Two-Way Analysis of Variance (ANOVA)

Unit Overview

Course: MPC-006 Statistics in Psychology · Block 3 · Unit 4
Pages: 96–112 · 📄 Download PDF
Prerequisite: One-Way ANOVA | t-Test for Independent & Correlated Samples

From One Factor to Two: Why We Need Two-Way ANOVA

One-way ANOVA answers: "Do groups differ on this outcome variable?" — but it can only examine one independent variable (IV) at a time. Real-world behaviour is shaped by multiple factors acting simultaneously, and often the combination of factors produces an effect that neither factor alone could predict.

Consider a pharmaceutical company testing a new drug to reduce smoking:

What is the independent effect of Drug A on smoking?
What is the independent effect of Drug B on smoking?
What is the joint (interaction) effect of Drug A × Drug B together?

No single one-way ANOVA can answer all three questions. We need Two-Way ANOVA — a technique that simultaneously tests:

The main effect of Factor A (e.g., Type of Personality)
The main effect of Factor B (e.g., Level of Anxiety)
The interaction effect A × B (the combined influence of both factors)

This is the essence of factorial experimental designs — studying how variables interact, not just act in isolation.

Historical Note

Two-way (and multi-way) ANOVA grew from Fisher's foundational agricultural research in the 1920s, where he needed to simultaneously test the effect of multiple soil treatments and crop varieties. The extension to psychology was pioneered by researchers at American universities in the 1940s–50s. Wikipedia: Factorial experiment

The Structure of Two-Way ANOVA

Factorial Design Notation

Two-way ANOVA uses a factorial design notation: a × b, where:

a = number of levels of Factor A
b = number of levels of Factor B
Total cells = a × b

Design	Factors	Cells	Example
2 × 2	2 IVs, 2 levels each	4	Gender × Anxiety (each High/Low)
3 × 2	IV₁ has 3 levels, IV₂ has 2	6	SES (High/Average/Low) × Gender
4 × 3	IV₁ has 4 levels, IV₂ has 3	12	4 teaching methods × 3 intelligence levels

Three Effects Tested Simultaneously

The Two-Way ANOVA Summary Table

Source of Variance	df	SS	MSS	F Ratio
Among All Groups	(k−1)	SS_A	—	—
Between Groups: Factor A	(k_a − 1)	SS_BFactorA	MSS_A	F₁ = MSS_A / MSS_W
Between Groups: Factor B	(k_b − 1)	SS_BFactorB	MSS_B	F₂ = MSS_B / MSS_W
Interaction A × B	(k_a−1)(k_b−1)	SS_AxB	MSS_AxB	F₃ = MSS_AxB / MSS_W
Within Groups (Error)	(N − k)	SS_W	MSS_W	—
Total	N − 1	SS_T	—	—

The within-groups (error) variance is the denominator for all three F-ratios — it represents the variability that cannot be attributed to any experimental factor.

Variance Partitioning in Two-Way ANOVA

The total variance is carved into more components than in one-way ANOVA:

$SS_{Total} = SS_{Between\,Groups} + SS_{Within\,Groups}$

$SS_{Between\,Groups} = SS_{Factor\,A} + SS_{Factor\,B} + SS_{A \times B}$

Therefore:

$SS_T = SS_A + SS_B + SS_{A \times B} + SS_W$

This elegant partition means that two-way ANOVA accounts for more sources of variance than one-way ANOVA, leaving a smaller (and more accurate) error term. This increases statistical power — the ability to detect true effects.

Step-by-Step Computation

Notation for a 2 × 2 Design

Symbol	Meaning
$k$	Total number of cells/groups
$k_a$	Levels of Factor A
$k_b$	Levels of Factor B
$n$	Cases per cell
$N$	Total cases = $k \times n$
$\sum X_1, \sum X_2 ...$	Cell sums
$Cx$	Correction term

The Computation Steps

Step 1: Correction Term $Cx = \frac{(\sum X)^2}{N}$

Step 2: Total Sum of Squares $SS_T = \sum X^2 - Cx$

Step 3: Sum of Squares Among All Groups (treats all $k$ cells as separate groups) $SS_{Among} = \frac{(\sum X_1)^2}{n} + \frac{(\sum X_2)^2}{n} + \cdots + \frac{(\sum X_k)^2}{n} - Cx$

Step 4: SS for Factor A (combine cells that share the same level of A) $SS_A = \frac{(\sum X_{A_1})^2}{n_{A_1}} + \frac{(\sum X_{A_2})^2}{n_{A_2}} + \cdots - Cx$

