Statistics
Module 1, Lecture 4
Inferential Statistics
MaryEllen Tancred, Ph.D.
Module 1 Learning Objectives
In this module, students will be able to:
1. Define basic terms in statistics including independent and dependent
variable; nominal, ordinal, interval and ratio measurements; and discrete
and continuous data.
2. Define descriptive statistics.
3. Describe and apply frequency distribution and measures of central
tendency.
4. Describe measures of variability.
5. Calculate the mean, variance, and standard deviation of a given data set.
6. Describe and apply bivariate descriptive statistics.
7. Define and describe the use of inferential statistics
and give examples of inferential statistics.
8. Define multivariate analysis.
Let’s Recall Some Key Concepts from
Lectures 1, 2, and 3
• In Lecture 1, the types of variables and levels of
measurement were introduced.
• In Lecture 2, descriptive statistics were introduced
with an emphasis on measures of central tendency
and variability that provided great ways to illustrate
characteristics of a distribution.
• In Lecture 3, the emphasis was on relationships
between variables, ending with the concept of
analyzing multiple variables.
• In this presentation, inferential statistics will be
introduced.
Inferential Statistics
•Recall that descriptive statistics are used to “describe”
characteristics of a sample. Inferential statistics are
techniques used to draw conclusions (inferences) about a
population by using sample data (characteristics) from
the population.
•This was not stressed in the previous presentations,
but in statistical literature, there are different symbols
used to indicate whether the characteristics reflect the
population or the sample.
Inferential Statistics
Examples of symbols distinguishing characteristics of
population versus sample:
Inferential Statistics
Examples of symbols distinguishing characteristics of
population versus sample:
Inferential Statistics
Examples of symbols distinguishing characteristics of
population versus sample:
Inferential Statistics
Examples of symbols distinguishing characteristics of
population versus sample:
Inferential Statistics
Examples of symbols distinguishing characteristics of
population versus sample:
Inferential Statistics
• Inferential statistics are widely used in
research and in comparison-of-methods
(COM) studies and are useful for drawing
conclusions about two sets of data. Examples
of data sets are two sets of means or two
sets of SDs.
• The distribution (shape) of the data helps to
determine what type of inferential statistics
to use.
Inferential Statistics
• For example, parametric tests involve
normally (Gaussian) distributed data and are
at least at the interval level of measurement.
• Examples of parametric tests include the t-
test (also known as the Student’s t-test) and
analysis of variance (ANOVA).
Inferential Statistics
Let’s look at some basic steps in performing an inferential statistical
analysis
[Link] the null hypothesis and research (alternative) hypothesis
[Link] a significance level associated with the null hypothesis
[Link] the appropriate test statistic
[Link] the test statistic value (obtained value)
[Link] the value needed to reject the null hypothesis using the
correct table of critical values for the statistic used
[Link] conclusions (inferences): Using the results, reject or fail to
reject the null hypothesis
Inferential Statistics
• Step 1. Generally speaking, inferential statistics
begins by making two hypotheses: the null
hypothesis and the research hypothesis.
o null hypothesis: states that the two sets of
data are the same. This is the evidence that
either supports or does not support the
research hypothesis.
o Research (alternative) hypothesis: states that
the two sets of data are different.
• Remember that we are estimating population
characteristics by examining sample characteristics.
Inferential Statistics
• Null Hypothesis: there will be no difference in the average score
of online MLS students and the average score of traditional face-
to-face MLS students on the ASCP BOC Exam (there is
essentially equality between variables):
H0 :μonline = μtraditional
• Research (Alternative) Hypothesis: the average score of online
MLS students is different from the average score of traditional
face-to-face MLS students on the ASCP BOC Exam (there is
inequality between variables):
H1 : μonline ≠ μtraditional
Inferential Statistics
Step 2. Select a significance level (α) associated with the
null hypothesis.
This is the amount of risk a researcher is willing to endure
that he/she will reject the null hypothesis when it is
actually true! This referred to as a Type I (α) error.
