Inferential Statistics

How to choose the appropriate test to answer your research

question?

Presented by
 Learning Outcomes
• Identify inferential statistical analyses.
• Compare statistical significance and clinical importance of results.
• Critically appraise the data analysis and statistical tests reported in
published studies.
 Statistical Analysis
• Descriptive statistics: summary statistics that allow the researcher
to organize data in ways that give meaning and facilitate insight.
• Calculated to describe the sample and key variables.
• Inferential statistics: designed to address objectives, questions, and
hypotheses in studies to allow inference/generalization from the
study sample to the target population.
• Conducted to identify relationships, examine predictions, and
determine group differences in studies.
 Appropriateness of inferential statistics in studies
• Why do we carry inferential statistical analysis?
• Examine relationship
• Make prediction
• Determine causality or differences
• Inferential statistics used determined by:
• Purpose
• Hypotheses, questions, or objective
• Design (# of groups, relationship between groups, and # of
data measurement points)
• Level of measurement of the variables (nominal, ordinal,
interval, or ratio)
 Inferential Statistics
Use of Inferential Statistical technique
statistics

Examine relationships ➢Correlation

✓ Person product-moment correlation
✓ Spearman Roh and Kendall’s tau
Predict outcomes ➢Regression analysis
Examine differences ➢Chi-Square Test of
Independence/Fisher's exact
➢t-Test
➢Analysis of Variance (ANOVA) - F-Test
 Examine relationship – Correlation

• Person product-moment correlation:

• Determine relationship between variables
• Correlation coefficient ( r )
• Values -1 to + 1
 Examine relationship – Correlation
• Person product-moment correlation
 Examine relationship – Correlation
• Person product-moment correlation
 Examine differences
•Parametric vs nonparametric analysis
Variables Analysis
• Normally distributed Parametric
• Interval or ratio measurement
level
• Nominal or ordinal measurement Nonparametric
level
• Not normally distributed
 Examine differences -Independent vs
dependent groups

Participants Group Type

- The selection of one subject is unrelated to the Independent
selection of other subjects (e.g., Males and
females, diagnosed and not diagnosed,
employed and non-employed)
- Subjects or observations selected for data Dependent
collection are related in some way to the (paired/matched
selection of other subjects or observations (e.g., paired)
Husbands and their wives, Mothers and their
children, twins, children and their grandparents)
 Examine differences -Independent vs
dependent groups
Group Type Parametric test (Ratio/interval
level and data is normally
distributed)
The mean of two Independent t-test
independent groups
The mean of Paired t-test
two dependent/paired
(matched) groups

The mean of 3 One-way ANOVA

or more independent groups
 Examine Differences Using Parametric versus Non-
Parametric Tests-Statistical decision tree
Group Type Dependent Independent Parametric test (data is Non-parametric test
(outcome) variable (explanatory) variable normally distributed) (ordinal/ skewed
data)

The mean of two Scale (Ratio/interval) Nominal (Binary) Independent t-test Mann-Whitney test
INDEPENDENT groups
The mean difference Scale (Ratio/interval) Time/ Condition Paired t-test Wilcoxon signed rank
between two paired variable test
(matched) groups
The mean difference Scale (Ratio/interval) Nominal One-way ANOVA Kruskal-Wallis test
among 3 or
more INDEPENDENT
groups
The mean difference Scale (Ratio/interval) Time/ condition Repeated measures Friedman test
among 3 or variable ANOVA
more measurements on
the same subject
 Examine Differences Using Non-
Parametric Tests

Group Type Non-parametric test

(ordinal/Nominal/ skewed data)

Two independent groups Chi-squared test

(Categories/Cells are 5 and
more)
Fisher’s exact (Categories/Cells
less than 5)
 Examine Differences Using Non-
Parametric Tests- Statistical decision tree

Dependent (outcome) Independent (explanatory) Non-parametric test

Comparing: Difference
variable variable (ordinal/ skewed data)

Chi-squared test
Two INDEPENDENT groups Nominal (Binary) Nominal (Binary)
Fisher’s exact
Two paired (matched)
Nominal (Binary) Nominal (Binary) McNemar test
groups
Three or more independent Chi-squared test
Nominal/Ordinal Nominal/Ordinal
groups Fisher’s exact

Three or more paired

Nominal/Ordinal Nominal/Ordinal Cochran’s Q statistics
(matched) groups
 Example 1
A researcher wants to assess the effectiveness of Corna virus awareness workshop on first
year MSNIC students' knowledge. She assigned students randomly into two groups
(experimental and control groups). Prior to giving the workshop to the experimental group,
the researcher want to know the following:
• Is there a difference in knowledge level between experimental group and control
group?
• Knowledge level is measured with knowledge level scale
Which test should I use??

