We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

econ.TH

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Economics > Theoretical Economics

Title: Screening $p$-Hackers: Dissemination Noise as Bait

Abstract: We show that adding noise to data before making data public is effective at screening $p$-hacked findings: spurious explanations of the outcome variable produced by attempting multiple econometric specifications. Noise creates "baits" that attract $p$-hackers, who engage in data mining with no prior information about the true cause behind the outcome, inducing them to report verifiably wrong results. But, noise only minimally impacts honest researchers who use data to test an ex-ante hypothesis about the causal mechanism. We characterize the optimal level of dissemination noise and highlight the relevant tradeoffs in a simple theoretical model. Dissemination noise is a tool that statistical agencies (e.g., the US Census Bureau) currently use to protect privacy, and we show this existing practice can be repurposed to improve research credibility.
Subjects: Theoretical Economics (econ.TH)
Cite as: arXiv:2103.09164 [econ.TH]
  (or arXiv:2103.09164v2 [econ.TH] for this version)

Submission history

From: Kevin He [view email]
[v1] Tue, 16 Mar 2021 16:05:19 GMT (70kb,D)
[v2] Fri, 11 Jun 2021 20:24:54 GMT (84kb,D)

Link back to: arXiv, form interface, contact.