Flexibility in analysis inflates false positives
P-hacking (or researcher degrees of freedom) occurs when you test many hypotheses, exclude participants after seeing results, adjust thresholds, or try multiple statistical approaches on the same dataset. If you run 20 statistical tests on random data, one will appear significant at p < 0.05 by chance alone. Most p-hacking is unconscious: researchers genuinely believe they're pursuing legitimate analyses but didn't pre-commit to stopping rules.
The diagram shows 20 p-values from the same dataset with no real effect. Two surpass 0.05, yet both are false discoveries. Real experiments contain exactly one true hypothesis and one planned test.
Pre-registration locks the analysis plan
Pre-registration means writing down your primary hypothesis, sample size, statistical approach, and stopping rules before data collection begins. This document is time-stamped on a public registry, making post-hoc deviation detectable. Researchers can still explore data, but those exploratory findings are labeled as such and require separate confirmation in new data. Pre-registration does not prevent new discoveries, only prevents false discoveries dressed as confirmatory findings.