One-way ANOVA: A statistical method to test whether the means of three or more groups are equal.
- State null and alternative hypotheses
- Calculate group means and overall mean
- Compute sum of squares (between, within, total)
- Calculate mean squares
- Compute F-statistic
- Determine critical value or p-value
- Make decision and draw conclusion
H₀: μ_A = μ_B = μ_C (all means are equal)
H₁: At least one mean differs
Method A: (85+87+82+88+86)/5 = 85.6
Method B: (78+80+76+82+79)/5 = 79.0
Method C: (92+90+94+91+89)/5 = 91.2
Overall mean: (85.6 + 79.0 + 91.2)/3 = 85.27
SSB = Σn_i(ȳ_i - ȳ)²
SSB = 5(85.6-85.27)² + 5(79.0-85.27)² + 5(91.2-85.27)²
SSB = 5(0.11) + 5(39.24) + 5(35.49) = 0.55 + 196.2 + 177.45 = 374.2
SSW = ΣΣ(y_ij - ȳ_i)²
Method A: (85-85.6)² + ... + (86-85.6)² = 21.2
Method B: (78-79)² + ... + (79-79)² = 20.0
Method C: (92-91.2)² + ... + (89-91.2)² = 17.2
SSW = 21.2 + 20.0 + 17.2 = 58.4
df_between = k - 1 = 3 - 1 = 2
df_within = N - k = 15 - 3 = 12
MSB = SSB/df_between = 374.2/2 = 187.1
MSW = SSW/df_within = 58.4/12 = 4.87
F = MSB/MSW = 187.1/4.87 = 38.4
F_critical(2, 12) at α = 0.05 = 3.89
Since F = 38.4 > 3.89, we reject H₀
We reject the null hypothesis. There are significant differences in mean test scores among the three teaching methods at α = 0.05.
• ANOVA principle: Compare between-group to within-group variance
• F-test: F = MSB/MSW
• Decision rule: Reject H₀ if F > F_critical
• Between groups SS = 374.2, df = 2, MS = 187.1
• Within groups SS = 58.4, df = 12, MS = 4.87
• F-statistic = 38.4, p-value < 0.001
Equal sample sizes: When each group has the same number of observations, calculations are simplified.
Fertilizer A: (12+14+13+15+14+12)/6 = 13.33
Fertilizer B: (10+11+12+11+10+12)/6 = 11.00
Fertilizer C: (16+17+15+16+17+15)/6 = 16.00
Fertilizer D: (13+14+13+14+15+13)/6 = 13.67
Overall mean: (13.33 + 11.00 + 16.00 + 13.67)/4 = 13.50
SSB = 6(13.33-13.5)² + 6(11.0-13.5)² + 6(16.0-13.5)² + 6(13.67-13.5)²
SSB = 6(0.028) + 6(6.25) + 6(6.25) + 6(0.028) = 0.17 + 37.5 + 37.5 + 0.17 = 75.34
Fertilizer A: (12-13.33)² + ... + (12-13.33)² = 6.67
Fertilizer B: (10-11)² + ... + (12-11)² = 4.00
Fertilizer C: (16-16)² + ... + (15-16)² = 4.00
Fertilizer D: (13-13.67)² + ... + (13-13.67)² = 3.33
SSW = 6.67 + 4.00 + 4.00 + 3.33 = 18.00
df_between = 4 - 1 = 3
df_within = 24 - 4 = 20
MSB = 75.34/3 = 25.11
MSW = 18.00/20 = 0.90
F = 25.11/0.90 = 27.90
F_critical(3, 20) at α = 0.01 = 4.94
Since F = 27.90 > 4.94, we reject H₀
We reject the null hypothesis. There are significant differences in mean plant growth among the four fertilizers at α = 0.01.
• Equal sample sizes: Simplifies SSB calculation
• Higher α: Stricter significance level requires larger F
• Statistical significance: Evidence of differences exists
• Between groups SS = 75.34, df = 3, MS = 25.11
• Within groups SS = 18.00, df = 20, MS = 0.90
• F-statistic = 27.90, p-value < 0.001
Non-significant result: When F-statistic is less than critical value, we fail to reject the null hypothesis.
Condition 1: (250+255+248+252)/4 = 251.25
Condition 2: (249+253+251+250)/4 = 250.75
Condition 3: (252+248+251+253)/4 = 251.00
Overall mean: (251.25 + 250.75 + 251.00)/3 = 251.00
SSB = 4(251.25-251)² + 4(250.75-251)² + 4(251.00-251)²
SSB = 4(0.0625) + 4(0.0625) + 4(0) = 0.25 + 0.25 + 0 = 0.50
Condition 1: (250-251.25)² + ... + (252-251.25)² = 12.75
Condition 2: (249-250.75)² + ... + (250-250.75)² = 6.75
Condition 3: (252-251)² + ... + (253-251)² = 10.00
SSW = 12.75 + 6.75 + 10.00 = 29.50
df_between = 3 - 1 = 2
df_within = 12 - 3 = 9
MSB = 0.50/2 = 0.25
MSW = 29.50/9 = 3.28
F = 0.25/3.28 = 0.076
F_critical(2, 9) at α = 0.05 = 4.26
Since F = 0.076 < 4.26, we fail to reject H₀
There is insufficient evidence to conclude that lighting conditions significantly affect reaction time.
