Understanding One-Way ANOVA: Logic, Assumptions, and Application

1. Between-Groups Variability ($SS_{BG}$)

This represents the systematic difference between the group means. If we are testing a new drug, the $SS_{BG}$ captures the effect of that drug. It tells us how much the group means deviate from the overall grand mean.

2. Within-Groups Variability ($SS_{WG}$ or $SS_E$)

Also known as error variance, this represents the random, unsystematic variation among individuals within each specific group. It is the "noise" in our data that exists regardless of whether our treatment works or not.

The relationship is expressed mathematically as: $SS_T = SS_{BG} + SS_{WG}$

The F-Ratio: The Core Logic of ANOVA

The goal of the ANOVA is to determine if the systematic difference ($SS_{BG}$) is large enough to be distinguished from the random noise ($SS_{WG}$). We do this by calculating the F-ratio:

$F = \frac{\text{Variance Between Groups}}{\text{Variance Within Groups}}$

The logic follows two possible scenarios regarding our hypotheses:

• The Null Hypothesis ($H_0$): All population means are equal ($\mu_1 = \mu_2 = \mu_3...$). In this case, the only reason the group means would differ is due to random chance. Therefore, the between-groups estimate and the within-groups estimate should be roughly equal, resulting in an F-ratio of approximately 1.
• The Research/Alternative Hypothesis ($H_1$): At least one population mean is different. Here, the between-groups estimate includes both random chance and the treatment effect. This makes the numerator larger than the denominator, resulting in an F-ratio greater than 1.

Exam Tip: If you calculate an F-value and it is less than 1, you have likely made a calculation error, as the F-distribution only contains positive values.

Key Assumptions for Validity

For the F-ratio to provide a valid test, four assumptions must be met:

1. Scale of Measurement: The dependent variable must be interval or ratio scale.
2. Normality: The populations from which samples are drawn should be normally distributed.
3. Homogeneity of Variance (Homoscedasticity): All populations should have approximately the same variance.
4. Independence: All observations must be independent (no paired or matched samples).

Worked Example: The Jury Perception Study

To see this in practice, consider a study investigating how knowledge of a defendant's criminal record affects jury perception.

The Setup: Researchers assigned 15 volunteers to three groups:

1. Criminal Record Group: Told the defendant has a history of crime.
2. Clean Record Group: Told the defendant has a clean record.
3. No Information Group: Given no info on the record.

The dependent variable was a guilt rating from 1 to 10.

Hypotheses:

• $H_0: \mu_1 = \mu_2 = \mu_3$ (Criminal record knowledge has no effect on guilt ratings).
• $H_1$: At least one mean is different.

The Decision Process: After calculating the Sum of Squares for both between-groups and within-groups, the researcher determines the Degrees of Freedom (df):

• $df_{between} = \text{number of groups} - 1 = 3 - 1 = 2$
• $df_{within} = \text{total participants} - \text{number of groups} = 15 - 3 = 12$

Using an F-table at the $\alpha = .05$ significance level, the researcher finds the critical value for $F(2, 12)$ is 3.89.

Conclusion:

• If the calculated $F_{stat} > 3.89$, we reject the null hypothesis and conclude that the criminal record information significantly changed jury perceptions.
• If $F_{stat} \leq 3.89$, we fail to reject the null hypothesis, concluding there is no evidence that the record information affected the ratings.