Understanding Post Hoc Comparisons

Introduction to Post Hoc Comparisons

  • Post hoc comparisons are analyses conducted after a significant ANOVA result to determine which specific groups differ from each other.

  • These comparisons can explore various hypotheses that were not explicitly planned in advance.

  • Distinction of comparisons:

    • Comparisons against the Control Group: Evaluate how each experimental group relates to the control.

    • Specific Comparisons: Select the groups to compare based on data trends or exploratory objectives.

    • All Possible Pairwise Comparisons: Analyze every combination of group comparisons.

Overview of Post Hoc Tests

  • Post Hoc Tests: Designed and conducted after data collection, focusing on controlling inflated family wise error rates during multiple comparisons.

  • Flexibility of Post Hoc Comparisons: Can range from broad (all possible comparisons) to specific (selective) comparisons.

  • Family Wise Error (FWE): Refers to the probability of making at least one Type I error across multiple comparisons.

Type I Error and Family Wise Error Rate

  • Type I Error (α): Occurs when the null hypothesis is incorrectly rejected when it is true.

    • Significance Level (α = 0.05): Indicates a 5% chance of making a Type I error per comparison.

  • Inflated Family Wise Error Rate: Involves increased risk of Type I error when making multiple comparisons, leading to a greater probability of incorrect conclusions.

Factors Influencing Family Wise Error

  • Number of Groups (k): More groups lead to more potential comparisons.

  • Number of Comparisons (c): Increases as group numbers increase, significantly impacting Type I error rates.

Example Calculation of Pairwise Comparisons

  • For 5 groups:

    • Total pairwise comparisons = 10 (calculated as (\binom{n}{2}=\frac{n(n-1)}{2}))

    • Explicit comparisons include:

    • Group 1 with Groups 2, 3, 4, 5 (4 comparisons)

    • Subsequent groups make remaining comparisons leading to a total of 10.

Visualization of Comparisons and Error Rates

  • As the number of groups increases, the number of comparisons grows rapidly:

    • Example: 18 groups result in 150 comparisons; thus, the likelihood of committing multiple Type I errors is high.

Family Wise Error Rate Formula and Adjustment

  • Family Wise Error Rate (FWER): Can be formulated as: FWER = 1 - (1 - \alpha^{\prime})^{c} Where (\alpha^{\prime} ) is the adjusted error rate per comparison.

    • The formula calculates the probability of making at least one Type I error across multiple comparisons.

  • Adjusting Alpha (α'): Adjusting the significance level can help control the inflated family wise error, increasing the likelihood of obtaining reliable results.

Comparison Strategies and Post Hoc Tests

  • Important to consider:

    • Which groups to compare (broad vs. selective comparisons).

    • How to manage inflated family wise error using appropriate statistical tests.

Types of Post Hoc Tests

  1. Dunnett's Test:

    • Utilized when comparing a control group to several experimental groups.

    • Focuses solely on comparisons of each experimental group to the control group.

    • Formula Utilized: Incorporates parts of a t-test and involves calculating t-statistics based on differences between group means.

      • Formula generally is:
        t = \frac{\overline{X}{c} - \overline{X}{j}}{\sqrt{\frac{MSE}{n}}}

      • Where ( \overline{X}{c} ) = mean of control group, ( \overline{X}{j} ) = mean of experimental group, and ( MSE ) = mean squared error from ANOVA summary table.

Conclusion and Implications

  • Understanding post hoc tests and family wise error is crucial for interpreting ANOVA results correctly.

  • Post hoc tests provide pathways to explore significant differences between group means while mitigating the risk of Type I errors.

  • These methods allow researchers to make informed decisions based on statistically robust comparisons, enhancing the validity of their findings in experimental settings.