HARKing (Hypothesising After Results are Known)

HARKing — Hypothesizing After the Results are Known, a term introduced by Norbert Kerr in 1998 — is the practice of presenting a hypothesis discovered through exploring the data as though it were the a priori hypothesis that motivated the study’s original design. The paper’s Introduction and Methods are written to imply a confirmatory test that never actually happened; the real sequence (explore data, find a pattern, then write a hypothesis to match it) is hidden from the reader.

What it is not

HARKing is structurally distinct from p-hacking: p-hacking changes what analysis is run — trying multiple tests, exclusion criteria, or subgroups until one yields significance — while HARKing changes what the paper claims was hypothesized, without necessarily altering the analysis performed at all. A study can HARK without any p-hacking, and vice versa, though the two often occur together. HARKing is also not the same as legitimate exploratory research: reporting an unexpected finding and clearly labeling it as exploratory or hypothesis-generating, rather than presenting it as confirmation of a stated prior hypothesis, is standard and valuable scientific practice. The problem is not exploration itself but disguising it as confirmation.

Why it matters

Because a confirmatory test’s evidential value depends on the hypothesis having been specified before the data were seen, HARKing removes the falsifiability that gives a “confirmed” result its meaning, while leaving the analysis and data themselves untouched — which is exactly why it is harder to detect after the fact than an outright data manipulation would be. Presenting one measure, chosen after inspecting several candidate outcomes, as though it had always been the pre-specified primary endpoint is a common pattern.

In practice

Pre-registration (time-stamping the hypothesis, design, and primary outcome before data collection, e.g. via OSF or a trial registry) and the registered-report publication format, in which peer review of the hypothesis and design happens before results exist, are the principal defenses journals and funders now rely on. Explicitly labeling analyses as exploratory versus confirmatory in Methods and Results is the minimum transparency step available to any author, registered or not.

Related terms

See also p-hacking, pre-registration, questionable research practices, and reproducible research practices.

References

• Kerr, ‘HARKing: Hypothesizing after the results are known’ (Personality and Social Psychology Review, 1998); Rubin, ‘The costs of HARKing’ (British Journal for the Philosophy of Science, 2022).