Step 5: SS for Factor B (combine cells that share the same level of B) $SS_B = \frac{(\sum X_{B_1})^2}{n_{B_1}} + \frac{(\sum X_{B_2})^2}{n_{B_2}} + \cdots - Cx$

Step 6: Interaction SS $SS_{A \times B} = SS_{Among} - SS_A - SS_B$

Step 7: Within-Groups SS (Error) $SS_W = SS_T - SS_{Among}$

Step 8: Degrees of Freedom

$df_A = k_a - 1$
$df_B = k_b - 1$
$df_{A \times B} = (k_a - 1)(k_b - 1)$
$df_W = N - k$
$df_T = N - 1$

Step 9: Mean Squares $MSS_A = \frac{SS_A}{df_A}, \quad MSS_B = \frac{SS_B}{df_B}, \quad MSS_{A \times B} = \frac{SS_{A \times B}}{df_{A \times B}}, \quad MSS_W = \frac{SS_W}{df_W}$

Step 10: F-Ratios $F_A = \frac{MSS_A}{MSS_W}, \quad F_B = \frac{MSS_B}{MSS_W}, \quad F_{A \times B} = \frac{MSS_{A \times B}}{MSS_W}$

Worked Example 1: Anxiety × Personality Type on Academic Achievement

Source: PDF p. 99–102

A researcher studied 20 undergraduates in a 2 × 2 design (Type of Personality × Level of Anxiety). Academic achievement scores:

	High Anxiety	Low Anxiety
Extroverts	12, 13, 14, 15, 14	14, 14, 13, 15, 15
Introverts	14, 16, 16, 16, 15	11, 10, 12, 12, 16

Cell Statistics:

Cell	Σ	M
Extroverts × High Anxiety (X₁)	68	13.60
Extroverts × Low Anxiety (X₂)	71	14.20
Introverts × High Anxiety (X₃)	77	15.40
Introverts × Low Anxiety (X₄)	61	12.20

$k = 4$ , $n = 5$ , $N = 20$

Step 1: $Cx = \frac{(68+71+77+61)^2}{20} = \frac{277^2}{20} = \frac{76729}{20} = 3836.45$

Step 2: $\sum X^2 = 930 + 1011 + 1189 + 765 = 3895$
$SS_T = 3895 - 3836.45 = 58.55$

Step 3: $SS_{Among} = \frac{68^2+71^2+77^2+61^2}{5} - 3836.45 = \frac{4624+5041+5929+3721}{5} - 3836.45 = 3863.00 - 3836.45 = 26.55$

Step 4 (Personality Type — rows):
Extroverts total = 68 + 71 = 139 (n = 10); Introverts total = 77 + 61 = 138 (n = 10)
$SS_{Personality} = \frac{139^2}{10} + \frac{138^2}{10} - 3836.45 = 1932.10 + 1904.40 - 3836.45 = 0.05$

Step 5 (Anxiety Level — columns):
High Anxiety total = 68 + 77 = 145 (n = 10); Low Anxiety total = 71 + 61 = 132 (n = 10)
$SS_{Anxiety} = \frac{145^2}{10} + \frac{132^2}{10} - 3836.45 = 2102.50 + 1742.40 - 3836.45 = 8.45$

Step 6: $SS_{A \times B} = 26.55 - 0.05 - 8.45 = 18.05$

Step 7: $SS_W = 58.55 - 26.55 = 32.00$

Summary Table:

Source	df	SS	MSS	F
Among Groups	3	26.55	—	—
Personality Type (A)	1	0.05	0.05	0.025
Anxiety Level (B)	1	8.45	8.45	4.225
A × B Interaction	1	18.05	18.05	9.025
Within Groups (Error)	16	32.00	2.00	—
Total	19	58.55	—	—

Critical F values (df₁ = 1, df₂ = 16): F₀₅ = 4.49, F₀₁ = 8.53

Decisions:

Personality Type: F = 0.025 < 4.49 → Not significant — personality type alone does not affect academic achievement
Anxiety Level: F = 4.225 < 4.49 → Not significant — anxiety level alone does not affect academic achievement
Interaction A × B: F = 9.025 > 8.53 → Significant at .01 level ✅ — the combination of personality type and anxiety level matters!

Interpretation: Although neither personality type nor anxiety individually predicts achievement, their joint effect does. Specifically: Introverts with high anxiety perform best (M = 15.40), while Introverts with low anxiety perform worst (M = 12.20). Extroverts show a more moderate, reversed pattern. This cross-over interaction is the hallmark of a significant A × B effect.

