One-Way ANOVA: The Model and Assumptions
What you'll learn: How to formally represent a one-way ANOVA problem using mathematical notation and understand the key assumptions required for valid inference.
The One-Way ANOVA Model
After introducing ANOVA conceptually, we now formalize it mathematically. One-way ANOVA compares means across k groups (or treatments). Each observation is modeled as:
y_ij = μ_i + ε_ij
Where:
- y_ij is the j-th observation in the i-th group
- μ_i is the true mean of group i
- ε_ij is the random error for that observation
Think of it like this: every data point equals its group's "true average" plus some random noise.
An alternative representation uses an overall mean:
y_ij = μ + τ_i + ε_ij
Where μ is the grand mean and τ_i is the "treatment effect" (how much group i differs from the grand mean).
The Three Critical Assumptions
For one-way ANOVA to produce valid results, the errors ε_ij must satisfy:
1. Normality
The errors (and thus observations within each group) follow a normal distribution. Each group is drawn from a normally distributed population.
2. Equal Variances (Homoscedasticity)
All groups share the same variance: σ²₁ = σ²₂ = ... = σ²_k. The spread is consistent across groups.
3. Independence
Observations are independent of each other, both within and between groups. One measurement doesn't influence another.
These mirror the assumptions you learned for regression: normality, constant variance, and independence.
Key Takeaway: One-way ANOVA models each observation as its group mean plus random error, requiring normally distributed, equally variable, and independent errors for valid inference about group mean differences.