Risk is based on the concept of probability. Researchers
indicate the risk (probability) of not coming to the
correct conclusions in hypothesis testing. We refer to
this risk (probability) as “p.”
Inferential Statistics
Step 2. (continued)
Probabilities can range from 0 to 1. If the probability is at 0-this
means that there is no chance of an event occurring, and a
probability of 1 means that there is 100% probability that the event
will occur.
Commonly, researchers will indicate the probability of being wrong
about their hypothesis testing at the .05 or .01 level. At the .05
level, researchers are willing to be wrong 5 times out of 100 when
failing to reject the null hypothesis. Stated alternatively, the
researcher wants to draw the correct conclusions 95 times out of
100.
Likewise, a probability level of .01 means that the researcher wants
to come to the correct conclusions 99 times out of 100.
Inferential Statistics
Step 2. (continued)
Now that we know that significance level is based on
probability, how do we determine the significance level?
It is based on how much error is allowable which is usually
based on known practice. For example, studies in social
sciences are usually set at the .05 level, whereas medically
related studies have stricter levels of significance (.01, .001,
.0001, etc) since the consequences of being wrong could be
tragic (such as in studying a new drug and its effect).
Inferential Statistics
Step 2. (continued)
Let’s look at our example:
H0 :μonline = μtraditional
Since the consequences of this study do not pose
substantial risks, the significance level can be set at .05 for
demonstration purposes.
Inferential Statistics
Step 3. Select the appropriate test statistic
H0 :μonline = μtraditional
H1 : μonline ≠ μtraditional
For this study, we are going to look at the means of MLS
ASCP BOC scores of students in the online courses and MLS
ASCP BOC scores of students in the traditional face-to-
face courses.
Inferential Statistics: t-test
Step 4. Compute the test statistic value (obtained value)
H0 :μonline = μtraditional
H1 : μonline ≠ μtraditional
Let’s take a moment to review what we are doing:
We determined that the appropriate test is the t-test. The
t-test is a parametric test (which assumes normally
distributed data that is at least at the interval level of
measurement) and we will use it to compare 2 independent
means (ASCP BOC scores from MLS students who were
taught online and students who were taught traditionally) to
determine if we should accept or reject the null hypothesis.
Inferential Statistics: t-test
Step 4. Compute the test statistic value (obtained value)
Here is the formula for the t-test
Inferential Statistics: t-test
Step 4. Compute the test statistic value (obtained value)
Online Traditional
400 430
550 570
600 420
425 575
720 600
Online Mean Score = 539
630 500
Traditional Mean Score = 543 520 750
620 520
475 475
450 585
Inferential Statistics: t-test
Step 4. Compute the test statistic value (obtained value)
When the numbers are plugged
in, here are the results
t = .078
Online Mean Score = 539
Traditional Mean Score = 543
Inferential Statistics: t-test
Step 5. Determine the value needed to reject the null hypothesis
using the correct table of critical values for the statistic used
Online Mean Score = 539 What is needed to determine
the critical value
Traditional Mean Score = 543
dF = 18
p = < .05 (determined from
step 2)
dF is referred to as the degrees of freedom. It is the number of
values that are free to vary. It is simply (n1-1) + (n2-1). In our data
set, each group had 10. 10-1 =9. 9+9 = 18.
Inferential Statistics: t-test
Step 5. Determine the value needed to reject the null hypothesis
using the correct table of critical values for the statistic used
Online Mean Score = 539 What is needed to determine
Traditional Mean Score = 543 the critical value
dF = 18
p = < .05 (determined from
step 2)
t is referred to as the obtained value
dF is referred to as the degrees of freedom. It is the number of
values that are free to vary. It is simply (n1-1) + (n2-1). In our data
set, each group had 10. 10-1 =9. 9+9 = 18.
Inferential Statistics: t-test
Step 5. (continued) Here
are the values again:
t = .078, dF = 18
From the critical value
table (found in any
statistics book) for the t-
test: dF (degrees of
freedom) of 18 and a p-
value of .05, the critical
value is 2.10.