Worked Example 2: Sex × Intelligence on Mathematical Creativity

Source: PDF p. 103–106

40 high school students in a 2 × 2 design: Sex (Boys/Girls) × Intelligence (High/Low). Dependent variable: mathematical creativity test scores.

Cell Summary:

Cell	n	Σ	M
Boys × High Intelligence	10	153	15.30
Boys × Low Intelligence	10	137	13.70
Girls × High Intelligence	10	139	13.90
Girls × Low Intelligence	10	111	11.10

Key Computations:

$Cx = \frac{(153+137+139+111)^2}{40} = \frac{540^2}{40} = 7290$

$SS_T = (2381+1889+1939+1247) - 7290 = 166$

$SS_{Among} = \frac{153^2+137^2+139^2+111^2}{10} - 7290 = 92$

$SS_{Sex} = \frac{(153+137)^2}{20} + \frac{(139+111)^2}{20} - 7290 = \frac{290^2+250^2}{20} - 7290 = 40.00$

$SS_{Intelligence} = \frac{(153+139)^2}{20} + \frac{(137+111)^2}{20} - 7290 = \frac{292^2+248^2}{20} - 7290 = 48.40$

$SS_{A \times B} = 92 - 40 - 48.40 = 3.60$

$SS_W = 166 - 92 = 74.00; \quad MSS_W = \frac{74}{36} = 2.056$

Summary Table:

Source	df	SS	MSS	F
Sex (A)	1	40.00	40.00	19.46 ✅
Intelligence (B)	1	48.40	48.40	23.54 ✅
Sex × Intelligence (A×B)	1	3.60	3.60	1.75 ❌
Within Groups (Error)	36	74.00	2.056	—
Total	39	166.00	—	—

Critical F (df₁ = 1, df₂ = 36): F₀₅ = 4.12, F₀₁ = 7.42

Decisions:

Sex: F = 19.46 > 7.42 → Significant at .01 — Boys score higher in mathematical creativity (M = 14.50 vs 12.50)
Intelligence: F = 23.54 > 7.42 → Significant at .01 — High-intelligence students score higher (M = 14.60 vs 12.40)
Sex × Intelligence: F = 1.75 < 4.12 → Not significant — the advantage of high intelligence is roughly equal for both boys and girls

Understanding Interaction Effects

The interaction effect is the most conceptually rich aspect of two-way ANOVA. It answers: "Does the effect of Factor A depend on the level of Factor B?"

Visualising Interaction: Parallel vs Crossing Lines

Rule of thumb for interaction graphs:

Parallel lines → No interaction (the effect of A is the same at every level of B)
Non-parallel / crossing lines → Possible or significant interaction

Practical Examples of Interaction

Domain	Factor A	Factor B	Possible Interaction
Clinical	Drug dosage	Patient age group	Drug effective for adults but not children
Education	Teaching method	Intelligence level	Discussion works for bright students; drilling works for others
Organisational	Work schedule	Job type	Flexible hours boosts satisfaction for creative jobs but not routine ones
Developmental	Parenting style	Child temperament	Authoritative parenting helps anxious children more than easy-going ones

The fertiliser analogy from the PDF is instructive: Urea and Phosphate each have independent effects on crop growth, but their combined application in the right ratio produces results neither achieves alone — the joint effect (interaction) is not predictable from individual effects.

Wikipedia: Interaction (statistics)

Decision Rules for Two-Way ANOVA Results

Three possible outcomes and what to do:

All F Non-Significant
All F Significant
Partial Significance

All three obtained F values (F_A, F_B, F_AxB) are below the critical value.

Conclusion: Neither variable alone nor in combination has a significant effect on the dependent variable. Retain all null hypotheses. No post-hoc tests needed. Report results and stop.

Example: In Example 2, if the interaction F (1.75) had also been non-significant, you would conclude that neither sex, intelligence, nor their combination affects mathematical creativity.

One-Way vs Two-Way ANOVA: A Comparison

Feature	One-Way ANOVA	Two-Way ANOVA
Independent variables	1	2
Effects tested	1 (main effect)	3 (2 main + 1 interaction)
Design type	Simple random design	Factorial design
Interaction analysis	Not possible	Core feature
F-ratios computed	1	3
Efficiency	Tests one IV at a time	Tests two IVs simultaneously
Statistical power	Lower (error includes B variance)	Higher (partitions more variance)
Complexity	Moderate	High

Key insight: Two-way ANOVA is more efficient than running two separate one-way ANOVAs because: (a) it controls for the second factor when testing the first, and (b) it reveals interaction effects that separate analyses cannot detect.

