The Garden of Forking Paths: Why Multiple Comparisons Can Be a Problem, Even When There Is No "Fishing Expedition" or "P-Hacking" and the Research Hypothesis Was Posited Ahead of Time
⚡ Contested📌 Appears in:
Plain English Summary
Imagine you're walking through a garden where the path keeps branching, and at each fork you make a choice that feels obvious — but a different person might have chosen differently at every turn. That's the powerful metaphor this paper introduced to explain how researchers can unknowingly inflate their results. Even without deliberately cheating or fishing for significance, the sheer number of small analytical decisions (how to split groups, which outliers to drop, which measure to emphasize) means the one analysis a scientist reports isn't really just "one" test — it's the survivor of many invisible alternatives. Using Bem's precognition studies as a prime example, the authors argue that pre-registration — publicly committing to your analysis plan before seeing data — is the best antidote to this hidden flexibility.
Research Notes
Foundational replication-crisis paper that uses Bem (2011) as a central example. Introduced the influential 'garden of forking paths' metaphor, now a standard reference in debates about analytic flexibility in psi research and the Feeling the Future controversy.
Researcher degrees of freedom can produce a multiple comparisons problem even when scientists perform only a single analysis on their data. Using case studies from published psychology — including Bem's (2011) precognition experiments, menstrual-cycle effects on voting, and upper-body strength and political attitudes — a four-level typology of testing procedures is proposed, distinguishing deliberate fishing from the more common pattern where a single analysis path is chosen that appears predetermined but is actually contingent on the observed data. Pre-registration and pre-publication replication are recommended as solutions.
Related Papers
Cites
- False-Positive Psychology: Undisclosed Flexibility in Data Collection and Analysis Allows Presenting Anything as Significant — Simmons, Joseph P (2011)
- Why Psychologists Must Change the Way They Analyze Their Data: The Case of Psi — Wagenmakers, Eric-Jan (2011)
- Must Psychologists Change the Way They Analyze Their Data? — Bem, Daryl J (2011)
- Power failure: why small sample size undermines the reliability of neuroscience — Button, Katherine S (2013)
Companion
- Estimating the Reproducibility of Psychological Science — Open Science Collaboration (2015)
- A Bayes Factor Meta-Analysis of Bem's ESP Claim — Rouder, Jeffrey N (2011)
- Registered Reports: A Method to Increase the Credibility of Published Results — Nosek, Brian A (2014)
- An Agenda for Purely Confirmatory Research — Wagenmakers, Eric-Jan (2012)
- Why Most Research Findings About Psi Are False: The Replicability Crisis, the Psi Paradox and the Myth of Sisyphus — Rabeyron, Thomas (2020)
- Why Most Published Research Findings Are False — Ioannidis, John P.A (2005)
More in Skeptical
Cognitive Styles and Psi: Psi Researchers Are More Similar to Skeptics Than to Lay Believers
Searching for the Impossible: Parapsychology's Elusive Quest
False-Positive Effect in the Radin Double-Slit Experiment on Observer Consciousness as Determined with the Advanced Meta-Experimental Protocol
Cross-Examining the Case for Precognition: Comment on Mossbridge and Radin (2018)
N,N-Dimethyltryptamine and the Pineal Gland: Separating Fact from Myth
📋 Cite this paper
Gelman, Andrew, Loken, Eric (2013). The Garden of Forking Paths: Why Multiple Comparisons Can Be a Problem, Even When There Is No "Fishing Expedition" or "P-Hacking" and the Research Hypothesis Was Posited Ahead of Time. Columbia University Department of Statistics Working Paper.
@article{gelman_2013_forking_paths,
title = {The Garden of Forking Paths: Why Multiple Comparisons Can Be a Problem, Even When There Is No "Fishing Expedition" or "P-Hacking" and the Research Hypothesis Was Posited Ahead of Time},
author = {Gelman, Andrew and Loken, Eric},
year = {2013},
journal = {Columbia University Department of Statistics Working Paper},
}