Skip to content

Is my treatment different from my control?

Compare two groups

Choosing and running a two-group comparison, and reporting it usefully.

The default is an unpaired t-test, and Welch’s version of it should be your standard rather than the equal-variance one — it costs almost nothing when variances match and saves you when they do not. Use a paired test only when each measurement genuinely has a partner: the same dish before and after, the same animal on both days. If your data are clearly not normal and n is small, Mann–Whitney is the usual fallback, though with very small n it has little power to detect anything.

The larger point is what you report. A p-value tells you almost nothing on its own. Give the difference between means and its confidence interval, and show the individual points. An estimation plot does all three at once.

If you measured several dishes from each of three independent experiments, your n is three, not the number of dishes. Pooling technical replicates as though they were independent is the single most common way biology papers report significance that is not there.

What to use

If you script

Command-line and library options. No page here — they need an environment rather than a browser.

Drafted from the literature and not yet reviewed by a specialist. Treat it as a starting point, and tell us if something here is wrong. Last touched .

All tasks →