Group Number Mean

Knowledge Score
Control 40 10.137

Experimental 40 10.207
 Example 1 – Answer

• To measure the difference in knowledge level between experimental group and

control group we will use independent t-test why?
• We have two independent/unrelated groups and we have mean score for
each group that' why we use independent t-test
Group Number Mean t-stat P- value
Control
40 10.137 1.4401 0.089

Experimental 40 10.207

There is no significant difference between the control and experimental groups in

knowledge level at the significant level of 0.05.
 Example 2
• A group of husbands and wives were enrolled in a study and asked to rate their
partner's degree of support. You want to assess the difference between
husbands and wives based on their mean score of support.
• Partner support level was measured with support scale

Which test should I use??

Group Number Mean
Support Score
Husbands 120 26.5

Waives 120 22.8

 Example 2 – Answer
• To measure the difference in support level between husbands and wives we will
use Paired t-test why?
• We have two related groups (husband and their wives)and we have mean
score for each group that' why we use paired t-test.
Group Number Mean t-stat P- value

Husbands 120 26.5 1.657 0.035

Waives 120 22.8

There is a significant difference between support level in husbands and wives at

the significant level of 0.05.
 Example 3
• We want to see if female students performs better than male
students in an undergraduate level biostatistics course. After
a midterm exam, we want to see if is there a difference between
male and female students in passing the biostatistics midterm
exam?
Which test to use??
Group Passed Failed

Females 30 10

Males 18 25
Total 48 35
 Example 3 – Answer

To measure the difference between female and male students in passing the biostatistics
midterm exam, we will use Chi-squared test why?
• We have two unrelated groups. The passing of the midterm exam is a nominal
variable (passed/failed) and we have number/count of those who passed the
midterm and those who didn't for each group that's why we use Chi-squared test.

Group passed Failed

Females 30 10
Males 18 25
Total 48 35
 Example 3 – Answer
What if the number in the cells are different from the previous example and we
still wants to measure the difference between female and male students in
passing the biostatistics midterm exam?
• We will use Fisher’s exact, why??
As you can see, one of the cell has a count less than 5, that is why we should
use Fisher’s exact test in this case.

Group passed Failed

Females 44 10
Males 4 25
Total 48 35
 Example 4 with answer
• We want to know the difference in cigarette smoking based on gender. We
asked 193 males and 209 female participants to indicate their cigarette smoking
status.
• Since we want to know the difference in numbers between unrelated groups
and smoking status is measured by a nominal scale, we use Chi-squared test.
 Summary
Descriptive Statistics
Level of measurement Statistical Tests
Used
Nominal
Non-parametric
Gender, Marital Status,
Mode Chi Square
Residence, Nationality, Medical
Fisher Exact Test
Diagnoses​
Ordinal
Non-parametric
Likert scale, levels of coping,
Mode and Median Chi Square
educational level, severity of
Fisher Exact Test
illness, dress sizes, age groups
Parametric
Mode, Median and Mean,
Interval Independent t-test ​
variance, standard
Temperature Paired t-test
deviation, range
ANOVA
Parametric
Ratio​ Mode, Median and Mean,
Independent t-test ​
Weight, height, age in years, variance, standard
Paired t-test
number of children​ deviation, range
ANOVA
 Other Important Tests/analysis
• Two-way ANOVA : involves two distinct IVs (each has two or more independent
groups) and One DV (interval or ratio)
• Analysis of Covariance (ANCOVA): Used to compare between two or more group
means after (statistically) removing the effect of a variable called a covariate (one or
more). 1 DV (Interval of Ratio), 1 IV (nominal or ratio) (e.g., experimental versus
control), and 1 CV (Interval or Ratio) (or dichotomous with near 50-50 split).
Dummy-variable covariates can be included as additional independent variables.
• Multivariate ANOVA (MANOVA): Used to test mean group differences on two or
more related dependent variables
• Multivariate Analysis of Covariance MANCOVA: Allows us to examine differences on
more than one DV, while controlling for covariate(s).
• Mixed Design RM-ANOVA: Two or more groups are measured at multiple points in
time (three or more). When there are only two data collection points (pre and
posttest), ANCOVA is often used; the group means are equal after removing the
effect of pretest scores.
 Other Important Tests/analysis

• Multiple Linear Regression Analysis: It is used when we want to predict the value of
a variable (DV at interval or ratio level) based on the value of independent variables
(IVs regardless the level of measurement). Categorical IVs should be dummy coded
to generate C-1 variables.
• Binary Logistic regression analysis: Analyzes the relationship between multiple IVs
(regardless the level of measurement and a single DV (categorical of two groups
only). Categorical IVs do not require dummy coding; the SPSS will do it.
• Multinomial Logistic regression analysis: Analyzes the relationship between multiple
IVs (regardless the level of measurement and a single DV (categorical of three or
more groups). Categorical IVs do not require dummy coding; the SPSS will do it.
Thank you,
Questions???