1

This question applies to both data set I generated from my own research and public data set I reproduce in teaching.

Suppose in a data intensive course, I used a publicly available data set as an example (with proper citation), and, for purely non-scientific reasons (e.g. to avoid a politically sensitive issue), I removed a variable (among many variables) from the data set. The modified copy is stored in the course management system with no explanation given as to why a variable was removed. The removal of the variable is obvious --- anyone who actually looked at the data would notice.

Strictly speaking, a data set is not faithfully reproduced. Of course, if I simply select a handful of variables that are interesting and discarded the rest, and I explained why I do that, that would certainly not be a misconduct. But I am now less certain if all but one variable are reproduced and no explanation is given. (Of course, no good explanation exist)

In this context, is the removal of a variable from public data set a misconduct?

4
  • 4
    Couldn’t you attach a notice that says “this data set is adapted from [public data set X] for pedagogical purposes” or something along those lines?
    – Dan Romik
    Dec 28, 2021 at 15:48
  • 3
    Did you comply with the terms of the license (if there is one) under which the dataset is made public?
    – Dan Romik
    Dec 28, 2021 at 15:48
  • 1
    It really depends on the content of the dataset and how it might be used. There's a big difference between deleting the "diagnosed with cancer" variable and deleting the "liked on Twitter" variable. Not all data is important enough to merit ethical scrutiny. Dec 28, 2021 at 17:28
  • Because you don't say what the variable removed was, nor the reasons, I am left to imagine. This is a situation that reflects poorly on you becaue when I hear "political reason" my imagination runs towards non-generous regions.
    – Dan
    Dec 28, 2021 at 23:45

3 Answers 3

3

I think it would only be misconduct if you tried to publish the revised data without comment and without permission. And, citation is needed, of course.

But, for purposes of teaching it is good to let the students know the original source, but sufficient to label the revision as "extracted from ..." or "revised from ...", or similar.

Some data sets can actually be artificially constructed so as to make certain insights possible and easier to gain than they would be from a more "messy" version. It depends on the goals. At some point students need to be able to deal with the world as it is, but not necessarily from the very start.

So, point to the original, and label the revision as such. Don't publish the revision outside the student's view without some form of permission.

But even in research, it is allowed to ignore certain variables if they aren't germaine to the current needs.

1

If you changed the dataset in any way, you should at the very least mention that you did so. And you should make sure that you are complying with the terms of any legal license under which the dataset is made public.

I don’t think you have an ethical duty to explain why you removed the variable, assuming it’s for a good faith, ethical reason. (E.g., an unethical reason would be if your intention is to mislead people into an inaccurate interpretation of what the data means.) Wanting to avoid a politically charged subject doesn’t strike me as an unethical reason, at least for the kinds of politically charged subjects I am able to imagine.

Is it misconduct if you didn’t state that you changed the dataset, and/or didn’t comply with the legal license? I don’t think so, not unless you acted out of malice or an unethical motivation, which it sounds like you didn’t. It wouldn’t be an ideal practice and would mean you made a small mistake, but not any small mistake that anyone makes is misconduct. Correct the mistake if you can, and move on.

0

No. It would be inaccurate to describe it as the same data set, either explicitly or implicitly.

I have taken the same putative data set from two different archival sources and found that the data sets were different. The difference was that a small number of variables were recoded as missing in one of the two data sets. There was no documentation of the difference except in the data itself. There was no explanation of why this was done.

I thought it would be more responsible to give some description of changes that were made by one of the archives.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .