ANOVA

ANOVA

ANOVA (Analysis of Variance) is a statistical method used to test if there are statistically significant differences among the means of three or more
groups. It helps researchers determine whether observed differences in sample means across various groups can be attributed to true differences in the
population or if they are likely due to random variation.
ANOVA Assumptions

1.Independence of Observations: Each sample is independent of the others.

2.Normality: The data in each group are approximately normally distributed.
3.Homogeneity of Variances: The variances in each group are similar.
4.Scale level: The scale level of the dependent variable must be metric, whereas the independent variable must be nominally scaled.
Assuring Independence in Experimental Design

• Randomly assigning subjects to treatment groups.

• Ensuring that measurements are taken in a way that one measurement doesn’t influence another.
• Structuring data collection to avoid spillover effects (where one participant’s response affects another’s).
Normality Test

Shapiro-Wilk Test:
This is one of the most widely used normality tests.

• Null Hypothesis: The data is normally distributed.

• Interpretation: If the p-value is less than the chosen significance level (e.g., 0.05), we reject the null hypothesis and conclude that the data is not
normally distributed.
Normality Test

Shapiro-Wilk Test:
This is one of the most widely used normality tests.

• Null Hypothesis: The data is normally distributed.

• Interpretation: If the p-value is less than the chosen significance level (e.g., 0.05), we reject the null hypothesis and conclude that the data is not
normally distributed.
Homogeneity of Variances

• homogeneity of variances (or homoscedasticity) is an assumption that the variances within each group or treatment level are equal.
Source: https://stats.stackexchange.com/questions/76151/what-is-an-intuitive-explanation-of-why-we-want-homoskedasticity-in-aregression
Test for homogeneity of variances

• Levene's Test: A common test for checking homogeneity of variances, where the null hypothesis assumes that the variances are equal across
groups.

• Bartlett's Test: Another test that assumes normally distributed data; it’s more sensitive to departures from normality than Levene’s Test.
• Brown-Forsythe Test: A variation of Levene’s Test, which is also commonly used if the data does not follow a normal distribution.
Levene's Test

• Null Hypothesis: The variances are equal across groups.

• Alternative Hypothesis: At least one group has a significantly different variance.
• If the p-value from Levene's Test is greater than the chosen significance level (e.g., 0.05), we fail to reject the null hypothesis, meaning we assume
equal variances.

• If the p-value is less than the significance level, we reject the null hypothesis, indicating that there is a significant difference in variances across
groups.
Key Concepts

• Factors: Independent variables that define the groups or treatments. For example, in a study of plant growth, "fertilizer type" could be a factor.

• Levels: The different categories or values within a factor. For instance, if "fertilizer type" is the factor, the levels could be
"organic" and "chemical."
Key Concepts

• Hypotheses in ANOVA:
 Null Hypothesis (H₀): Assumes there is no difference in the group means (all group means are equal).
Alternative Hypothesis (H₁): At least one group mean is different from the others.
Key Concepts

• Within-Group Variability: The variation within each group or treatment condition, reflecting individual differences and random error.

• Between-Group Variability: The variation between the group means, which may be due to the effect of the factor(s) being studied.
Source: https://camilo-mora.github.io/GEO380/anova-reasoning.html
Types of ANOVA

• One-Way ANOVA: Used when there is a single factor with multiple levels, and the objective is to determine if the means of the different levels are
different.

• Two-Way ANOVA: Used when there are two factors, and it can test:

• The main effect of each factor.

• The interaction effect between the two factors.

• Repeated Measures ANOVA: Used when the same participants are measured multiple times under different conditions (e.g., time points or
treatments).

• MANOVA (Multivariate ANOVA): Extends ANOVA to multiple dependent variables, allowing researchers to see if the combination of means differs
across groups.
How ANOVA Works

• ANOVA calculates two types of variances:

• Between-Groups Variance: Measures the difference between the group means and the overall mean. A large between-groups variance suggests a
significant effect of the factor(s).
• Within-Groups Variance: Measures the variability within each group. This is essentially "noise" and is used as a baseline to compare against the
between-groups variance.
• These variances are compared using an F-ratio:

𝐹=
Between−Groups Variance

Within−Groups Variance

• A high F-ratio suggests that the differences between group means are larger than would be expected by chance, supporting the alternative
hypothesis.
• A low F-ratio suggests no significant difference between group means, supporting the null hypothesis.
Example of One-Way ANOVA
• Study Title: Examining the Effect of Different Diets on Weight Loss

• Objective: To determine if there are significant differences in weight loss among three different diet groups: low-carb, low-fat, and
Mediterranean diets.

