I have reached out to a researcher about using their dataset from a paper published 20+ years ago, but they have not responded to my email. There are a variety of reasons why they might have not responded, ranging from being retired, data is destroyed, not trusting unsolicited emails, coveting the data's future research potential, and probably a myriad of other possibilities.

Because their study

  • used a constrained scale (Likert [1,5]),
  • used a relatively small sample size,
  • used a small number of questions,
  • reported multiple statistics giving location, scale, and correlations,
  • and model fit statistics,

it is possible for me to recreate a dataset that is identical to the original.

The purpose of creating this dataset would be to provide a hypothetical analysis with updated modeling and statistical methods to compare if the conclusions would be similar with new methods. The original dataset would have been preferred, but a reconstruction would provide an equivalent result for the newer methods.

The measurements are survey responses, so on the face of it one might be concerned about privacy. However, the sampled population is large and the survey questions are not specific, so actually identifying individuals would not be possible from this reconstruction.

Assuming that I am transparent with what I am doing in my reporting of the data and analysis, is it ethical to reconstruct a dataset without explicitly obtaining permission?

  • 13
    What makes you even concerned that it could be unethical? That's what's routinely done in replication studies. One way that science progresses it by testing existing hypotheses and models on new, yet-unseen data from reality. Commented Feb 21, 2022 at 10:35
  • 10
    (To be clear, what you have in mind here is technically not "reconstructing dataset X" but "constructing a new dataset, based on the methodical setup of X".) Commented Feb 21, 2022 at 10:37
  • 16
    Do you actually mean "identical" or "identical in summary statistics"? Commented Feb 21, 2022 at 15:26
  • 7
    "able to reconstruct their dataset" this probably makes a number of assumptions about the original researchers' methods w.r.t. rounding, handling of outliers, other data-wrangling stuff that they may or may not have mentioned in their report. You should just be aware that, due to all of this indirection, your analysis is now subject to such limitations of THEIR study, and that you should discuss this in your report.
    – Him
    Commented Feb 21, 2022 at 18:00
  • 4
    Honestly, this whole approach sounds concerning: using a small sample size study's summary statistics (that's 6 s- in a row!) to then use some "newer methods" on it without having access to underlying distributions... Ouch!
    – Lodinn
    Commented Feb 22, 2022 at 10:03

3 Answers 3


Yes. This is published work, so you are free to do whatever you want with it, modulo the usual caveats. The fact that you are "reconstructing their dataset" is irrelevant; the key point is that you are building on their prior work using only published, publicly-available information. As you mention, you'll have to be transparent in both directions: give credit for the parts you take but be sure not be imply that you have any "inside information" about the original study.

  • 15
    I agree, but first I think OP should tell the researcher that they are able to reconstruct the dataset and what they are planning to do with it. The researcher might reply when told this, and might even give the original data, which would be useful as a check, or some other useful information.
    – gib
    Commented Feb 21, 2022 at 11:23
  • 2
    I would also mention the methodology/steps/code you followed to reconstruct the data set (maybe in apendix) and explicitly mention the reconstruction. Commented Feb 22, 2022 at 15:07

Talk to your IRB.

The original participants in the study may or may not have had some expectation of privacy w.r.t. their data. This existed as a real, official agreement between the original researchers and the participants, and likely went through some kind of IRB approval process. The authors almost certainly never even considered the possibility that the data could be perfectly reconstructed from the summary stats they provided, and anticipated that not providing the raw data was sufficient to keep the participants' individual responses anonymized. Possibly, the participants were explicitly told that the raw data would not be publicly available! You say that determining identities is likely not possible, but this probably depends on various things. Often, small studies are convenience-sampled, and the participants are easily identifiable as the students taking Psych2053 at PolyTech U under professor Study Author the semester before the study was published.

Anyhow, all of this is something you should probably bring up with your own IRB. I'm sure they'll be fine with it if you take the usual precautions of not making the data publicly available, or what have you. Still, it would suck to have the fact that you didn't get approval to come back to haunt you.

  • 7
    I can appreciate this answer, but would also be a little bit concerned that IRBs tend to be extremely conservative. The OP did say explicitly "the sampled population is large" (so the small sample is presumably a small sample of respondents from a large surveyed population). If the OP is really worried they could analyze the reconstructed data set and only publish summaries/results of the analysis (i.e., leaving it up to others who want the original data to reconstruct it for themselves). In any case, the information is publicly available now ...
    – Ben Bolker
    Commented Feb 22, 2022 at 0:19
  • 3
    Unfortunately I feel like any expectation of "privacy" the participants may have had is rendered moot once the information has been already published with their informed consent ("informed" to an extent reasonable at the time). In fact I feel that such a post-facto restriction would not only hinder the advancement of science, but potentially create a perverse incentive for future unscrupulous researchers to obtain poorly-anonymized datasets in a similar manner as the original authors, precisely to make it more difficult for future researchers to question their results...
    – user541686
    Commented Feb 22, 2022 at 12:50
  • 9
    From a standpoint of personal ethics, I think it's certainly kind of the OP to ponder this, and perhaps refrain from publishing if the data appears sensitive in his own eyes, but otherwise, from the standpoint of professional ethics, given the above, I don't see why an IRB should have any input in this.
    – user541686
    Commented Feb 22, 2022 at 12:54
  • 1
    Let's make some reasonable assumptions. Personal identities have been replaced in original publication. Original publication and deriving works use aggregates. OP wants to reconstruct original data. The only identities OP can use are the published pseudonyms. OP is surely not stealing the original un-pseudonymized dataset from a safe deposit! So, there seems to be no way for OP to breach the identities of the study participants. But there are two questions: Is it ethical towards the original study participants? And: Is it ethical towards the original paper authors? Commented Feb 22, 2022 at 17:41
  • "OP is surely not stealing the original un-pseudonymized dataset" this is fair, but the fact remains that the OP has no idea what the original study participants were told would happen to this data. It's also not entirely clear from the description how identifying the data are. Certainly this qualifies as some level of de-anonymization over and above providing summary statistics. How un-anonymous is un-anonymous enough? This is precisely the sort of thing that IRBs are for!
    – Him
    Commented Feb 22, 2022 at 19:00

I don't see how ethics is involved here, especially as the data set is 20 years old.

How can you be sure that any result (positive or not) is caused by the new method and is not an artifact of your potentially insufficient data set reconstruction?

Why do you cannot create a new data set, use the old methods, check that the result confirms the old paper, and then try modern methods? And publish your new data set. I think that would be the better, more scientific approach.

You must log in to answer this question.

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