8.4 Two-Way ANOVA and Factorial Designs

Two-way ANOVA lets us study how two factors affect an outcome together. It's like looking at how both fertilizer and sunlight impact plant growth, not just one at a time. This method reveals if factors work independently or interact in complex ways.

The analysis involves calculating various sums of squares and F-ratios to test for main effects and interactions. It's a powerful tool, but it requires careful planning and interpretation, especially when dealing with significant interactions between factors.

Two-Way ANOVA and Factorial Designs

Factorial designs and interaction effects

• Factorial designs study effects of multiple factors simultaneously allowing examination of main and interaction effects
• Main effects measure impact of one independent variable on dependent variable averaged across other factors
• Interaction effects occur when effect of one factor depends on level of another revealing complex relationships
• Two-way ANOVA analyzes impact of two independent variables on one dependent variable extending one-way ANOVA

Conducting two-way ANOVA analysis

• Steps for conducting two-way ANOVA:
  1. Formulate hypotheses for main effects and interaction
  2. Collect and organize data
  3. Calculate sum of squares (SS) for each variation source ($SS_{total}$, $SS_{factor A}$, $SS_{factor B}$, $SS_{interaction}$, $SS_{error}$)
  4. Compute degrees of freedom (df) for each source
  5. Calculate mean squares (MS) by dividing SS by df
  6. Compute F-ratios for main effects and interaction
  7. Determine p-values for each F-ratio
• Assumptions include independence of observations, normality of residuals, and homogeneity of variances

Interpretation of two-way ANOVA results

• Main effects interpretation:
  • Significant main effect indicates factor has overall effect on dependent variable (crop yield, test scores)
  • Non-significant main effect suggests no overall effect of the factor
• Interaction effects interpretation:
  • Significant interaction shows effect of one factor depends on level of the other (fertilizer type and soil pH)
  • Non-significant interaction implies factors act independently
• Effect size measures:
  • Partial eta-squared quantifies proportion of variance explained by each factor
  • Omega-squared offers less biased alternative to eta-squared
• Post-hoc tests used for pairwise comparisons when main effects are significant (Tukey's HSD, Bonferroni correction)
• Interaction plots graphically represent interaction between factors with non-parallel or crossing lines suggesting potential or strong interaction

Advantages vs limitations of factorial designs

• Advantages:
  • Efficiency in studying multiple factors simultaneously (temperature and humidity on plant growth)
  • Ability to detect interaction effects
  • Increased external validity
  • Cost-effective compared to multiple single-factor studies
• Limitations:
  • Increased complexity in design and analysis
  • Potential for confounding effects if not properly controlled
  • Larger sample sizes required for adequate statistical power
  • Difficulty interpreting higher-order interactions (3+ factors)
• Considerations for use:
  • Research question suitability
  • Available resources and sample size
  • Potential for meaningful interactions between factors (drug dosage and patient age)
  • Balance between complexity and interpretability