2025

Effect Size

  • How big is an effect?

  • The answer depends on the context and the statistical test being performed.

  • In a single-sample test, we ask whether the true population mean of a variable is meaningfully different from a hypothesized value (e.g., \(C\)).

    \[ H0: \mu_{X} = C \\ H1: \mu_{X} < C \]

  • The effect size quantifies the magnitude of that difference, using a standardized metric.

  • This helps us assess whether a result is not just statistically significant, but also meaningful in practical terms.

Visualizing Effect Size

  • Effect size can be thought of as the area of overlap the raw data Null distribution and the raw data distribution estimated from the observed sample.

Effect size is computed from the Raw Data Distribution

  • Effect size is computed in the space of the raw data, not the test statistic.

  • It answers the question: How far is the observed mean from the hypothesized mean, in meaningful units?

  • In contrast, use test statistic distribution for:

    • Hypothesis testing (NHST decisions)
    • Power analysis

Cohen’s d (Single-Sample Version)

  • Cohen’s d is a widely used measure of effect size.

  • For a single-sample test (with known population standard deviation), it is:

\[ d = \frac{\bar{x} - \mu_0}{\sigma} \]

  • This tells us how far the sample mean is from the hypothesized mean, in standard deviation units.

  • For example, if \(\bar{x} = 1\), \(\mu_0 = 0\), and \(\sigma = 1\), then \(d = 1.0\).

Why effect size and not just p-value?

Key Takeaway

  • Effect size helps answer: Is the result meaningful, not just significant?

  • Most scientific journals now require reporting effect size estimates along with p-values etc.

  • We’ll learn different version of Cohen’s d and other effect size measures as we cover more statistical tests.