Summary of Comparing Two Means
Key Learning Goals
Differentiate between paired & unpaired designs.
Conduct paired and unpaired two-sample t-tests & understand their assumptions and alternatives.
Estimate significance from overlap of Confidence Intervals.
Test if between group variances differ, and learn how to proceed if they do.
Types of Comparisons
Variables: 1 numerical & 1 categorical with 2 groups.
Goals: Compare mean of numerical variable between two groups.
Test Versions: Paired comparisons and unpaired comparisons.
Paired vs. Unpaired Comparisons
Paired:
Treatments applied to every sample unit under different conditions.
Measurements on the same unit are not independent.
Control for extra variation.
Unpaired:
Each group is an independent random sample.
Measurements on different units are independent.
Easier to collect data externally.
Procedure for Paired t-Test
Hypothesis Testing Steps:
Null hypothesis: mean of differences = 0.
Alternative hypothesis: mean of differences ≠ 0.
Data Collection: Sample data, calculate the differences.
Calculate Test Statistic: Use mean differences and standard deviation.
Determine Null Sampling Distribution: Compute degrees of freedom (df = n - 1).
Calculate P-value: Compare to significance level (α).
Conclusion: Biological significance based on statistical results.
Unpaired t-Test Procedure
Hypothesis Testing Steps:
Null hypothesis: means of two groups are equal (D = 0).
Alternative hypothesis: means are different (D ≠ 0).
Calculate Test Statistic:
Use means and standard error of differences.
Determine Null Sampling Distribution: Calculate df as df1 + df2 - 2.
Calculate P-value: Compare to significance level (α).
Conclusion: Assess if results are statistically significant.
Assumptions of t-Tests
Random samples.
Normal distributions for populations.
Equal variances; robust for n > 30.
Variance Comparison
Use F test to check if variances differ:
Null hypothesis: variances are equal.
Critical value determined from F distribution.
Summary
Paired Designs: Analyze mean differences directly between pairs.
Unpaired Designs: Compare group means, requires independent samples.
Proper statistical conclusions stem from direct group comparisons and adherence to assumptions for the tests used.