Statistical Analysis you can defend
From choosing the right test to interpreting the output, we make sure your statistics are correct, appropriate to your design, and explainable in your viva and to reviewers — analysed on your real data, never fabricated.
Statistics are where many otherwise-strong studies come undone. The wrong test for the design, an assumption left unchecked, a misread p-value, or a table that reviewers cannot follow — any of these can undermine months of careful data collection and invite a rejection that has nothing to do with the quality of your idea. Our role is to close that gap: to help medical, dental, nursing, and allied-health researchers analyse their data correctly and, just as importantly, understand what the numbers actually mean so they can own and defend every result.
We work as a consultant beside you, not a black box you hand data to. You keep authorship and every scientific decision. We bring the methodological judgement — matching the analysis to your question, checking that the data support it, and translating the output into plain language you can put into your results and discussion.
What we help with
Statistical support spans the full arc of a quantitative study, from first look at the data to the final table in your manuscript:
- Data cleaning and preparation — screening for entry errors, impossible values, and missing data, and deciding how to handle them defensibly
- Descriptive statistics — summarising your sample correctly for continuous, ordinal, and categorical variables
- Test selection — matching the analysis to your design, data type, and distribution rather than defaulting to a familiar test
- Group comparisons — t-tests, one-way and repeated-measures ANOVA, ANCOVA, Mann–Whitney, Wilcoxon, Kruskal–Wallis, chi-square, and Fisher’s exact
- Correlation and regression — Pearson and Spearman, and simple, multiple, and logistic regression with confounder adjustment
- Diagnostic accuracy — sensitivity, specificity, predictive values, likelihood ratios, and ROC/AUC analysis
- Survival analysis — Kaplan–Meier curves, log-rank tests, and Cox proportional-hazards models
- Reliability and agreement — Cronbach’s alpha, Cohen’s and weighted kappa, and intraclass correlation coefficients
Getting the assumptions right
A test is only valid if its assumptions hold, and this is where many analyses quietly go wrong. Before we report any comparison we check the conditions the test depends on — normality of residuals, homogeneity of variance, independence of observations, linearity, and, for regression, multicollinearity and influential points. When assumptions are not met we do not ignore it; we either transform the data appropriately, switch to a non-parametric or robust alternative, or model the structure explicitly. Every one of these choices is documented so you can explain why a particular test was used if an examiner or reviewer asks.
Software we work in
We use SPSS, R, and GraphPad Prism, with Excel for data handling. The choice depends on your study, your supervisor’s and journal’s expectations, and what you are comfortable defending. SPSS remains the standard in most medical and health-science departments; R is preferred for reproducible, script-based workflows and advanced modelling; GraphPad is common in laboratory and bench research. We can also translate an analysis from one platform to another if your journal or supervisor prefers it.
Common mistakes we help you avoid
- Using a t-test or ANOVA on clearly skewed data without checking normality
- Running many pairwise tests without correcting for multiple comparisons
- Treating ordinal scores as if they were continuous without justification
- Confusing statistical significance with clinical importance, or reporting a bare p-value with no effect size or confidence interval
- Ignoring clustering or repeated measures and analysing correlated data as independent
- Over-fitting a regression model with too many predictors for the sample size
Reporting to journal standard
How results are reported matters as much as how they are computed. We present findings following recognised reporting guidelines — CONSORT for trials, STROBE for observational studies, STARD for diagnostic accuracy — and your target journal’s specific format. That means exact statistics: effect sizes with 95% confidence intervals, precise p-values rather than “p<0.05”, and clearly labelled tables and figures that a reader can interpret without wading back through the text.
How the process works
You share your dataset and research questions. We confirm the study design, check the data and assumptions, run the appropriate analyses, and return annotated output with publication-ready tables, figures, and a written interpretation of each key result. We then walk you through the findings so you can present and defend them in your own words. If reviewers later request additional or sensitivity analyses, we help you run and explain those too.
Whether you are at the protocol stage and want your analysis plan reviewed, sitting on a finished dataset you are unsure how to analyse, or facing statistical queries from reviewers, a short consultation is usually enough for us to tell you exactly what your study needs.