4

I was wondering how people deal with data represented with plots (of any kind) in a published paper, in particular for what regards Physics. When doing theoretical research (but not only), often you have to compare your model and its predictions with other models and/or experimental data. In fact, I usually find many papers in which the authors "update" some existing results with their own findings, changing very complicated plots and superimposing new graphics (i.e. the constraints they have found on a given quantity etc.).

How do people usually do that?

I mean, I do not really think that people (especially theorists) get all the possible databases and re-analyze everything, it would be simply impossible when one has to deal with dozens of experiments and the very messy data analysis behind each of those. I was thinking that maybe researchers just digitize published graphs to have several lists of coordinates which can be used at will to create new plots.

What do you think about this and what is your experience?

Thank you very much!

  • In the dark ages, one used a copier to enlarge the picture, then would either (1) hand measure where data points were, or (2) got a hold of one of the fancy precision input devices for CAD, or (3) used an HP pen plotter with a special optical target thingy in place of the pen (you could manually move and record coordinates on some HP plotters). – Jon Custer Nov 1 '19 at 19:38
  • automeris.io/WebPlotDigitizer – astronat Nov 2 '19 at 16:23
  • I already know and use WebPlotDigitizer, but my question was more about the way researchers use to do... – Lele Nov 2 '19 at 18:45
2

I know of four approaches, and to my knowledge, all are commonly used. I'm coming from theoretical chemistry, so not quite theoretical physics, but not so different.

  1. Often the data is made public, especially from large collaborations. For example: http://opendata.atlas.cern/ Otherwise, ask the original authors for their data. Often they'll share.
  2. Recalculate the results -- especially the case with theoretical/computational approaches. Within a given field, many methods are usually quite similar to each other, so code written to do one can be easily modified to handle another. It just costs you some computer time.
  3. Have a standard dataset that you compare to. If you're developing new methods, you test them against a standard that has been studied deeply. I've especially seen this in computational work, where a single massive brute-force simulation is used for years as a benchmark, and that data is shared with the community.
  4. Graph digitizers, e.g., http://markummitchell.github.io/engauge-digitizer/. Scan and estimate where the data is. (Actually, this is the only approach I personally have not really used.)
  • Ok, I don't know how stuff works in your field, but I am quite sure that in Physics the experiments are so complicated that it is nearly impossible for someone not involved to analyze properly the data (it takes months of work and large teams of people!), not to mention that you have to compare them... for a theoretical physicist is not something affordable (it would mean learning a new job in practice...). They could provide you with the final list of data (the one that reproduces the plot), but in that case, it would be exactly as digitizing the plot and deriving the list by yourself! – Lele Nov 2 '19 at 8:51
  • I've updated my answer a bit (downloading data; explaining what I meant by standard datasets. I'm not sure what kinds of experiments you're thinking of, but they don't get much more complicated than ATLAS. – dwhswenson Nov 2 '19 at 9:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.