• Data Collection: Participants are randomly assigned to one of the three diets. After 12 weeks, weight loss (in kilograms) is measured.
Hypotheses:

• Null Hypothesis (H₀): There is no difference in the average weight loss between the three diet groups.

• Alternative Hypothesis (H₁): At least one diet group has a different average weight loss.
Example of One-Way ANOVA
Reject or accept Null Hypothesis
Compare the F-
value to a critical
value from the
Fdistribution table

F-ratio F-critical reject

Null Hypothesis

Using p-value p-value <0.05 reject

Null
Hypothesis
How to find F-critical value

• Significance Level (α): Typically, researchers use a 0.05 or 0.01 significance level, but it can vary depending on the study.

• Degrees of Freedom for the Numerator (df₁): This is based on the number of groups or levels in your first factor (between-groups variance).

• Degrees of Freedom for the Denominator (df₂): This comes from the error term or within-groups variance, often based on the sample size and
number of groups.
Steps to Find the Critical F Value

• Identify your significance level (α): Typically 0.05 or 0.01 for a two-tailed test, but 0.05 is commonly used in ANOVA. For a one-tailed test, this
remains the same if you’re interested in only one direction of the effect.

• Calculate Degrees of Freedom:

•
three groups, 𝑑𝑓1=3-1=2
Degrees of Freedom for the Numerator (df₁): This is related to the number of groups (or treatment levels) minus one. For example, if you have

• Degrees of Freedom for the Denominator (df₂): This is the total number of observations minus the number of groups. For example, if you have 15
total
Steps to Find the Critical F Value

• Locate the Critical Value in the F-Table:

observations across three , 𝑑 =15-3=12

 • Find the column corresponding to 𝑑𝑓 (numerator degrees of freedom)
1

• Find the row corresponding to 𝑑𝑓 (denominator degrees of freedom)

2

• Look at the intersection of the selected row and column for the chosen significance level. This number is your critical F value.
Steps to Find the Critical F
Value - example

• Significance Level: 𝛼 = 0.05

• Degrees of Freedom for the Numerator (df₁): 2

• Degrees of Freedom for the Denominator (df₂):

12
 After performing ANOVA?
Post Hoc Tests

• Purpose: To determine which specific groups differ from each other.

• Common Tests:

• Tukey's HSD (Honestly Significant Difference): Often used for equal sample sizes; it controls for the overall Type I error rate and is appropriate
for multiple comparisons.

• Bonferroni Correction: A conservative approach that adjusts the significance level when making multiple comparisons.

• Scheffé Test: More flexible for unequal sample sizes and allows testing of all possible contrasts between group means.

• Dunnett's Test: Compares each group mean to a control group.

• Interpretation: These tests will show which specific pairs of groups have statistically
 EXPERIMENTAL DESIGN
• https://www.khanacademy.org/science/highschool-biology/hs-biology-foundations/hsbiology-and-the-scientific-
method/v/introduction-to-experimental-design
Experimental design ensures that an experiment is set up in a way that the results are reliable, valid, and can be attributed to the changes made by the
researcher.
Experimental design is a method used to plan and structure experiments in a way that allows researchers to test a hypothesis, measure outcomes, and
draw conclusions. The goal is to identify cause-and-effect relationships between variables
Variables

• Independent variable: The factor you change or manipulate (e.g., type of fertilizer).

• Dependent variable: The outcome or effect you measure (e.g., plant growth).

• Confounding variable: unwanted variable that may influence the dependent variable, affecting results

• Control variables: Factors that are kept the same to ensure they don’t influence the results
(e.g., amount of water).
Groups

• Experimental group: The group that receives the treatment (e.g., plants with fertilizer).

