1.Introduction
General-purpose statistical packages and spreadsheets are designed to compute whatever the user requests. They do not distinguish between a well-specified clinical analysis and a misapplied one: a spreadsheet will run a parametric test on non-normal data, or compute a p-value without an accompanying effect size, without objection. The responsibility for methodological correctness rests entirely with the user.
StatClinic is built around clinical and biomedical research workflows rather than generic calculation. Its structure reflects the sequence a careful analysis actually follows: define the research question, identify the variable types and study design, verify the assumptions of candidate tests, select an appropriate method, and report the result in a form suitable for a medical journal. The intent is not to replace statistical judgement, but to make the correct workflow the path of least resistance.
2.Philosophy of Evidence-Based Statistical Analysis
Evidence-based medicine depends on analyses that are appropriate to the data, transparent in their assumptions, and reported completely enough to be appraised by others. StatClinic is designed to align with the reporting and methodological expectations of clinical trials and observational studies, including the principles reflected in reporting guidelines such as CONSORT (trials), STROBE (observational studies), and PRISMA (systematic reviews).
In practice this means the platform favours:
- Complete reporting over isolated p-values — every applicable result includes an effect size and a 95% confidence interval alongside the test statistic and p-value.
- Assumption-aware test selection over one-size-fits-all defaults — the workflow routes users toward parametric or non-parametric methods based on the data.
- Interpretation in context over raw output — each result is accompanied by a plain-language reading and a reporting sentence consistent with medical-journal conventions.
3.Assumption Checking Before Test Selection
The most common statistical errors in the medical literature stem from applying a test whose assumptions are unmet — most often a parametric test on non-normal data, or a regression model with severe multicollinearity. StatClinic makes assumption checking an explicit, guided step rather than an afterthought.
- Normality. A dedicated normality-testing tool evaluates the distribution using an empirical-distribution-function test (Kolmogorov–Smirnov / Lilliefors), reports descriptive statistics, and — through its decision engine — recommends parametric or non-parametric tests accordingly. For small samples, the accompanying guidance references the Shapiro–Wilk test as the more powerful alternative available in full statistical packages.
- Homogeneity of variance. For two- and multi-group comparisons, the workflow flags the equality-of-variance assumption and directs users to Levene's test and the Welch correction where variances are unequal.
- Multicollinearity. Multiple linear regression reports the Variance Inflation Factor (VIF) for each predictor, with the conventional VIF > 10 threshold highlighted as indicating a problem.
- Linearity & homoscedasticity. Regression tools prompt inspection of residuals for constant variance and linear form before results are interpreted.
4.Appropriate Test Selection
StatClinic's Test Finder Wizard implements an explicit decision tree that mirrors standard biostatistical teaching. By asking about the research goal, variable types, number of groups, independence of observations, and distributional assumptions, it narrows the field to a defensible method — which discourages the practice of trying several tests and reporting the most favourable (a form of p-hacking).
Because the path to each recommendation is fixed and visible, two researchers with the same study design arrive at the same method — a property general-purpose menus do not enforce.
5.Publication-Ready Outputs
Results are structured to match the reporting conventions of medical journals rather than the raw layout of a statistics engine. For each applicable analysis, StatClinic reports:
- The test statistic and degrees of freedom (e.g. t(38) = −4.07).
- The exact p-value (with p < 0.001 used as the conventional lower bound, never p = 0.000).
- An effect size appropriate to the test (Cohen's d, odds ratio, R², Cramér's V, etc.).
- A 95% confidence interval for the estimate.
- A reporting sentence in APA-consistent style, ready to adapt for a Results section.
Citations for the software itself are provided in Vancouver, APA 7th, BibTeX, and RIS formats on the Cite StatClinic page, so the analytic tool and version can be recorded in a Methods section.
6.Clinical Research Workflow
StatClinic's method library is organised around the analyses that recur in clinical and biomedical research, rather than a generic list of procedures:
| Research task | Supported methods |
|---|---|
| Diagnostic accuracy | ROC curve & AUC, DeLong comparison, Youden index, sensitivity, specificity, PPV/NPV, decision curve analysis |
| Survival / time-to-event | Kaplan–Meier estimation, hazard ratios, Cox proportional-hazards regression |
| Risk & association | Odds ratio, relative risk, number needed to treat, logistic regression |
| Systematic review | Meta-analysis with forest plots, heterogeneity (I², τ²), funnel plots, meta-regression |
| Agreement & reliability | Cohen's κ, intraclass correlation, Cronbach's α, Bland–Altman analysis |
| Group comparison | t-tests, ANOVA family, non-parametric equivalents, post-hoc procedures |
7.Reproducibility & Transparency
Reproducibility means that a stated analysis, applied to the same data, yields the same result — and that a reader can determine exactly what was done. StatClinic supports this in the following concrete ways:
- Deterministic computation. Every calculation runs locally in the browser using standard published formulas. The same input always produces the same output; there is no hidden server-side state, and no result depends on an unseen random seed except in explicitly stochastic procedures.
- Validation against reference software. Methods are checked against results from established statistical software (such as SPSS and R) to confirm numerical agreement, so that a StatClinic result matches what a reviewer would obtain in a reference package.
- Documented methods and assumptions. Each tool states the statistic computed, the assumptions it relies on, and the reporting standard it follows, so the analysis can be described precisely in a manuscript.
- Versioned software identity. The exact software version is recorded and citable (see the Changelog and Cite pages), allowing the analytic environment to be pinned in a Methods section.
8.Academic Documentation & Citation Support
Every supported method is accompanied by documentation describing when it applies, the assumptions it requires, the quantities it reports, and worked clinical examples — written to be usable directly when drafting a Methods section. The Documentation page consolidates the available methods, a getting-started workflow, frequently asked questions, and the version history.
For the software itself, formal citations are provided in Vancouver, APA 7th, BibTeX, and RIS formats, with fields reserved for a DOI and, in future, a peer-reviewed software or validation paper — so that citations remain stable as the platform's academic record grows. Researchers can add the software citation to their reference list directly from the Cite StatClinic page.