Table obtained from: [Link]
Inferential Statistics: t-test
Step 6. Draw conclusions (inferences): Using your results,
reject or fail to reject (accept) the null hypothesis
Online Mean Score = 539
Traditional Mean Score = 543
t = .078, dF = 18, critical value 2.10
If the obtained value (.078) is less than the critical value
(2.10), it is not extreme enough to attribute the differences in
mean ASCP BOC scores to anything other than chance.
(Differences occur for other reasons such as sampling error,
rounding error, etc).
Therefore, the mean ASCP BOC scores for online and
traditional students are comparable: meaning no significant
difference.
Inferential Statistics: t-test
Step 6. Draw conclusions (inferences): Using your results,
reject or fail to reject (accept) the null hypothesis
Online Mean Score = 539
Traditional Mean Score = 543
t = .078, dF = 18, critical value 2.10
H0 :μonline = μtraditional (equal) Accept
H1 : μonline ≠ μtraditional (inequal)
Since the statistical analysis provided evidence that there was
no significant difference between the means of online and
traditional students with regard to ASCP BOC scores, we
(accept) fail to reject the null hypothesis.
Inferential Statistics: t-test
Step 6 (continued) Here is a
visual explanation
t = .078, dF = 18
p- value of .05, the critical
value is 2.10.
95% of all Obtained Values- 5%
any difference due to
chance and is not
significant
H0 not rejected
Inferential Statistics: t-test
Just one more point about this t –test. Let’s say that this same
data set was run through a statistical software package to
calculate at the .05 level of significance (the same level used
previously). The results given would include the same values we
calculated manually
t = .078, dF = 18
And, in addition, it also provides a p-value for this analysis. In this
instance, the p-value calculated from this analysis is .938
What does this indicate? The p-value is > 0.05 (5%): It is .938
(93.8%). This is interpreted as: The probability is 93.8% that the
difference between the two means is due to chance; or there is a
probability of of 6.2% that the difference between means is due
to something other than chance; providing further evidence that
there is no significant difference between the means of online and
traditional students with regard to ASCP BOC scores!
Inferential Statistics: t-test
If this data was going to be published in a journal you would
see this
t(18) = .078, p = .938
This shows that the t statistic was used, there were 18
degrees of freedom, and the obtained value is .078. The p-
value for the analysis is >.05 (it is actually .938).
In conclusion, in statistical analysis: a p-value > .05 means that
the null hypothesis is the best explanation, and will not be
rejected. There is no significant difference.
A p-value obtained that is <.05 means that the differences
are significant.
Inferential Statistics
For demonstration purposes, let’s look at the results of a different dataset
comparing Online Mean Scores and Traditional Mean Scores
t = 3.01, dF = 18, critical value = 2.10, p = .04
H0 :μonline = μtraditional (equal) reject
H1 : μonline ≠ μtraditional (inequal)
In this scenario, the t-value (3.01) is greater than the critical value (which
shows significant difference); the differences in means of online and
traditional students with regard to ASCP BOC scores cannot be explained by
chance, therefore we reject the null hypothesis, indicating that the
differences are significant. The p-value for this analysis is <.05 (the value is
.04) which means the probability is less than 5% that the difference in mean
scores is due to chance alone.
Inferential Statistics-Quick Review
• Parametric tests involve normally (Gaussian)
distributed data and are at least at the
interval level of measurement.
• Examples of parametric tests include the t-
test (also known as the Student’s t-test) and
analysis of variance (ANOVA).
Inferential Statistics
Analysis of variance (ANOVA)
• differs from t-test because there are more than
2 means
• looks for an overall difference between groups
• If the differences in means are significant,
ANOVA is followed by another procedure to
determine all possible comparisons among the
means
Inferential Statistics: Non-Parametric Tests
• Non-Parametric tests involve data that are
not normally (Gaussian) distributed and do not
require the interval level of measurement.