• Control group: The group that does not receive the treatment or receives a standard treatment (e.g., plants without fertilizer). This helps to
compare the effects.
 Randomization

Participants or subjects are randomly assigned to different groups to reduce bias and ensure the groups are comparable.
 Replication

The experiment is repeated several times to ensure the results are consistent and not due to chance.
 Data collection and analysis

After the experiment, data is collected and analyzed to see if the independent variable (treatment) had a significant effect on the dependent variable
(outcome).
 The classification of experimental designs *

* source: Pandey, P., & Pandey, M. M. (2021). Research methodology tools and techniques. Bridge Center.
Pre-experimental designs

There is no control over extraneous or situation variables (In an experiment, an extraneous variable is any variable that is not being investigated but has
the potential to influence the results of a research project.)
When to Use Pre-Experimental Designs

• Exploratory Research: When exploring a new concept or treatment and you need initial insights without investing in a full experiment.

• Pilot Studies: To gather preliminary data that can guide the design of a more controlled study.

• Constraints on Resources: When time, budget, or ethical considerations limit the use of true experimental methods.

Pre-experimental designs

Design No. 0: One-Shot Case Study Design

• Structure: In this design, a single group is exposed to a treatment, and then the outcome is measured.

• Notation: X → O
 • X: Treatment or intervention

• O: Observation or measurement of the outcome after the treatment

• Example: A teacher implements a new teaching method in a classroom and measures students' performance afterward. There is no pre-test or
control group for comparison, so it's unclear if changes in performance are due to the teaching method or other factors.

• Limitations: This design has significant validity threats, as it lacks a pre-test and control group. You cannot conclude that the outcome is a result
of the intervention alone.
Pre-experimental designs

This study lacks a control group, making it difficult to determine if the results are due to treatment or extraneous variables
 Pre-experimental designs – Design No.1

Imagine a researcher wants to test the effectiveness of a new study skills workshop for improving student test scores.
Steps in the One-Group Pretest-Posttest Design:
1.Pretest: The researcher gives a test to a group of students to assess their current study skills knowledge.
2.Intervention: The students attend a study skills workshop aimed at improving their techniques and understanding.
3.Posttest: After the workshop, the researcher administers the same test to the students to measure any improvement in their knowledge or skills.
By comparing the pretest and posttest scores, the researcher can observe any changes in the students' study skills. However, since there is no control
group, other factors (like motivation or additional study) could have contributed to any changes observed, so causality cannot be definitively
Design No.1

• Example: A manager wants to test a new training program. The manager assesses employees’ skills before the training, conducts the training
program, and then measures the employees' skills again afterward.

• Limitations: While this design allows comparison between pre- and post-intervention, it still lacks a control group. Changes in the outcome could
be due to factors like maturation or external events rather than the intervention itself.
Design No.2
 Pre-experimental designs -
The two groups are not selected randomly or based on matching. The lack of a pre-test prevents the researcher from measuring treatment-related
differences. This design compares post-test results in experimental and control groups.
Design No.2

Imagine a researcher wants to evaluate the effectiveness of a new reading program designed to improve literacy skills among middle school
students. Steps in the Two-Group Static Design:
1.Select two existing groups of students: The researcher selects two similar classes, where one is designated as the experimental group and the
other as the control group.
2.Implement the treatment: The experimental group participates in the new reading program, while the control group continues with the standard
curriculum.
3.Posttest measurement: At the end of the program, the researcher administers a reading comprehension test to both groups.
4.Compare results: The researcher compares the posttest scores of the experimental group and the control group to see if there was any
improvement in reading comprehension due to the program.
While the researcher can observe any differences in performance, the lack of random assignment means there could be other pre-existing
differences between the
Design No.2

• Example: A researcher introduces a new product feature to one group of customers and not to another, then measures customer satisfaction in
both groups afterward.

• Limitations: Since there is no pre-test, it’s hard to confirm that any differences in the outcomes are due solely to the treatment. Differences might
have pre-existed between the two groups.
Strengths and Limitations of Pre-Experimental Designs

• Strengths:

• Simple and Cost-Effective: These designs are easy to implement and require minimal resources.

• Exploratory Insights: Useful for testing hypotheses or interventions on a preliminary basis before conducting a more rigorous study.

• Quick Results: Suitable for early-stage research when rapid results are needed.

• Limitations:

• Lacks Randomization: Without random assignment, there is a risk that differences between groups (if any) may bias the results.

• No Control Over Extraneous Variables: Many factors beyond the treatment can influence the outcome, making it difficult to establish causality.

• Low Internal Validity: The designs are prone to threats

Notation

• M: Matching subjects on relevant characteristics.

• R: Random assignment to groups

• X: Treatment or intervention (only for the experimental group)

such as history, maturation, testing effects,

 True-experimental designs -
• O: Test measurement

Design No.3

Researcher attempts to control the effects of testing, measuring instruments etc.

