What do I use
Medicine and clinical research
Clinical trials, diagnostic studies, case-control and cohort work, hospital audit.
Your question, and what answers it
With the effect size and its interval, because a significant difference of half a point on a pain scale is not a clinical finding.
In ZStat: open your study, go to Analysis, and choose independent t.
A different question from the one above, needing a margin fixed before you look. A non-significant difference is not evidence of equivalence.
In ZStat: open your study, go to Analysis, and choose non inferiority.
Sensitivity, specificity and the predictive values at the prevalence of the setting you are recommending it for.
In ZStat: open your study, go to Analysis, and choose diagnostic accuracy.
DeLong's test, because both were measured on the same patients and treating them as independent hides a real improvement.
In ZStat: open your study, go to Analysis, and choose roc comparison.
Relative risk, odds ratio and number needed to treat, with intervals.
In ZStat: open your study, go to Analysis, and choose risk measures.
Survival curves with the log-rank test, then Cox regression when you need to adjust for other factors.
In ZStat: open your study, go to Analysis, and choose kaplan meier.
Matching on measured confounders, with the balance table that shows whether it worked.
In ZStat: open your study, go to Analysis, and choose propensity matching.
What goes wrong in medicine and clinical research
- p values in the baseline characteristics table of a randomised trial. Any imbalance is by definition chance, and CONSORT has advised against it since 2001.
- Intention-to-treat as the only analysis in a non-inferiority trial. It is conservative in a superiority trial and the opposite here.
- Reporting sensitivity and specificity to a clinician who needs predictive values.