Independent Measures ANOVA: Mean Squares and the F-Ratio Distribution

Introduction to Independent Measures ANOVA and Session Context

• This material covers the second day of Chapter 12: Independent Measures ANOVA.

• While conceptually the second day of face-to-face instruction, this content follows three previous videos from the first day.

• The lecture builds on foundational knowledge, transitioning from conceptual review to new calculation procedures and their specific purposes.

The Procedural Goal and Calculation Flow of ANOVA

• The ultimate objective in ANOVA is to calculate $F_{\text{obtained}}$.

• The calculation process is structured like a funnel, where data transitions from left to right across the ANOVA summary table:

  • SS (Sum of Squares): The process begins with calculating Sum of Squares, specifically $SS_{\text{between}}$, $SS_{\text{within}}$, and $SS_{\text{total}}$. These represent the sum of squared deviations, a concept introduced in Chapter 4.

  • DF (Degrees of Freedom): Every $SS$ value has a corresponding $df$ value ($df_{\text{between}}$, $df_{\text{within}}$, and $df_{\text{total}}$).

  • MS (Mean Squares): These two components (SS and DF) are combined to produce two $MS$ values ($MS_{\text{between}}$ and $MS_{\text{within}}$).

  • F-Obtained: The final step is to combine the two $MS$ values to reach the goal of $F_{\text{obtained}}$.

Conceptual Definition of Mean Squares ($MS$)

• Mean Squares is described as a "new name for an old thing."

• It is mathematically identical to Sample Variance ($s^2$).

• Historical Formula from Chapter 4:

  • Simple variance was defined as:     $s^2 = \frac{SS}{n-1}$

  • Since $n-1$ is the simplest formula for degrees of freedom ($df$), variance can be expressed as:     $MS = \frac{SS}{df}$

• ANOVA Application:

  • In ANOVA, the subscripts for $SS$ and $df$ must always match to calculate the specific $MS$:

  • $MS_{\text{between}} = \frac{SS_{\text{between}}}{df_{\text{between}}}$

  • $MS_{\text{within}} = \frac{SS_{\text{within}}}{df_{\text{within}}}$

• Because $MS$ is simply variance, the $F$-ratio is technically a ratio of sample variances (the division of one variance by another).

Terminology Changes in ANOVA Psychology Research

• Researchers change standard terms when using ANOVA:

  • Factor: Represents the Independent Variable or general treatment (e.g., the "dose of a pill").

  • Levels: Refers to the specific conditions or groups within the factor (e.g., "Placebo," "Low Dose," and "High Dose").

• While previously called "groups" or "treatments" in $t$-tests, ANOVA nomenclature prioritizes "Factors" and "Levels."

Understanding Variability: The "Million and One Reasons" Logic

• ANOVA distinguishes between within-group and between-group variability through different contributory factors:

• Within-Group Variability ($MS_{\text{within}}$):

  • Measures deviations between individuals inside the same level (e.g., comparing students within the same school).

  • If Student 1 at School 1 scores a $2$ and Student 2 at School 1 scores a $4$, they are different because of Random Variability.

  • There are a "million reasons" for this differences, including study habits, prior math knowledge, support systems, babysitting duties, or personal study skills.

• Between-Group Variability ($MS_{\text{between}}$):

  • Measures the extent to which group means (averages) differ from one another.

  • When comparing a student at one school (score of $4$) to a student at a different school (score of $5$), there are a "million and one" reasons for the difference.

  • These include all the previous random reasons PLUS the effect of the factor (e.g., the effect of the school itself or the effect of the drug dose).

The Logic of the F-Ratio Expected Value

• General Structure:     $F = \frac{\text{Random Variability} + \text{Treatment Effect}}{\text{Random Variability}}$

• Expected Value Under $H_0$:

  • In $z$-tests and $t$-tests, the expected value when the null hypothesis ($H_0$) is true is $0$. This is because the numerator (the difference between means) is expected to be $0$ if the treatment does nothing.

  • In ANOVA, the expected value of $F$ when $H_0$ is true is $1$.

  • If $H_0$ is true, there is no treatment effect (Treatment Effect = $0$).

  • Therefore, the formula becomes:     $F \approx \frac{\text{Random Variability}}{\text{Random Variability}} \approx 1$

  • When the numerator and denominator consist of the same set of random factors, dividing them yields a value of approximately $1$.

Characteristics of the F-Distribution

• Skewness: Unlike the symmetrical bell curves of $z$ or $t$, the $F$-distribution is right-skewed, characterized by a long right-hand tail.

• Lower Bound: The distribution has a hard cutoff at $0$. It cannot contain negative values.

• Why No Negative Values?:

  • $F$ is a ratio of variances ($MS$).

  • Variance deals with Sum of Squares ($SS$), which mathematically can never be negative.

  • Dividing two non-negative values ($MS_{\text{between}} / MS_{\text{within}}$) will never result in a negative number.

• Possibility of Zero:

  • An $F$-value of $0$ is theoretically possible but practically unlikely.

  • It would require $MS_{\text{between}}$ to be $0$, which only occurs if the means of all groups are perfectly identical.

  • For instance, if three different drug doses produced identical average reaction times down to the decimal, the between-group variability would be zero, resulting in $F = 0$.