Find Your Test
Select the variable you want to test against your outcome — organised the same way as your Data Collection Form. The recommendation updates live as you answer.
1Which variable do you want to test against your outcome?
Demographics
Age Continuous
Gender Nominal
Weight Continuous
BMI Continuous
Baseline Parameters / Classification / Study-Specific Details / Another Outcome
Baseline Parameters
Classification into Type
Study-Specific Details
Another Outcome Variable
2What type of data is your Outcome?
Nominal e.g. complication type
Dichotomous e.g. union yes/no
Ordinal e.g. functional grade
Continuous e.g. score, ROM, blood loss
3Is the continuous variable normally distributed?
Yes — Normalparametric
No — Not Normalnon-parametric
Not Checked Yet
Testing
Special Situations
These don't fit the group-comparison wizard above — click to expand.
Time-to-Event / Survival Data — e.g. time to union, implant failure, reoperation▶
| Purpose | Test |
|---|---|
| Describe survival over time | Kaplan-Meier survival curve |
| Compare 2 or more groups | Log-rank test |
| Adjust for confounders | Cox proportional hazards regression |
Diagnostic Accuracy — new test vs a reference standard▶
Sensitivity, specificity, positive predictive value, negative predictive value against the reference standard; overall discrimination summarised with a ROC curve and AUC.
Agreement / Reliability — inter-observer or intra-observer▶
| Data Type | Test |
|---|---|
| Categorical ratings, 2 raters | Cohen's kappa |
| Categorical ratings, >2 raters | Fleiss' kappa |
| Continuous measurements | Intraclass correlation coefficient (ICC), with Bland-Altman plot to show the pattern of disagreement |
Multivariable / Adjusted Analysis — controlling for confounders▶
| Outcome Type | Test |
|---|---|
| Continuous | Multiple linear regression |
| Binary | Logistic regression |
| Ordinal | Ordinal (proportional odds) logistic regression |
| Count (e.g. number of complications) | Poisson or negative binomial regression |
| Time-to-event | Cox regression |
Almost always needed once a significant univariate association is found, to check it survives adjustment for confounders.
Master Reference Table
| Data Type | 2 Independent Groups | 2 Related Groups | >2 Independent Groups | >2 Related Groups | Association / Correlation |
|---|---|---|---|---|---|
| Nominal / Dichotomous | Chi-square test (Fisher's exact if any expected cell <5) | McNemar's test | Chi-square test of independence | Cochran's Q test | Phi coefficient / Cramer's V; Odds ratio, Relative risk |
| Ordinal | Mann-Whitney U test | Wilcoxon signed-rank test | Kruskal-Wallis test (+ post-hoc Dunn's) | Friedman test (+ post-hoc) | Spearman's rho / Kendall's tau; Chi-square trend test |
| Continuous — Normal | Independent (unpaired) t-test | Paired t-test | One-way ANOVA (+ post-hoc Tukey) | Repeated-measures ANOVA | Pearson's correlation; Linear regression |
| Continuous — Non-normal | Mann-Whitney U test | Wilcoxon signed-rank test | Kruskal-Wallis test | Friedman test | Spearman's rho |
Worked Orthopaedic Examples
| Research Question | Data Type | Design | Test |
|---|---|---|---|
| Union rate: dual-plating vs single-plating | Dichotomous | 2 independent groups | Chi-square / Fisher's exact |
| Knee Society Score at 6 months, two techniques | Continuous, check normality | 2 independent groups | Unpaired t-test or Mann-Whitney |
| VAS pain score pre-op vs post-op, same patients | Ordinal / often-skewed continuous | 2 related groups | Wilcoxon signed-rank |
| Functional outcome across 3 AO/OTA fracture grades | Ordinal outcome, >2 groups | >2 independent groups | Kruskal-Wallis |
| Blood loss vs surgical time | Continuous vs continuous | Correlation | Pearson's or Spearman's, depending on normality |
| Predictors of surgical site infection (age, diabetes, BMI, surgical time) | Dichotomous outcome, multiple predictors | Multivariable | Logistic regression |
| Implant survivorship over 3 years | Time-to-event | Comparing 2 implant types | Kaplan-Meier + Log-rank |
| Two surgeons grading the same X-rays on Tönnis classification | Ordinal, same subjects rated twice | Agreement | Weighted Cohen's kappa |
Common Errors to Avoid
- Running a t-test on a 5-point Likert/ordinal score just because it "looks numeric" — the single most frequent mistake in PG theses; ordinal scores need Mann-Whitney/Kruskal-Wallis unless the scale has strong evidence of interval properties.
- Treating repeated measurements on the same patient as if independent (e.g. left and right knee from one patient analysed as 2 separate observations) — inflates sample size artificially and violates independence.
- Skipping the normality check and defaulting to a parametric test out of habit.
- Not correcting for multiple comparisons when running several post-hoc pairwise tests after ANOVA/Kruskal-Wallis.
- Reporting p-values without effect sizes (mean difference, odds ratio, correlation coefficient) — a significant p-value with a trivial effect size is not clinically meaningful, especially with large sample sizes.