What we do · Data analysis

Data analysis

Transforming complex, real-world data into clear, defensible evidence that leads to confident decisions.

Overview

Data analytics in the real world

Every important decision begins with data, yet turning data into reliable information remains one of the biggest challenges, whether in research or business. Real-world datasets are rarely complete, consistent, or analysis-ready. Missing values, outliers, and poor data quality can lead to misleading conclusions if they are not addressed correctly.

Yet data preparation is often rushed or overlooked. Even once the data have been cleaned, another challenge emerges: choosing the right statistical analysis. Modern statistical software offers hundreds of tests and modelling techniques, but deciding where to begin, whether assumptions are met, and how to interpret the results requires specialist expertise that many users simply do not have. As a result, valuable data are often underused, inappropriate methods are applied, or analyses become a “black box” that decision-makers struggle to trust.

The real challenge is not collecting more data—organisations already have plenty.

The challenge is transforming complex, messy data into clear, defensible evidence that people can understand and act upon with confidence.

Our solution: Inference-Stats

Inference-Stats was built to remove the uncertainty from statistical analysis. It offers a full suite of statistical tests. If you are a seasoned statistician, this large menu of statistical tests will feel intuitive. If you are a less experienced analyst, Inference-Stats provides a structured, guided workflow that takes you from raw data to reliable conclusions.

Inference-Stats helps prepare and clean your data, identifies potential quality issues, supports multiple methods for handling missing data, and leads you through exploratory data analysis before recommending appropriate statistical methods. At every stage, Inference-Stats checks whether the required assumptions are satisfied and guides you towards methods appropriate for your data—whether your work involves regression, clustering, survival analysis, time series modelling, ROC analysis, or machine learning.

An integrated AI assistant explains both the statistical methods and the results in clear, accessible language. Inference-Stats can generate a comprehensive report documenting the data, analyses performed, findings, and interpretations.

“Statistical software helps you run analyses. Inference-Stats helps you answer questions.”

Transforming data into evidence

With Inference-Stats, we do more than statistical analyses—we transform your data into evidence that informs better decisions.

Now available

Inference-Stats is live

Upload a spreadsheet and it will check your data, recommend the appropriate test, run it, and explain the result — with the effect size and confidence interval alongside the p-value. Free to start, and the analysis runs in your browser.

Open Inference-Stats

Or read the method guides first.

Method guides

Choosing the right test

The hardest part of an analysis is rarely the arithmetic. It is the decision made before it: which test, which effect size, what to do about the gaps. Get that wrong and the software still returns a number — a confident, precise, wrong number.

Each guide below came from a question we kept meeting in real analyses: a skewed outcome that makes a mean meaningless, two groups with wildly different variances, a p-value doing work an effect size should be doing. Each explains the decision in plain terms, shows what goes wrong when it is made badly, and says how to report the result. They are published on Inference-Stats, where you can also run the analysis being described.

Comparing groups

Kruskal-Wallis vs ANOVA: which test should you use? When skew, ordinal outcomes or outliers make ANOVA the wrong answer — and the third option most people miss. Welch’s t-test vs Student’s t-test With unequal variances and unequal group sizes, Student’s t-test returns a false positive in roughly one study in three. Welch’s holds 5%.

Reporting results honestly

P-value vs effect size: what to report Three studies can share an identical p-value and support completely different conclusions. Why the effect size and its interval carry the story. Handling missing data Why deleting incomplete rows is a decision, not a default — and when multiple imputation genuinely recovers the truth.

Finding structure & time-to-event

Which clustering algorithm should you use? K-means, DBSCAN or GMM — how the shape of your data decides, and how to choose k without fooling yourself. Competing risks in survival analysis When a patient can die of something else first, 1−Kaplan-Meier overstates your event rate. What to use instead. Which statistical test should I use? The decision guide the others sit under — from the shape of your question and your data to the test that answers it.
Back to all capabilities