• Examples of non-parametric tests include
o the chi-square test (X2)
o the Mann-Whitney test
Inferential Statistics: Non-Parametric Tests
• The chi-square (X2) test Below is the X2 formula:
o is used with two or more
groups where data is at X2 = Σ (0 - E)2
the nominal level and
calculating a mean is not
E
possible. It compares the
difference between what X2 = chi-square value
is expected to occur by Σ = summation
chance and what is actually O = observed frequency
observed. E = expected frequency
o A one-sample chi-square is
also referred to as
goodness-of-fit test.
Inferential Statistics: Non-Parametric Tests
Example of using the chi-square test (X2)
Suppose we wanted to survey MLS students to ask
them their learning modality preference. We want to
know if the number of participants are equally
distributed across all levels of preferences.
MLS Student Learning Modality Preferences
Prefer online Prefer Prefer blended Total
learning traditional (a combination
face-to face of online and
learning face-to-face)
learning
50 15 25 90
Inferential Statistics: Non-Parametric Tests
MLS Student Learning Modality Preferences
Online Traditional Blended Total
50 15 25 90
The null hypothesis in this situation would state
that there is no difference in frequency of
occurrences in each category
H0 : P1 = P2 = P3
“P” indicates the percentage of occurrences in any
individual category
Inferential Statistics: Non-Parametric Tests
MLS Student Learning Modality Preferences
Online Traditional Blended Total
50 15 25 90
The research hypothesis (alternative) in this
situation would state that there is a difference in
frequency of occurrences in each category
H1 : P1 ≠ P2 ≠ P3
Let us choose the level of significance as .05.
Inferential Statistics: Non-Parametric Tests
MLS Student Learning Modality Preferences
Online Traditional Blended Total
50 15 25 90
X2 = Σ (0 - E)2 H0 : P1 = P2 = P3
E
H1 : P1 ≠ P2 ≠ P3
X2 = chi-square value How do we know our expected frequency?
Σ = summation It is the observed frequency divided by the
O = observed frequency number of categories. In this case, it is
90/3 = 30
E = expected frequency
Inferential Statistics: Non-Parametric Tests
Category Observed Expected Difference (O-E) 2 ( O-E) 2
frequency frequency (E) (O-E) E
(O)
Online 50 30 20 400 13.33
Traditional 15 30 -15 225 7.50
Blended 25 30 -5 25 .833
Total 90 90 21.6
X2 = Σ (0 - E)2 = 21.6 Degrees of freedom are obtained by taking the
E number of categories minus 1
3-1 = 2
X2 = chi-square value
Σ = summation df = 2
O = observed frequency
E = expected frequency
Inferential Statistics: Non-Parametric Tests
Category Observed Expected Difference (O-E) 2 ( O-E) 2
frequency frequency (E) (O-E) E
(O)
Online 50 30 20 400 13.33
Traditional 15 30 -15 225 7.50
Blended 25 30 -5 25 .833
Total 90 90 21.6
X2 = Σ (0 - E)2 = 21.6 Degrees of freedom are obtained by taking the
E number of categories minus 1
3-1 = 2
X2 = chi-square value
Σ = summation df = 2
O = observed frequency
E = expected frequency
Inferential Statistics: Non-Parametric Tests
Category Observed Expected Difference (O-E) 2 ( O-E) 2
frequency frequency (E) (O-E) E
(O)
Online 50 30 20 400 13.33
Traditional 15 30 -15 225 7.50
Blended 25 30 -5 25 .833
Total 90 90 21.6
X2 = Σ (0 - E)2 = 21.6 Degrees of freedom are obtained by taking the
E number of categories minus 1
3-1 = 2
X2 = chi-square value
Σ = summation df = 2
O = observed frequency
E = expected frequency
Inferential Statistics: Non-Parametric Tests
Category Observed Expected Difference (O-E) 2 ( O-E) 2
frequency frequency (E) (O-E) E
(O)
Online 50 30 20 400 13.33
Traditional 15 30 -15 225 7.50
Blended 25 30 -5 25 .833
Total 90 90 21.6
