5 reactions
Physicists Really Ought to Share More Raw Data and Scripts
The recent superconductor papers make it seem pretty obvious to me that the field needs greater transparency when it comes to publishing data. And it's probably not just condensed matter physics that can improve, it's probably most subfields. If you publish results based on some data, you should really do your best to publish the raw data and scripts that produced your results. For the recent superconductivity papers it really shouldn't be hard to share all of this data and analysis from start to finish.
I got my PhD in physics and now I'm in computational biology. Biology has got its own problems, but the fact that there are methods researchers in biology who specialize in data analysis means that there is a large push to get people to share data and methods online. Many journals now require a "data availability" statement about where/how the data will be available online, and genetics datasets are huge so most physicists don't have that excuse.
Here's a list of data repositories in physics and other fields where you can share your data: https://www.nature.com/sdata/policies/repositories
There's also Github for smaller datasets.
The goal of the methods section is to enable people reading your paper to reproduce your experiment or analysis. If they can't do that then you haven't done your job.
Sure, there are exceptions. Some datasets are going to be too large to reasonably share, but surely you could still share scripts and some processed data. I don't need to see every pixel imaged by your cameras, just the ones that are relevant to your results. Each subfield needs to come up with its own norms.
In recent years I have just created a Github page for each paper, post the data or links to the data, post the scripts, and post a readme that outlines how to generate the results of the data with the scripts. It's not that hard.
Someone on the other thread said something like, "Oh, I have gigabytes of data and an incomprehensible lab notebook, good luck sorting through all of that if I share it." But I feel like this is exactly the problem that data sharing is designed to prevent, and it's also basically scientific misconduct to not keep reasonable lab notebooks of your data.
When I was a physicist I did a bad job of this myself. I never shared data or scripts online. My PI would never have allowed it. I asked other research groups for the data they used to generate published images and they said no. Today, as an outsider, I think it's easier for me to point out the problems that existed back then, and it's possible for me to see where different fields can learn from each other now that I've been a part of both of them for so many years.
I really don't have any skin in the game anymore, except for the fact that I got into science because I thought it was more important than an exercise in story-telling and picture-making.
The recent superconductor papers make it seem pretty obvious to me that the field needs greater transparency when it comes to publishing data. And it's probably not just condensed matter physics that can improve, it's probably most subfields. If you publish results based on some data, you should really do your best to publish the raw data and scripts that produced your results. For the recent superconductivity papers it really shouldn't be hard to share all of this data and analysis from start to finish.
I got my PhD in physics and now I'm in computational biology. Biology has got its own problems, but the fact that there are methods researchers in biology who specialize in data analysis means that there is a large push to get people to share data and methods online. Many journals now require a "data availability" statement about where/how the data will be available online, and genetics datasets are huge so most physicists don't have that excuse.
Here's a list of data repositories in physics and other fields where you can share your data: https://www.nature.com/sdata/policies/repositories
There's also Github for smaller datasets.
The goal of the methods section is to enable people reading your paper to reproduce your experiment or analysis. If they can't do that then you haven't done your job.
Sure, there are exceptions. Some datasets are going to be too large to reasonably share, but surely you could still share scripts and some processed data. I don't need to see every pixel imaged by your cameras, just the ones that are relevant to your results. Each subfield needs to come up with its own norms.
In recent years I have just created a Github page for each paper, post the data or links to the data, post the scripts, and post a readme that outlines how to generate the results of the data with the scripts. It's not that hard.
Someone on the other thread said something like, "Oh, I have gigabytes of data and an incomprehensible lab notebook, good luck sorting through all of that if I share it." But I feel like this is exactly the problem that data sharing is designed to prevent, and it's also basically scientific misconduct to not keep reasonable lab notebooks of your data.
When I was a physicist I did a bad job of this myself. I never shared data or scripts online. My PI would never have allowed it. I asked other research groups for the data they used to generate published images and they said no. Today, as an outsider, I think it's easier for me to point out the problems that existed back then, and it's possible for me to see where different fields can learn from each other now that I've been a part of both of them for so many years.
I really don't have any skin in the game anymore, except for the fact that I got into science because I thought it was more important than an exercise in story-telling and picture-making.
| 32,000 mlx | |
| 37 | |
| 5 |