Interpreting results after Greenhouse-Geisser or Huynh-Feldt adjustments

📚 Understanding Sphericity Corrections in Repeated Measures ANOVA

When conducting repeated measures ANOVA, an assumption called sphericity must be met. Sphericity implies that the variances of the differences between all possible pairs of related groups (levels of the independent variable) are equal. When sphericity is violated, the F-statistic in ANOVA can be artificially inflated, leading to a higher chance of a Type I error (false positive). The Greenhouse-Geisser and Huynh-Feldt corrections adjust the degrees of freedom to provide a more accurate p-value.

📜 History and Background

The need for sphericity corrections arose from the limitations of traditional ANOVA in handling repeated measures data. Greenhouse and Geisser proposed a conservative correction, while Huynh and Feldt offered a less conservative alternative, often providing a better balance between Type I and Type II error rates.

🔑 Key Principles

• 🔍 Assessing Sphericity: Mauchly's Test of Sphericity is used to determine if the assumption of sphericity is violated. A significant Mauchly's test (p < .05) indicates a violation.
• 🧪 Greenhouse-Geisser Correction: This correction estimates epsilon ($\epsilon$), which represents the degree to which sphericity is violated. The degrees of freedom are then multiplied by epsilon, reducing their value. This leads to a more conservative p-value.
• 📈 Huynh-Feldt Correction: The Huynh-Feldt correction also estimates epsilon ($\epsilon$), but it does so using a different method than Greenhouse-Geisser. It is generally less conservative and is preferred when epsilon is greater than .75.
• 🔢 Interpreting Corrected Output: In statistical software output (e.g., SPSS, R), look for the Greenhouse-Geisser or Huynh-Feldt corrected p-value. If this corrected p-value is less than your alpha level (usually .05), you reject the null hypothesis.
• ⚖️ Choosing a Correction: If Mauchly's Test is significant and epsilon is less than .75, use Greenhouse-Geisser. If epsilon is greater than .75, use Huynh-Feldt. Some researchers always report Greenhouse-Geisser due to its conservative nature.

🌍 Real-World Examples

Imagine a study examining the effect of a new drug on reaction time, measured at three time points (pre-treatment, post-treatment 1, post-treatment 2). A repeated measures ANOVA is used to analyze the data.

Example Output (Simplified):

In this example, the uncorrected p-value (0.02) is significant at an alpha of 0.05. However, Mauchly's test was significant, indicating a violation of sphericity. The Greenhouse-Geisser corrected p-value is 0.04, and the Huynh-Feldt corrected p-value is 0.03. Both corrected p-values are still less than 0.05, so you would still reject the null hypothesis and conclude that there is a significant effect of time on reaction time.

If the Greenhouse-Geisser corrected p-value had been greater than 0.05 (e.g., 0.06), you would fail to reject the null hypothesis, even though the uncorrected p-value was significant. This illustrates the importance of sphericity corrections.

💡 Conclusion

Greenhouse-Geisser and Huynh-Feldt corrections are essential tools for addressing violations of sphericity in repeated measures ANOVA. Properly interpreting the corrected p-values ensures more accurate and reliable conclusions are drawn from the data. Choosing the appropriate correction method (Greenhouse-Geisser or Huynh-Feldt) depends on the severity of the sphericity violation and the desired balance between Type I and Type II error rates.