Donald Kornfeld and Sandra Titus argue that misconduct should be considered when investigating irreproducible research (Nature 537, 29–30; 2016). In my view, this premise of 'guilty until proved innocent' risks turning a scholarly investigation into a witch hunt.

Distinguishing poorly designed research from faked data is hard, but it is generally more difficult to prove misconduct than to identify the cause of irreproducibility. Moreover, investigating bad science costs less than examining misconduct (in terms of money, time, careers and so on). One estimate put the direct cost of a misconduct enquiry at US$525,000 (A. M. Michalek et al. PLoS Med. 7, e1000318; 2010).

The authors assert that the US National Institutes of Health's (NIH) training mandate for responsible conduct of research failed to reduce misconduct. Perusing research on the rate of misconduct, I find estimates that span several orders of magnitude, so it is unclear whether misconduct is or has been rising or falling. Thus, we cannot say what effect the NIH training scheme has had.

The damage caused to the scientific record by publishing sloppy, plagiarized or fabricated research is ultimately the same.