Experimental Group: R → X → O
Control Group: R→—→O
 True-experimental designs -
Design No.3

Imagine a researcher is studying the effect of a new memory-enhancing drug on adults' memory performance.
Steps in the Two-Group, Randomized Subject, PostTest Only Design:
1.Random Assignment: The researcher randomly assigns 100 adults to two groups of 50. Group A (the experimental group) will receive the
memory-enhancing drug, while
Group B (the control group) will receive a placebo.
2.Intervention: Group A takes the drug as prescribed, while Group B takes the placebo.
3.Posttest Measurement: After a set period, both groups take a memory test to assess their memory performance.
4.Comparison: The researcher compares the posttest scores of Group A and Group B to determine if there is a significant difference in memory
performance between the two groups.
Design No.4

Experimental Group: M → R → X → O
 True-experimental designs -
Control Group: M → R → — → O

Design No.4

• Imagine researchers are studying the effect of a new therapy technique on reducing anxiety. They want to control for factors like initial anxiety
levels, age, and gender to ensure these don’t influence the results. They:

• Match participants into pairs based on their initial anxiety levels, age, and gender.
• Randomly assign one member of each pair to the experimental group (receiving the new therapy technique) and the other to the control group
(receiving no therapy or a placebo).

• Measure anxiety levels in both groups after the intervention to assess its effectiveness.
Design No.5
True-experimental designs -
Experimental Group: R → O1 → X → O2
Control Group: R → O1 → — → O2
Design No.5

• Suppose researchers are studying the effect of a new educational program on students' math test scores. They:

• Randomly assign students to either the experimental group (receives the program) or control group (does not receive the program).

• Administer a pre-test to all students to measure their math skills before the program.

• The experimental group receives the new educational program, while the control group continues with regular instruction.

• Administer a post-test to all students to measure any changes in math skills.

• By comparing the pre- and post-test scores between groups, researchers can determine if the program had a significant impact on math
performance.
True-experimental designs Design No.6

Group 1: R → O1 → X → O2
Group 2: R→ X → O2
Group 3: R → O1 → O2
Design No.6

• Suppose researchers are testing a new training program’s effect on job performance. The design would be structured as follows:

• Group 1 takes a pre-test, completes the training program (treatment), and then takes a post-test.
 • Group 2 skips the pre-test, completes the training program, and then takes a post-test.

• Group 3 takes the pre-test, does not complete the training, and then takes the post-test.

• This setup enables researchers to isolate the effect of the training program itself (comparing post-test scores across groups) and also observe
whether the pre-test influenced participants' responses to the training.
Design No.6

The Solomon Three-Group Design is useful when there is concern that pre-testing might affect participants' responses to the treatment. By isolating the
effects of the pre-test, treatment, and their interaction, this design provides a clearer picture of the true effect of the treatment.
True-experimental designs
 Group 1: R → O1 → X → O2
Group 2: R → O1 → → O2
Group 3: R → → X → O2
Group 4: R → → → O2
Design No.7

• Suppose researchers are studying the effect of a new training program on employee productivity. The design would be set up as follows:
• Group 1 takes a pre-test, completes the training (treatment), and then takes a post-test.
• Group 2 takes a pre-test, skips the training, and then takes a post-test.
• Group 3 skips the pre-test, completes the training, and then takes a post-test.
• Group 4 skips both the pre-test and the training, taking only the post-test.
• This arrangement allows the researchers to compare results across all groups to see if any observed differences in post-test scores are
Factorial Design

Factorial design is a systematic method used in experiments to evaluate the effects of multiple factors simultaneously. This design is particularly valuable for
studying interactions between factors, which can provide a more comprehensive understanding of how variables influence outcomes. Below, I’ll provide
an overview of factorial design, including its characteristics, benefits, types, and an example.
 Types of Factorial Design

1. Full Factorial Design: Every possible combination of factors and levels is tested. For example, a 2x2 design tests two factors, each at two levels,
resulting in four treatment combinations.

2. Fractional Factorial Design: Only a subset of the treatment combinations is tested. This is useful when resources are limited or when there are
many factors, allowing researchers to focus on the most significant factors and interactions.

