Methods
Grouped by outcome type. Not sure which one you need? Start the test chooser.
Continuous outcomes
BP, HbA1c, length of stay
- ANCOVA (adjusting for baseline)Compares groups on a continuous follow-up outcome while adjusting for the baseline value of that same outcome.Adjusted difference in means
- Friedman testCompares a skewed measurement or an ordered score across three or more conditions in the same patients, using ranks.
- Kruskal-Wallis testRank-based test of whether a skewed measurement or an ordered score differs between three or more independent groups.
- Linear mixed modelAnalyzes a continuous outcome measured repeatedly in the same patients, or in clustered patients, accounting for the correlation between measurements.Difference in means
- Linear regressionEstimates how a continuous outcome changes with one or more factors, holding the others fixed.Difference in means per unit / vs reference
- Mann-Whitney U test (Wilcoxon rank-sum test)Tests whether values in one of two independent groups tend to be higher than in the other, using ranks instead of raw values. Used for skewed measurements and ordered scores.Probability of superiority (and the Hodges-Lehmann estimate)
- One-sample t-testCompares the mean of a measurement in one group with a fixed reference value that was specified in advance.
- One-way ANOVATests in one step whether the mean of a measurement differs between three or more independent groups.Differences in means (post hoc)
- Paired t-testCompares the mean of a measurement between two time points or conditions in the same patients, using each patient's difference.Mean within-patient difference
- Pearson correlationMeasures how strongly two numeric measurements are related in a straight line, from −1 to 1.
- Sign test or Wilcoxon signed-rank testCompares the centre of a skewed measurement in one group with a known reference value, using signs or ranks.
- Spearman rank correlationMeasures how strongly two measurements move together using ranks, so skew and curves that keep one direction (monotone) are fine.
- Summary-measures analysisReduce each patient to one number, then use a standard two-group test
- Two-way ANOVACompares the mean of a measurement between the groups formed by two factors, and tests whether the effect of one factor depends on the other (an interaction).Differences in means, and the interaction
- Unpaired t-test on log-transformed dataCompares two independent groups on the log scale for right-skewed data, and reports a ratio of geometric means.Ratio of geometric means
- Welch's t-testCompares the mean of a measurement between two independent groups, without assuming both groups have the same spread (equal variances). It gives practically the same answer as the classic t-test when the spreads are equal.Difference in means
- Wilcoxon signed-rank testCompares paired measurements using the ranks of the within-patient differences, for skewed differences or ordered scores.Pseudo-median of differences
Binary outcomes
Death, complication, readmission
- Chi-square testCompares the proportion of patients with a yes/no outcome between independent groups.Risk difference / risk ratio
- Chi-square test (r × c table)Compares a category or yes/no outcome across three or more independent groups.
- Cochran's Q testCompares a yes/no outcome across three or more conditions in the same patients, as an extension of McNemar's test.
- Conditional logistic regressionEstimates an odds ratio in matched case-control data, adjusted for other variables, by comparing within matched sets.Odds ratio
- Fisher's exact testCompares the proportion of patients with a yes/no outcome between independent groups when some groups have very few events.Odds ratio
- GEE and mixed-effects logistic regressionAnalyzes a yes/no outcome measured repeatedly in the same patients, or in clustered patients, accounting for the correlation within patients or clusters.Odds ratio (marginal vs conditional)
- Log-binomial regressionEstimates a prevalence ratio in a cross-sectional study or a risk ratio in a cohort or trial with fixed follow-up. Poisson with robust SE if it fails to converge.Prevalence ratio or risk ratio
- Logistic regressionEstimates how the odds of a yes/no outcome change with one or more factors, holding the others fixed.Odds ratio
- McNemar testCompares a yes/no outcome measured twice in the same patients, or in matched pairs, using only the pairs that changed.Difference in paired proportions
- z-test for a single proportionCompares the proportion of patients with a yes/no outcome in one group with a known reference proportion.
Categorical outcomes
Blood group, other unordered categories
Time-to-event outcomes
Survival, relapse, time to discharge
- Competing risks: cumulative incidence and Fine-GrayEstimates the cumulative incidence of an event when another event can prevent it, instead of treating the competing event as censoring.Subdistribution hazard ratio; cumulative incidence
- Cox proportional hazards regressionEstimates how the hazard of an event changes with one or more factors, holding the others fixed.Hazard ratio
- Kaplan-Meier curves and log-rank testDescribes the probability of staying event-free over time in each group, and compares groups with the log-rank test.Survival probability at time t; median survival
Ordinal outcomes
mRS, pain scores, Likert items
- Friedman testCompares a skewed measurement or an ordered score across three or more conditions in the same patients, using ranks.
- Kruskal-Wallis testRank-based test of whether a skewed measurement or an ordered score differs between three or more independent groups.
- Mann-Whitney U test (Wilcoxon rank-sum test)Tests whether values in one of two independent groups tend to be higher than in the other, using ranks instead of raw values. Used for skewed measurements and ordered scores.Probability of superiority (and the Hodges-Lehmann estimate)
- Spearman rank correlationMeasures how strongly two measurements move together using ranks, so skew and curves that keep one direction (monotone) are fine.
- Wilcoxon signed-rank testCompares paired measurements using the ranks of the within-patient differences, for skewed differences or ordered scores.Pseudo-median of differences
- Proportional odds (ordinal logistic) regressionCompares an ordered score between groups while adjusting for other variables, giving one common odds ratio across the cut-points of the score.Common odds ratio
Counts and rates
Exacerbations, admissions per year
Accuracy and agreement
A new test, two raters, two devices
- Bland-Altman plot and ICCShows how closely two methods that measure the same quantity agree, with the average difference and the range in which most differences fall.Bias and 95% limits of agreement; ICC
- Cohen's kappaMeasures agreement between two raters or methods that assign categories to the same patients, beyond what chance alone would give.Kappa (weighted for ordered categories)
- Diagnostic accuracy: sensitivity, specificity, ROCQuantifies how well a new test identifies patients with the condition, compared with a reference standard.Sensitivity, specificity, AUC with CI
Prediction
Risk for individual patients