Assumptions of Two-Way ANOVA

Two-way ANOVA inherits all the assumptions of one-way ANOVA:

Normality: The dependent variable is normally distributed within each cell
Homogeneity of variance: All cells have equal population variance (Levene's test)
Independence: Observations are independent within and across cells
Equal cell sizes (balanced design): Makes computation simpler and the test more robust

Additional consideration for two-way designs: With very small cell sizes (n < 5), the test is especially sensitive to violations. Balanced designs (equal n per cell) are strongly preferred.

Wikipedia: Analysis of variance — Assumptions

Merits and Limitations

✅ Merits
⚠️ Limitations

Tests two main effects and one interaction in a single analysis — three hypotheses for the price of one
Reveals interaction effects that would be completely invisible to separate one-way ANOVAs
Higher statistical power than one-way ANOVA — partitioning the B variance out of error reduces the error term
Ideal for factorial designs (2×2, 3×2, 4×3 etc.) which closely mimic real-world multi-causal phenomena
Produces a richer interpretation: not just "does A matter?" but "does A matter more at some levels of B?"
Foundation for more complex designs: three-way ANOVA, MANCOVA, mixed-model ANOVA all build on this

Indian Research Context

Two-way ANOVA is the backbone of much factorial research in Indian educational and social psychology:

Educational Psychology: Studies examining the joint effects of gender × school type (government vs private) on academic performance are ubiquitous in Indian M.Ed. dissertations. Frequently, the interaction reveals that gender gaps are larger in private schools than government ones — a finding invisible to separate one-way analyses.

Clinical Psychology: Indian clinical researchers have used 2×2 designs to study how diagnosis × socioeconomic status jointly affect treatment adherence — finding significant interactions at NIMHANS, PGI Chandigarh, and affiliated centres.

Developmental Psychology: Research on parenting style × child birth order as predictors of adolescent self-concept is a recurring theme in Indian family psychology, reflecting the culturally distinct role of joint families and sibling hierarchies in India.

Occupational Psychology: Studies on job sector (public vs private) × gender as determinants of work-life balance report consistent interaction effects, with gender differentials far more pronounced in the private sector — a pattern with significant policy implications.

Real-World Applications

Pharmacology RCTs: Testing drug efficacy across gender × dosage cells — the interaction reveals if a drug works differently for men vs women
Educational interventions: Instruction method × learning style combinations to personalise teaching
Neuropsychology: Memory training × age group interactions in cognitive rehabilitation
Consumer psychology: Ad format × product category interactions in purchase decision research
Health psychology: Exercise type × dietary regimen combinations in weight loss interventions

Memory Aids

🧠 The "Main + Interaction" Triangle

         Two-Way ANOVA
        /              \
   Main Effect A    Main Effect B
        \              /
         Interaction
           (A × B)

Two-Way ANOVA always gives you THREE effects. Never forget the third — the interaction.

🎵 Parallel Lines Rule

Parallel lines = No interaction (the gap stays the same)
Crossing lines = Interaction present (the relationship reverses)
"Lines that cross mean factors interact; lines that don't means they act apart."

🔢 df Quick Reference for 2×2 Designs

Effect	df
Factor A (2 levels)	$2-1=1$
Factor B (2 levels)	$2-1=1$
A × B interaction	$1 \times 1=1$
Within groups (N=20)	$20-4=16$
Total (N=20)	$19$

⚡ Interaction vs Main Effect Quick Rule

If main effects are not significant but interaction is significant → The factors only matter together, not alone (Example 1: Personality × Anxiety)
If main effects are significant but interaction is not → The factors work additively, not jointly (Example 2: Sex + Intelligence)

External Resources

📖 Essential Reading

Wikipedia: Two-way analysis of variance — comprehensive mathematical treatment
Wikipedia: Interaction (statistics) — conceptual and graphical explanation
Statistics How To: Two-Way ANOVA — step-by-step worked examples with Excel output

🎓 Video Lectures

Crash Course Statistics #34: Two-Way ANOVA — accessible conceptual introduction
StatQuest: Two-Way ANOVA (Josh Starmer) — visual variance partitioning
Khan Academy: Interactions in factorial ANOVA — interactive exercises

📚 Research Papers (2020–2025)

Lakens, D. (2022). Sample size justification. Collabra: Psychology, 8(1). [doi:10.1525/collabra.33267] — covers power for factorial designs
Maxwell, S.E., Delaney, H.D., & Kelley, K. (2018). Designing Experiments and Analyzing Data: A Model Comparison Perspective (3rd ed.). Routledge. — standard advanced reference for factorial ANOVA
Shieh, G. (2020). Power analysis for two-way balanced ANOVA. Journal of Statistical Computation and Simulation, 90(7). — for planning two-way designs

