57

 EFFECT SIZE
• The answer is, yes. This is because the chance of
obtaining a statistically significant result depends not just
on the population effect size, but also on the sample size
you have used in your research (and, as mentioned earlier,
on the significance level adopted).

University of Science & Technology of Southern Philippines 58

 EFFECT SIZE

• With two very small samples of the salaries of men and women, you
are unlikely to find that the difference between the means of the two
samples is statistically significant even if there is actually a very big
difference in the mean salaries of men and women (i.e. a big population
effect size).

• And conversely, if you take two very large samples of men and women,
you are likely to find that even if the difference in the salaries of men
and women is very small indeed, you will still obtain a statistically
significant result.

University of Science & Technology of Southern Philippines 59

 EFFECT SIZE
Example:
Someone might try to sell your organization a psychometric test to
use in the selection of employee. They might have data to show that the
correlation between the score that people obtain on the test and their job
performance is statistically significant. This would imply that the test is
worth buying since it will helpfully predict job applicants who will perform
well or perform badly at work. However, if the sample size used to examine
the association between the test scores and job performance was very
large (say 1,000 or more) this would be likely to be statistically significant
even if the actual correlation between the test scores and job performance
is very small (say about 0.1).

University of Science & Technology of Southern Philippines 60

 How do we know that we made
a correct decision, reject or do
not reject the null hypothesis?

University of Science & Technology of Southern Philippines 61

 Statistical Power

University of Science & Technology of Southern Philippines 62

 Statistical Power

• The statistical power is 1- β,

where Beta( β) is the probability of committing type II error, that you
won’t reject the null hypothesis when it is false.

• Power is your probability of not making a type II error.

University of Science & Technology of Southern Philippines 63

 Statistical Power
• Statistical power is the ability of statistical tests to demonstrate that a
difference or association of a particular magnitude would be unlikely if the
null hypothesis is true, under circumstances in which the null hypothesis is
actually false.

• Similarly, the more statistical power you have the more likely you are to
obtain evidence for the alternative hypothesis if it is true.
• In other words, a high statistical power means that the results are likely
valid. As the power increases, the probability of making a Type II
error decreases.
• A low statistical power means that the results are questionable.

• Statistical power is comparable to Sherlock Holmes’ magnifying glass; the

more powerful the glass the more likely he is to detect fingerprints on the
murder weapon if they are there.
University of Science & Technology of Southern Philippines 64
 Statistical Power

• Statistical power depends on 3 factors:

1. the significance level adopted in the study (e.g. is

the .05 or the .01 level of significance adopted?),
2. the effect size of the association between the
populations (or difference in the means of the
populations) you are trying to detect; and
3. most important of all, the sample size used in the
study.

University of Science & Technology of Southern Philippines 65

 Statistical Power

What then, are the implications of increasing/decreasing

significance level, effect size, sample size ?

University of Science & Technology of Southern Philippines 66

 Statistical Power

• Perhaps the most important implication is that it is possible to identify a variety

of strategies for increasing statistical power by.

1. adopting the .05 significance level, as is the convention because as

discussed earlier, as the significance level is reduced, the probability
of making a Type I error reduces.

2. increase the population effect size.

3. increasing your sample size.

• Finally, we can effectively answer one of the most common questions that
researchers ask: ‘How many people (participants/respondents) do I need to
include in my study?’

University of Science & Technology of Southern Philippines 67

 Statistical Power

How to work out the sample size?

University of Science & Technology of Southern Philippines 68

 Statistical Power

• In order to work out the sample size you will require in your study you need to
✓ adopting the .05 significance level
✓ assuming the 80% the statistical power, and
✓ estimate the population effect size.

Three ways to estimate the effect size before you run your research.

1) carry out a small-scale pilot study, using the same measures, types of
people, and testing conditions that you intend to use in the study proper,
and use the results you obtain from this to estimate the effect size.

University of Science & Technology of Southern Philippines 69

 Statistical Power
Example:
You are interested in whether there is an association between the salaries and
job satisfaction. To estimate the effect size between these variables you carry out a small-
scale pilot study, measuring the salaries and job satisfaction of 30 employees. You obtain
a correlation coefficient of .24. You use the correlation of .24 as an estimate of the effect
size of the population when calculating the sample size you will need in your subsequent
full-scale study. Note that it is essential that the data obtained in the pilot study are not
included in the analysis of the data from the full-scale study.

University of Science & Technology of Southern Philippines 70

 Statistical Power

2) carry out a review of published studies which are similar

to the one you are proposing to carry out (i.e. ones
which used the same measurement scales you intend to
use, were carried out in an environment similar to the
one you plan to use in your own study, and involved
similar people) and to use the effect size found in these
as a guide to what you can expect in your own study.

University of Science & Technology of Southern Philippines 71

 Statistical Power
Example:
You wish to investigate whether the conscientiousness of train
drivers, as measured with a personality questionnaire, is associated with
the extent to which they are punctual. You examine the literature and find
two studies that seem relevant. One examined the relationship between the
conscientiousness of lorry drivers and the number of times they arrived late
for work over a six-month period. The other was concerned with the
relationship between the conscientiousness of junior managers and, again,
measured the number of times they arrived late for work over a six-month
period. The correlation between conscientiousness and punctuality was .33
in the first study and .53 in the second study. You average these and
predict an effect size of .43 in your own study.

University of Science & Technology of Southern Philippines 72

 Statistical Power

3) adopt a medium effect size.

For example, we can say that for a difference between two groups, if you expect
a medium population effect size of .3, you adopt the .05 significance level, and you
The ideal
effect size
are prepared to accept that you will have an 80 per cent chance of rejecting the null
to assume. hypothesis if it is false (i.e. a power of 0.80 or 80 per cent), you will require a sample
size of exactly 65 people.

The ideal power is .80.

I will not discussed how 65 people was obtained because the formula for
calculating the sample size depends on the type of statistical tools.
Cohen (1988) provide the approximate sample size required in each
statistical tools, which will be presented as we discussed each type of
statistical tool.

Sample size required in independent t-test by Cohen (1988)

University of Science & Technology of Southern Philippines 73