Two distinguishing traits of science are that there are errors all the time and that almost no-one can reproduce anything

I got annoyed and tweeted:

”If you can’t reproduce a result, it isn’t science” … so we’re at that stage now, when we write things that sound righteous but are nonsense.

Hashtag subtweet, I guess. But it doesn’t matter who first wrote the sentence I was complaining about; they won’t care what I think, and I’m not out to debate them. I only think the quoted sentence makes sense if you take ”science” to mean ”the truth”. The relationship between science and reproducibility is messier than that.

The first clause could mean a few different things:

You have previously produced a result, but now, you can’t reproduce it when you try …

Then you might have done something wrong the first time, or the second time. This is an everyday occurrence of any type of research, that probably happens to every postdoc every week. Not even purely theoretical results are safe. If the simulation is stochastic, one might have been interpreting noise. If there is an analytical result, one might have made an odd number of sign errors. In fact, it is a distinguishing trait of science that when we try to learn new things, there are errors all the time.

If that previous result is something that has been published, circulated to peers, and interpreted as if it was a useful finding, then that is unfortunate. The hypothetical you should probably make some effort to figure out why, and communicate that to peers. But it seems like a bad idea to suggest that because there was an error, you’re not doing science.

You personally can’t reproduce a results because you don’t have the expertise or resources …

Science takes a lot of skill and a lot of specialised technical stuff. I probably can’t reproduce even a simple organic chemistry experiment. In fact, it is a distinguishing trait of science that almost no-one can reproduce any of it, because it takes both expertise and special equipment.

No-one can ever reproduce a certain result even in principle …

It might still be science. The 1918 influenza epidemic will by the nature of time never happen again. Still, there is science about it.

You can’t reproduce someone else’s results when you try with a reasonably similar setup …

Of course, this is what the original authors of the sentence meant. When this turns out to be a common occurrence, as people systematically try to reproduce findings, there is clearly something wrong with the research methods scientists use: The original report may be the outcome of a meandering process of researcher degrees of freedom that produced a striking result that is unlikely to happen when the procedure is repeated, even with high fidelity. However, I would say that we’re dealing with bad science, rather than non-science. Reproducibility is not a demarcation criterion.

(Note: Some people reserve ”reproducibility” for the computational reproducibility of re-running someone’s analysis code and getting the same results. This was not the case with the sentence quoted above.)