Statistical Analysis and Interpretation
What the exam tests
Statistical analysis and interpretation tests descriptive versus inferential statistics, confidence intervals, hypothesis tests, regression, survival analysis, missing data, and multiplicity. You must interpret effect sizes and uncertainty, not only p-values.
Core rules
- A p-value is the probability of data at least as extreme under the null; it is not the probability the null is true or the effect is large.
- Confidence intervals show estimate precision and are usually more informative than a dichotomized p-value.
- Report absolute effects: risk difference, number needed to treat, and baseline risk.
- Pre-specify the primary analysis; adjust or account for multiple comparisons.
- Missing data need a mechanism assumption; complete-case analysis can bias results.
Worked example
In a trial, readmission is 18% with usual care and 12% with an intervention. The absolute risk reduction is 6 percentage points; the relative risk is 0.67, and NNT is about 17. If the 95% CI for the risk difference is 1 to 11 points, the result is statistically significant but the benefit could be small. The reasoning: the relative reduction sounds larger than the absolute benefit, and the CI shows uncertainty. Report all three.
Traps
- Interpreting p=0.049 as clinically important and p=0.051 as no effect.
- Testing baseline characteristics for significance in an RCT.
- Ignoring multiplicity after many subgroup tests.
- Treating odds ratios as risk ratios when the outcome is common.