week8.3_ANOVA
1. ANOVA Overview
Developed by L. H. Huhuhu, PhD, Assistant Professor of Business Administration at Washington and Lee University.
2. Example Scenario
An airline aims to determine which sound keeps callers on hold the longest:
(a) Advertisement about the airline
(b) Muzak
(c) Classical music (e.g., Vivaldi’s Four Seasons)
Rationale: Callers on hold lead to potential customer retention.
3. Conducting the Experiment
A randomized experiment played:
(a) Advertisement
(b) Muzak
(c) Classical music
Callers' hold times were recorded for analysis.
4. Issue with Multiple t-Tests
Inappropriate to use two-sample t-tests for:
Advertisement vs. Muzak
Advertisement vs. Classical
Classical vs. Muzak
Reason: Increases the risk of Type I error.
5. Confusion Matrix and Errors
Confusion Matrix Components:
True Positive: Correctly identifies a condition (e.g., pregnancy).
False Positive: Incorrectly identifies a condition.
Type I Error: Rejecting a true null hypothesis.
Type II Error: Failing to reject a false null hypothesis.
6. Type I and Type II Errors
When null hypothesis is true or false:
Type I Error (α): Rejecting null when true.
Type II Error (β): Failing to reject null when false.
Correct decision probabilities are equal to 1 - α and 1 - β, respectively.
7. Controlling for Confidence with Multiple Comparisons
Increased comparisons raise the error risk:
Example: If g = 10, there are 45 pairs for comparison.
Error probability per comparison leads to higher average incorrect intervals.
8. Analysis of Variance (ANOVA)
Hypothesis Testing:
Null Hypothesis (H0): Average hold times are equal (μ1 = μ2 = μ3).
Alternative Hypothesis (H1): At least two population means are different.
9. F-Test
F-statistic formula:
F = Between-group variability / Within-group variability
Larger variability between groups implies a larger F test statistic.
10. Degrees of Freedom
Calculating degrees of freedom:
Between-groups df1 = g - 1
Within-groups df2 = N - g (N = total observations; g = groups)
11. Example Calculation
For 3 groups and 15 total samples:
df1 = 3 - 1 = 2
df2 = 15 - 3 = 12
12. F-Value Calculation
F Calculation Steps:
Between Group Mean Square = Between sum of squares / (g-1)
Within Group Mean Square = Within sum of squares / (N-g)
13. ANOVA Table Structure
Standard metrics included in an ANOVA table:
Source | Sum of Squares (SS) | Degrees of Freedom (DF) | Mean Square (MS) | F-value | p-value | Effect Size (np2)
Values guide the conclusion on the null hypothesis.
14. Interpretation of Results
If p-value < 0.05, reject H0, implying the effects of the sounds are not equal.
15. Assumptions for ANOVA
Independence Assumption for groups and data.
Equal Variance Assumption: Works well if largest/ smallest standard deviation ratio <2.
Normality Assumption: Less of a concern with larger samples.
16. Post-Hoc Tests
Purpose: Multiple comparisons between groups (e.g., Advertisement vs. Muzak).
T-test and confidence intervals may be utilized.
17. Multiple Comparison Methods
Tailored to assess confidence levels for sets of comparisons, focusing on:
Bonferroni Method: Adjusts threshold p-value based on number of comparisons.
Tukey Method: Offers slightly narrower confidence intervals, potentially more complex.
18. Example Outputs from Tukey Method
Direct comparisons yield p-values indicating significance of group differences:
Comparisons include pairs such as C–A, M–A, M–C, etc.