Module 06 Activity
Scenario
Compare groups with the method the design and the data actually support.
What you build
A comparison with its assumptions checked and its power stated.
Steps
- Decide whether the comparison is one-sample, two-sample, paired, or across three or more groups, and justify it from the design.
- For paired data, report the between-unit SD and the within-unit change SD so the gain from pairing is visible.
- For proportions, report the absolute difference with its interval and the relative change alongside.
- For categorical data, report the chi-square test, the expected counts and the standardised residuals — and switch to Fisher when expected counts fall below about 5.
- For three or more groups, run the F-test and follow with Tukey's HSD, never with unadjusted pairwise tests.
- Before interpreting any null result, state the effect size the study could have detected.
Evidence to hand in
- The design justification for the test chosen.
- The two SDs for any paired analysis.
- The absolute and relative differences for any proportion.
- The chi-square output with expected counts and residuals.
- The F-test and its adjusted follow-up.
- The detectable effect size behind any null result.
Review checklist
- Paired data is analysed as paired.
- Welch's test is used rather than the equal-variance version.
- Chi-square is reported with residuals, and expected counts were checked.
- Pairwise comparisons after ANOVA are adjusted.
- Every null result is accompanied by what the study could have detected.