METHODOLOGY
how the numbers are produced.
Researchers come to us from statistical software they already trust, so the question that matters is not what the platform can do — it is how it arrives at a number, and whether that number can be defended. This page answers both.
The engine computes. The AI explains.
Every statistic on this platform is calculated in an audited Python engine built on pandas, scipy and statsmodels, covered by a regression test suite that cross-checks its output against those libraries and against canonical datasets.
The language model never computes a value. It reads your study objective, helps phrase the research question, explains what a finding means, and drafts the report narrative around results the engine has already produced. That separation is deliberate: a model that writes prose cannot invent a p-value.
What happens to your data
01
Describe the study
You state your research question and objectives in plain language. The AI layer turns that into a numbered set of objectives and proposes a suitable family of tests, which you confirm or correct before anything runs.
02
Check and clean the data
A deterministic engine runs an audited suite of data checks and reports what it finds with the real affected rows, columns and values. Each finding shows the exact before and after. Nothing is changed until you approve it, and you can always keep the original.
03
Inspect the cleaned dataset
The cleaned table is shown read-only, with the row count and the changed cells, plus a cleaning log and — where questions were renamed — a question codebook. Analysis stays locked until you confirm the data is right.
04
Decide how missing values are handled
Missingness is shown explicitly. You either approve a stated imputation or keep the values missing. Nothing is filled in or dropped silently.
05
Run descriptive and inferential statistics
The engine computes descriptive statistics, then the tests appropriate to your design, checking the assumptions each test requires before it is fitted. Effect sizes and confidence intervals accompany the p-values.
06
Produce the report
Results are assembled into APA 7th edition tables and figures. The AI layer drafts the narrative around the numbers the engine has already computed; it does not produce the numbers.
What the platform will not do
- It will not change your data without your approval, and it keeps the original upload whatever you decide.
- It will not guess which of your columns are paired. That is your knowledge, so paired tests are chosen by you.
- It will not fill in or delete missing values silently. You either approve a stated approach or keep the values missing.
- It will not replace your judgement. It documents the decisions so you can defend them — the analysis remains yours to sign.
Common questions
- Does an AI calculate the statistics?
- No. Every statistic is computed in an audited Python engine using pandas, scipy and statsmodels, with a regression test suite that cross-checks results against those libraries and against canonical datasets. The language model reads your objective, explains findings and drafts report prose around results the engine has already produced. It cannot produce or alter a number.
- Which statistical tests are available?
- Independent and paired t-tests, Mann-Whitney U, Wilcoxon signed-rank, one-way ANOVA, Kruskal-Wallis, chi-square, Fisher exact, Pearson and Spearman correlation, linear and logistic regression, Kaplan-Meier survival analysis, post-hoc comparisons, ROC and diagnostic accuracy, reliability analysis, and multiple-comparison corrections including Bonferroni, Holm, Sidak and Holm-Sidak.
- Can the platform change my data without telling me?
- No. Every cleaning change is optional and is shown with the exact rows, columns and before and after values it would affect. You approve or decline each one, and you must inspect the cleaned table before analysis can begin. The original upload is always preserved.
- Are the results reproducible?
- Yes. Cleaning runs as a recorded, replayable recipe, and the same input with the same accepted decisions produces the same cleaned dataset. Every run keeps a cleaning log, and transformations are designed to agree with the equivalent steps in standard spreadsheet and statistical software.
- Which paired tests are chosen automatically?
- None. Which columns are paired is knowledge only the researcher has, so paired tests are always selected manually rather than inferred from the data.
- How should I cite Analysez AI in a paper?
- A citation format is provided on the How to cite page, and each completed analysis can produce a certificate recording what was run. Analysez AI is a tool used in the analysis; the researcher remains the author of the work.