3 ms·
A 2016 poll of 1,500 scientists reported that 70% of them had failed to reproduce at least one other scientist's experiment (50% had failed to reproduce one of
by lazyjones 6y ago
A 2016 poll of 1,500 scientists reported that 70% of them had failed to reproduce at least one other scientist's experiment (50% had failed to reproduce one of their own experiments).
- RhysU 6y agoI would love to see this number for computational scientists.
- aewens 6y agoWhile anecdotal, its easy to run into issues like this[0] where scientist with little CS background assume certain behaviors in code that don't actually exist. [0] https://www.vice.com/en_us/article/zmjwda/a-code-glitch-may-have-caused-errors-in-more-than-100-published-studies https://www.vice.com/en_us/article/zmjwda/a-code-glitch-may-...
- Fragoel2 6y agoI don't have a clear reference to point you to but in many CS research fields these number are equally bad. Often code is not even published and is a complete mess when it is.
- fiftyacorn 6y agoReminds me of doing chemistry labs at uni. You follow the experiment plan and get nothing like the expected result so you add big error bars and your result matches. i remember getting a clear linear relationship once that as supposed to be a curve. Solution was error bars and a curve to match expectation. Pointless