Understanding the Correlation Ratio E²

The correlation ratio, often denoted by E² (eta squared), is a statistical measure that describes the proportion of variance in a dependent variable that is explained by the categories of an independent variable. It is used when you want to measure the relationship between a quantitative variable (like height, income, test scores) and a categorical variable (like gender, treatment group, geographic region). It's essentially a measure of effect size in analysis of variance (ANOVA).

To understand the range of the correlation ratio E², let's look at how it is calculated based on the partitioning of variance.

Variance Decomposition and E² Calculation

The total variance of the dependent quantitative variable ($y$) can be split into two main components:

• Between-group variance ($\sigma_B^2$): This is the variance of the group means around the overall mean. It represents the part of the total variance that is due to the differences between the categories of the independent variable.
• Within-group variance ($\sigma_W^2$): This is the average variance within each category. It represents the part of the total variance that is due to variation within each group, often considered random error or variation not explained by the categorical variable.

The total variance ($\sigma_y^2$) is the sum of the between-group variance and the within-group variance:

$\sigma_y^2 = \sigma_B^2 + \sigma_W^2$

The correlation ratio E² is defined as the ratio of the between-group variance to the total variance:

$E^2 = \frac{\sigma_B^2}{\sigma_y^2} = \frac{\sigma_B^2}{\sigma_B^2 + \sigma_W^2}$

Determining the Range of Correlation Ratio E²

Since variance is a measure of spread, it is always non-negative. Thus, $\sigma_B^2 \ge 0$, $\sigma_W^2 \ge 0$, and $\sigma_y^2 \ge 0$.

Considering the formula $E^2 = \frac{\sigma_B^2}{\sigma_B^2 + \sigma_W^2}$:

• The numerator ($\sigma_B^2$) is always non-negative.
• The denominator ($\sigma_B^2 + \sigma_W^2$) is the total variance ($\sigma_y^2$) and is also always non-negative.
• The numerator ($\sigma_B^2$) is a component of the denominator ($\sigma_B^2 + \sigma_W^2$). Therefore, the numerator can never be greater than the denominator. $\sigma_B^2 \le \sigma_B^2 + \sigma_W^2$.

Based on these points, the ratio $\frac{\sigma_B^2}{\sigma_B^2 + \sigma_W^2}$ must be between 0 and 1, inclusive.

• Lower Bound (E² = 0): This occurs when the between-group variance ($\sigma_B^2$) is zero. This means there are no differences between the means of the groups. All the variation is within the groups ($\sigma_W^2$).
• Upper Bound (E² = 1): This occurs when the within-group variance ($\sigma_W^2$) is zero. This means there is no variation within each group. All the variation in the dependent variable is accounted for by the differences between the group means.

Therefore, the correlation ratio E² always takes values between 0 and 1.

Analyzing the Given Options

Let's evaluate the provided options for the range of the correlation ratio E²:

• Option 1: E takes the values between -1 and 1. This is incorrect. E² represents a proportion of variance, which cannot be negative. While Pearson's correlation coefficient ($r$) ranges from -1 to 1, E² is conceptually different and non-negative.
• Option 2: E takes the values between 0 and ∞. This is incorrect. As explained, the between-group variance cannot exceed the total variance, so the ratio E² cannot be greater than 1.
• Option 3: E takes the values between -∞ and ∞. This is incorrect. Variance ratios are always non-negative and bounded above by 1.
• Option 4: E takes the values between 0 and 1. This is correct. The mathematical definition and interpretation of E² as a proportion of variance explained confirm this range.

Therefore, the correlation ratio E² is always a value between 0 and 1, inclusive.

Revision Table: Key Aspects of Correlation Ratio E²

Additional Information on Correlation Ratio in Statistics

While E² measures the *proportion* of variance explained, its square root, E (eta), can also be used, but E² is more commonly reported as it directly relates to the variance decomposition. The correlation ratio is particularly useful because it does not assume a linear relationship between variables, unlike Pearson's correlation coefficient, which is suitable for two quantitative variables and linear relationships. E² captures any kind of relationship between the mean of the quantitative variable and the categories of the independent variable. However, it only measures the extent to which the categorical variable explains the variance of the quantitative variable, not the direction of the relationship (as there isn't a simple direction with categories).