🛠 Online Tools

Social Science Statistics: Two-Way ANOVA Calculator — free browser-based computation
Real Statistics: Two-Factor ANOVA in Excel — detailed Excel tutorials with interpretation
JASP Software — free, user-friendly statistics software with beautiful ANOVA output for students

Self-Assessment Questions

🟢 Knowledge (Recall)

Define two-way ANOVA and explain how it differs from one-way ANOVA. What additional information does it provide?
Name the three effects tested in a two-way ANOVA. Define each one clearly.
State the formula for the interaction sum of squares ( $SS_{A \times B}$ ) in terms of other SS components. Why is the interaction variance calculated this way?

🟡 Comprehension (Understanding)

Explain the interaction effect using a concrete example from psychology. What does it mean when two lines on an interaction graph are parallel versus when they cross?
In Example 1 (Anxiety × Personality), the main effects of both factors were non-significant, yet the interaction was highly significant at .01 level. How is this possible? What does this tell us about the relationship between personality type, anxiety, and academic achievement?
Compare Example 1 and Example 2 in terms of their interaction findings. In which example do the two factors act additively? In which do they interact? How do the graphs for each differ?

🔴 Application (Problem-Solving)

Solve this problem: A researcher studies the effect of teaching method (Traditional vs Activity-based) and class size (Small: ≤20 vs Large: >40) on student satisfaction scores. Data from a 2×2 design with n = 6 per cell:

	Traditional	Activity-Based
Small Class	14, 13, 15, 12, 14, 13	17, 18, 16, 19, 17, 18
Large Class	10, 11, 9, 12, 10, 10	12, 13, 11, 14, 12, 12

Apply two-way ANOVA. Calculate $Cx$ , $SS_T$ , $SS_{Among}$ , $SS_{Method}$ , $SS_{ClassSize}$ , $SS_{M \times C}$ , and $SS_W$ . Complete the summary table and determine significance for each effect.

Summary

Concept	Key Point
Two-way ANOVA	Tests 2 main effects + 1 interaction simultaneously
Factorial design	a × b notation (levels of A × levels of B)
Main effect	Effect of one IV while controlling for the other
Interaction effect	Effect of A depends on the level of B (and vice versa)
$SS_{A \times B}$	$SS_{Among} - SS_A - SS_B$
Parallel lines (graph)	No interaction — effects are additive
Crossing lines (graph)	Significant interaction — effects are multiplicative/conditional
All F non-significant	Retain all H₀; no further testing
Some F significant	Partial rejection; post-hoc t-tests for factors with 3+ levels
Advantage over 1-way	Detects interaction; higher power; tests two IVs in one analysis
Assumption	Normality, homogeneity of variance, independence, balanced cells

Source PDFs:

📄 Block-3/Unit-4.pdf — Pages 96–112
📚 MPC-006 Statistics in Psychology, IGNOU

Two-Way Analysis of Variance (ANOVA)

From One Factor to Two: Why We Need Two-Way ANOVA​

The Structure of Two-Way ANOVA​

Factorial Design Notation​

Three Effects Tested Simultaneously​

The Two-Way ANOVA Summary Table​

Variance Partitioning in Two-Way ANOVA​

Step-by-Step Computation​

Notation for a 2 × 2 Design​

The Computation Steps​

Worked Example 1: Anxiety × Personality Type on Academic Achievement​

Worked Example 2: Sex × Intelligence on Mathematical Creativity​

Understanding Interaction Effects​

Visualising Interaction: Parallel vs Crossing Lines​

Practical Examples of Interaction​

Decision Rules for Two-Way ANOVA Results​

One-Way vs Two-Way ANOVA: A Comparison​

Assumptions of Two-Way ANOVA​

Merits and Limitations​

Indian Research Context​

Real-World Applications​

Memory Aids​

🧠 The "Main + Interaction" Triangle​

🎵 Parallel Lines Rule​

🔢 df Quick Reference for 2×2 Designs​

⚡ Interaction vs Main Effect Quick Rule​

External Resources​

📖 Essential Reading​

🎓 Video Lectures​

📚 Research Papers (2020–2025)​

🛠 Online Tools​

Self-Assessment Questions​

🟢 Knowledge (Recall)​

🟡 Comprehension (Understanding)​

🔴 Application (Problem-Solving)​

Summary​