How does the tool choose between Welch and Mann–Whitney?

For two independent groups, it checks each group's shape using a fixed rule: absolute skewness above 1 or a mean–median gap above half an SD flags skewness. The judgement is shown for you to confirm. The tool uses Welch's t-test for roughly normal data and Mann–Whitney for skewed data, and always runs the other method as a robustness check. A normality-test p value never decides the method on its own.

Why use Welch when the data are roughly normal?

Welch does not require the groups to have equal variance. When the shape rule flags skewness, the tool uses a rank-based method instead; for roughly normal measurements, Welch compares means without a separate variance test.

Research workbench

Auto Stats

Paste your data; the test is chosen from how the data are laid out

See how it chooses ↓

Paste data

Data table (first row is the header)Paste CSV or spreadsheet data, or upload a CSV / TSV file.

Upload CSV / TSV

Your data is parsed and analysed in your browser and is never sent to a server.

Illustrative samples

Simple comparison

Clinical datasets

What the tool sees

Paste a data table with at least two columns.

Result

Your analysis will appear here.

How it chooses the method

How a spreadsheet finds its own statistical test

All you do is paste your data. Every step after that follows a decision table written in advance, and the tool tells you why it chose what it chose.

First, read the table

First the tool works out what each column is: text labels are groups, numbers are measurements, and a code that repeats marks the same subject. Where it cannot tell, it asks you to pick instead of guessing.

One question only you can answer

One thing cannot be read from the numbers: whether they come from the same samples measured twice. Guess wrong and the whole result is wrong, so this is the one question the tool asks you directly.

Walk the decision table

Four things guide the method: outcome type, number of groups, pairing, and each group's distribution shape. A fixed rule checks skew and the mean–median gap, then shows the judgement for you to confirm.

Check shape, then choose

When the data are roughly normal, Welch compares the means. When they are skewed, the tool compares ranks instead. You can see the shape judgement for each group and confirm it before the analysis.

Check again with another method

Whichever method is selected, the tool always runs the other method as a robustness check. Agreement adds confidence; disagreement is a reason to look back at the data points, not to pick the smaller p value.

When many scores are zero

Immunohistochemistry often produces many zero scores. When zeros are common, the tool splits the question in two — how many samples stained at all, and how strongly the stained ones stained — and then combines them into one answer.

A result you can put in a paper

What you get at the end is a result you can put in a paper: the method, the sample sizes, the test statistic, degrees of freedom, the exact p value and the effect size — with a 95% confidence interval wherever a standard method for one exists.

Back to the top and paste your data ↑