3. Mixed Factorial Design: Combines elements of both fixed and random effects, accommodating both withinsubject and between-subject factors.
Example of a Factorial Design

• Study Title: The Impact of Fertilizer Type and Watering Frequency on Plant Growth

• Objective: To assess how different types of fertilizer and varying watering frequencies affect the growth of tomato plants.
 Example of a Factorial Design
• Factors and Levels:

1.Factor A: Fertilizer Type

1.Level 1 (−): Organic fertilizer
2.Level 2 (+): Chemical fertilizer
2.Factor B: Watering Frequency
1.Level 1 (−): Once a week
2.Level 2 (+): Twice a week

• Treatment Combinations:
(−, −): Organic fertilizer with once-a-week watering
(−, +): Organic fertilizer with twice-a-week watering
(+ , −): Chemical fertilizer with once-a-week watering
(+ , +): Chemical fertilizer with twice-a-week watering
 Example of a Factorial Design
• Experimental Design:

1.Participants: 40 tomato plants are randomly assigned to one of the four treatment combinations, with 10 plants per treatment group.
2.Data Collection: After a predetermined growing period (e.g., 8 weeks), measure the height of the plants and record the data.

• Analysis: Use ANOVA to analyze the data and assess:

• The main effect of fertilizer type on plant growth.

• The main effect of watering frequency on plant growth.

• The interaction effect between fertilizer type and watering frequency.

Factorial design is a powerful research method that allows for an efficient and comprehensive examination of multiple factors and their interactions. It is
widely applicable across various fields, including agriculture, psychology, medicine, and industrial processes, making it a fundamental tool in
experimental research. By employing factorial designs, researchers can derive meaningful insights and optimize processes based on their findings.
Types of Factorial Designs

• 2x2 Factorial Design: Two factors with two levels each.

• 3x3 Factorial Design: Two factors with three levels each.

• 2x3 Factorial Design: One factor with two levels, and the other with three levels.

2x2 Factorial Design is a common and straightforward type of factorial design in experiments. In a 2x2 factorial design:

• There are two factors (independent variables).

• Each factor has two levels (e.g., high/low, yes/no).

 2x2 Factorial Design
Structure of a 2x2 Factorial Design
This design results in four possible combinations, allowing researchers to observe:

• The main effect of Factor A (e.g., the effect of one variable on the outcome, regardless of the other variable's level).

• The main effect of Factor B.

• The interaction effect between Factors A and B, which examines how the combination of levels influences the outcome.

Example
Imagine a study investigating the effects of Exercise Type (Factor A: aerobic or strength training) and Diet Type (Factor B: low-carb or high-carb) on weight
loss. The four conditions would be:

• Aerobic exercise + low-carb diet

• Aerobic exercise + high-carb diet

• Strength training + low-carb diet

• Strength training + high-carb diet

 2x2 Factorial Design
3x3 Factorial Design

In a 3x3 factorial design:

• Factor A has three levels (e.g., low, medium, high).

• Factor B also has three levels.

Each combination of Factor A and Factor B is tested, resulting in nine conditions in total. This allows researchers to observe:

• The main effect of Factor A.

• The main effect of Factor B.

• The interaction effect between Factors A and B.

 3x3 Factorial Design
Example
Suppose researchers want to study the effects of Medication Dosage (Factor A: low, medium, high) and Exercise Intensity (Factor B: none, moderate,
intense) on blood pressure reduction. The nine conditions would be:

• Low dosage + no exercise

• Low dosage + moderate exercise
• Low dosage + intense exercise
• Medium dosage + no exercise
• Medium dosage + moderate exercise
• Medium dosage + intense exercise
• High dosage + no exercise
• High dosage + moderate exercise
• High dosage + intense exercise
2x3 Factorial Design

Structure of a 2x3 Factorial Design In this design:

• Factor A has two levels (e.g., presence vs. absence of a treatment).

• Factor B has three levels (e.g., low, medium, and high).

• Each combination of the levels for Factors A and B is tested, creating six distinct groups or conditions.

2x3 Factorial Design

Example
Suppose researchers are studying the effects of Diet
Type (Factor A: low-carb, high-carb) and Exercise Duration (Factor B: short, moderate, long) on weight loss. The six conditions would be:

• Low-carb diet + short exercise duration

• Low-carb diet + moderate exercise duration

• Low-carb diet + long exercise duration

• High-carb diet + short exercise duration

• High-carb diet + moderate exercise duration

• High-carb diet + long exercise duration