We fail to reject the null hypothesis. There is no significant difference in mean reaction times among the three lighting conditions at α = 0.05.
• Non-significant result: Fails to reject H₀
• No evidence: Insufficient evidence of differences
• Not proof: Failing to reject ≠ proving equality
• Between groups SS = 0.50, df = 2, MS = 0.25
• Within groups SS = 29.50, df = 9, MS = 3.28
• F-statistic = 0.076, p-value > 0.05
Analysis of Variance (ANOVA): A statistical method to compare means of multiple groups simultaneously.
Null hypothesis (H₀): All group means are equal.
Alternative hypothesis (H₁): At least one group mean differs from others.
Sum of squares between (SSB): Measures variation between group means and overall mean.
Sum of squares within (SSW): Measures variation within each group.
F-statistic: Ratio of between-group variance to within-group variance.
- Verify assumptions: Normality, equal variances, independence
- State hypotheses: H₀: μ₁ = μ₂ = ... = μₖ vs H₁: Not all means equal
- Calculate group statistics: Means and standard deviations
- Compute sums of squares: SSB, SSW, SST
- Calculate degrees of freedom: df_between, df_within, df_total
- Compute mean squares: MSB = SSB/df_between, MSW = SSW/df_within
- Calculate F-statistic: F = MSB/MSW
- Make decision: Compare F with critical value or p-value
- Draw conclusion: Interpret results in context
• Independence: Observations within and between groups are independent
• Normality: Data in each group is normally distributed
• Equal variances: Homogeneity of variance across groups
• ANOVA assumes equal variances across groups
• Large sample sizes help meet normality assumption
• Outliers can significantly affect results
• ANOVA is robust to minor assumption violations
P-value approach: Compare the calculated p-value with the significance level α to make a decision.
Program 1: (80+82+78+84+81)/5 = 81.0
Program 2: (85+87+83+86+84)/5 = 85.0
Program 3: (79+81+77+83+80)/5 = 80.0
Overall mean: (81.0 + 85.0 + 80.0)/3 = 82.0
SSB = 5(81.0-82.0)² + 5(85.0-82.0)² + 5(80.0-82.0)²
SSB = 5(1) + 5(9) + 5(4) = 5 + 45 + 20 = 70
Program 1: (80-81)² + ... + (81-81)² = 20
Program 2: (85-85)² + ... + (84-85)² = 10
Program 3: (79-80)² + ... + (80-80)² = 20
SSW = 20 + 10 + 20 = 50
df_between = 3 - 1 = 2
df_within = 15 - 3 = 12
MSB = 70/2 = 35
MSW = 50/12 = 4.17
F = 35/4.17 = 8.39
Using F-distribution with df = (2, 12)
p-value ≈ 0.005
Since p-value (0.005) < α (0.05), we reject H₀
We reject the null hypothesis. There are significant differences in mean performance among the three training programs at α = 0.05.
• P-value method: Compare p-value with α
• Decision rule: Reject H₀ if p-value ≤ α
• Statistical significance: Strong evidence of differences
ANOVA table: Organized summary of ANOVA calculations showing sources of variation, degrees of freedom, sum of squares, mean squares, and F-statistic.
df_between = k - 1 = 4 - 1 = 3
df_within = N - k = 32 - 4 = 28
df_total = N - 1 = 32 - 1 = 31
SST = SSB + SSW = 120 + 240 = 360
MSB = SSB/df_between = 120/3 = 40
MSW = SSW/df_within = 240/28 = 8.57
F = MSB/MSW = 40/8.57 = 4.67
F_critical(3, 28) at α = 0.05 = 2.95
Since F = 4.67 > 2.95, we reject H₀
| Source | df | SS | MS | F |
|---|---|---|---|---|
| Between | 3 | 120 | 40 | 4.67 |
| Within | 28 | 240 | 8.57 | |
| Total | 31 | 360 |
We reject the null hypothesis. There are significant differences among the four treatment means at α = 0.05.
• ANOVA table: Organized presentation of results
• Relationships: df_total = df_between + df_within
• Decision making: Compare F with critical value
Experimental design: The framework for organizing data collection to enable statistical analysis.
Factor: The variable being manipulated (independent variable) in an experiment.
Levels: The different values or categories of a factor.
Treatment groups: The groups defined by the different levels of the factor.
Response variable: The dependent variable being measured.
Randomization: Assigning subjects to treatment groups randomly to minimize bias.
- Design planning: Plan the experiment with appropriate sample sizes
- Randomization: Randomly assign subjects to treatment groups
- Data collection: Gather data according to experimental protocol
- Descriptive analysis: Calculate group statistics and visualize data
- Assumption checking: Test normality and equal variance assumptions
- ANOVA computation: Calculate all ANOVA components
- Statistical inference: Conduct hypothesis test
- Post-hoc analysis: If significant, determine which groups differ
- Practical interpretation: Connect statistical findings to real-world implications
• ANOVA principle: Compare systematic variation to random variation
• Multiple comparisons: Control family-wise error rate
• Effect size: Report practical significance along with statistical significance
• Assumption checking: Verify conditions before inference
• Interpretation: Connect statistical results to research question
• ANOVA can be extended to multiple factors and interactions
• Post-hoc tests (Tukey's HSD, Bonferroni) identify specific differences
• Effect size measures help assess practical significance
• Consider alternatives if assumptions are severely violated