Interactive Exploration

The Garden of
Forking Paths

Same dataset. Different analysis choices. How many paths lead to a "significant" result?

Your dataset

You're analysing data from a study on whether 6 weeks of mindfulness meditation increases hippocampal grey matter volume. You have MRI scans from 48 participants (24 meditation, 24 control). The raw data are ambiguous — there's a hint of an effect, but it's noisy. Now you need to make five analysis decisions before you can report a p-value. Each choice is defensible. None is "wrong." But they lead to very different conclusions.

Your analysis path produced

0
Paths → p < .05
0
Paths → p ≥ .05
0%
Significant rate

All 32 analysis paths

This is the Garden of Forking Paths. The term comes from statistician Andrew Gelman. Even without any intention to cheat, the flexibility in data analysis means that many defensible pipelines exist for the same dataset. If a researcher (consciously or not) tries a few different approaches and reports the one that "works," the published p-value no longer means what it claims to. This is why pre-registration matters — it locks in analysis decisions before seeing the data, closing off the forking paths.
Dr Marcella Montagnese
Junior Research Fellow, Christ's College · University of Cambridge
Website · Google Scholar · GitHub · Source