X2 = Σ (0 - E)2 = 21.6 Degrees of freedom are obtained by taking the
E number of categories minus 1
3-1 = 2
X2 = chi-square value
Σ = summation df = 2
O = observed frequency
E = expected frequency
Inferential Statistics: Non-Parametric Tests
Category Observed Expected Difference (O-E) 2 ( O-E) 2
frequency frequency (E) (O-E) E
(O)
Online 50 30 20 400 13.33
Traditional 15 30 -15 225 7.50
Blended 25 30 -5 25 .833
Total 90 90 21.6
X2 = Σ (0 - E)2 = 21.6 Degrees of freedom are obtained by taking the
E number of categories minus 1
3-1 = 2
X2 = chi-square value
Σ = summation df = 2
O = observed frequency
E = expected frequency
Inferential Statistics: Non-Parametric Tests
Ddfdf = 2
X2 = 21.6
Level of significance set at .05
Critical value needed to reject
the null hypothesis is 5.991
Table retrieved from
[Link]/search?q=chi+square+distribution&source=lnms&tbm=isch&sa=X&ei=t
bVGVbH3LKu0sATcwIC4Dw&ved=0CAcQ_AUoAQ&biw=1299&bih=713#tbm=isch&q
=chi+square+distribution+table&revid=784073021&imgrc=JD49uYWAQybeIM%253A%3
Brp2R6ZhgVvzw6M%3Bhttp%253A%252F%[Link]%252Fdkernler%252F
statistics%252Fch09%252Fimages%252Fchi-square-
[Link]%3Bhttp%253A%252F%[Link]%252Fdkernler%252Fstatisti
cs%252Fch09%[Link]%3B540%3B540
Inferential Statistics: Non-Parametric Tests
Draw conclusions (inferences) based on the results
df = 2, X2 = 21.6, Level of significance set at .05
Critical value needed to reject the null hypothesis is 5.991
Using statistical software, the p-value of this analysis is .001.
The obtained value using the chi-square formula (21.6) is greater
than the critical value (5.991), and the p-value is <.05; means we can
reject the null hypothesis because the differences in the frequency
of occurrences in each category are not equally distributed and
results are significant. When it comes to learning modality
preferences, there is a difference in the frequency of students
selecting ONLINE, TRADITIONAL, and BLENDED, and this
difference is not attributed to chance.
H0 : P1 = P2 = P3 reject
H1 : P1 ≠ P2 ≠ P3
Inferential Statistics: Non-Parametric Tests
Draw conclusions based on the results
df = 2, X2 = 21.6, Level of significance set at .05
Critical value needed to reject the null hypothesis is 5.991
Another way to think of this:
There is a significant difference between the expected frequencies
and the observed frequencies in more than one of the categories
(Online, Traditional, Blended), and it is a difference that is not
random, but due to other reasons.
H0 : P1 = P2 = P3 reject
H1 : P1 ≠ P2 ≠ P3
Inferential Statistics: Non-Parametric Tests
The other non-parametric test mentioned previously
is the Mann-Whitney U test:
U = the Mann-Whitney test
N1 = sample size of group 1
N2 = sample size of group 2
Ri = Rank of the sample
Inferential Statistics: Non-Parametric Tests
The Mann-Whitney U test is used to determine if
two groups differ based on ranked scores:
• Not normally distributed
• Used to compare 2 independent samples (not
related, such as gender)
• The level of measurement is at least ordinal
(ranked values)
• The observations are independent (the same
participants cannot be in more than one group
being studied)
Inferential Statistics: Non-Parametric Tests
An example of how to use Mann-Whitney U test:
Perhaps a researcher would like to administer a
survey to determine if MLS student learning
modality preferences differ significantly by gender.
In this case, each modality (online, traditional,
blended) determined by the survey could be ranked
and mean rank for both groups (males and females) is
determined. Based on the calculations, the Mann-
Whitney U test would determine whether a
significant difference in learning modality
preference exists based on gender.