Assumptions and Diagnostics for Difference Intervals
What you'll learn: How to verify that your data meets the requirements before calculating confidence intervals for differences between groups.
Why Assumptions Matter
Just like you wouldn't build a house on unstable ground, you shouldn't construct confidence intervals without checking your statistical foundation. When comparing two groups (whether means or proportions), specific assumptions must hold for your interval to be trustworthy.
The Three Key Assumptions
1. Independence
Your observations must be independent—both within and between groups. One person's measurement shouldn't influence another's. Paired data violates independence between groups (use paired intervals instead).
2. Normality
For difference-in-means intervals, each group's data should follow a normal distribution, especially with small samples. You've already learned to check this using Q-Q plots, histograms, and tests like Shapiro-Wilk. The Central Limit Theorem helps when sample sizes are larger (n ≥ 30 per group).
3. Equal Variances (Homogeneity)
When using the pooled variance approach, both groups should have similar variability. If variances differ substantially, use the unpooled variance approach instead. Visual comparison of spread in boxplots can reveal obvious differences.
Diagnostic Workflow
- Plot your data — boxplots and histograms for each group
- Check normality — Q-Q plots or Shapiro-Wilk test per group
- Assess variance equality — compare standard deviations or use Levene's test (not yet covered formally, but intuitive: do the spreads look similar?)
- Verify independence — review your sampling method and data collection
Key Takeaway: Always verify independence, normality (for means), and equal variances before constructing difference confidence intervals—violating these assumptions can make your intervals unreliable and your conclusions wrong.