AI summary by TwinMind
StatQuickie: Which t test to use
StatQuest with Josh Starmer
5m 10s
T-Test Categories Overview
- Two main categories: paired and unpaired t-tests
- Selection depends on data structure and measurement type
Paired T-Tests
- Used for before/after measurements from same test subjects
- Example: blood pressure drug study
- Measure blood pressure before drug administration
- Give drug to subjects
- Measure blood pressure after drug treatment
- Each individual has pair of measurements (before + after)
- Use when data comes in matched pairs from same subjects
Unpaired T-Tests
- Used when comparing two separate groups without paired measurements
- Example: height comparison between Group A and Group B
- Group A: measure heights of one set of people
- Group B: measure heights of different set of people
- No overlap between groups, no paired measurements
Unpaired T-Test Subcategories
- Equal variance assumption: assumes variation in Group A equals variation in Group B
- Unequal variance assumption: allows different variance levels between groups
- Recommendation: use unequal variance test
- More conservative approach
- If data passes more conservative test = rock solid results
One-Tailed vs Two-Tailed T-Tests
Two-Tailed (Two-Sided) T-Tests
- Tests both directions of difference
- Example with height data:
- Tests if Group A significantly higher than Group B
- Also tests if Group A significantly smaller than Group B
- Agnostic approach - no preconceived notions about direction
- Lets data speak for itself
One-Tailed T-Tests
- Less conservative approach
- Requires pre-specifying expected direction
- Example: must predict Group A will be higher than Group B
- Generally not recommended for academic journals
Final Recommendations
For Paired Data
- Use paired t-test when subjects have before/after measurements
For Unpaired Data
- Use unpaired t-test for separate groups
- Select unequal variance assumption (more conservative)
- Use two-tailed test (more conservative, data-driven)
Conservative Approach Benefits
- More stringent testing standards
- If data passes conservative tests = rock solid results
- Preferred for academic credibility
