8

I am currently engaged in some volunteer data work with a charity, I would like to create a write up of the work that can be shared with a few partners who do not have full access to the data.

The write up will be mostly statistical overview, which is not sensitive. The statistics are drawn from data-sets which are GDPR protected. In order to illustrate the data pipeline a few examples would be useful, so I will create a small fictional data-set with the same form as the real data.

To avoid any confusion, I need to declare that the examples are fiction (illustrative only) but the statistics are real. Is there a standard way to go about this?

It will look like;

The result of the data pipeline will be to tranform mixed type data into a float vector the input is .... Output fields are .... For example;

"Joe Bloggs|Carmarthen|06:00|Negative|1995|May|Finance|271946"

will become

"267.3|82.|168.|-35.|.0777|1648".

Note the example is not an anonymized record drawn from my data, it is pure fiction.

  • 1
    Is this project under the supervision of your IRB? If not, why not? – Nate Eldredge Apr 24 at 15:41
  • 2
    If you want to have a chance of publishing the work academically in the future, then the project likely needs to be approved by your IRB from the outset. (And if you do not intend to publish it or use it for some other academic purpose, then your question would be outside the scope of this site.) – Nate Eldredge Apr 24 at 16:05
  • 3
    @Clumsycat Ethics committees are not a US-only thing. – Azor Ahai Apr 24 at 16:42
  • 2
    @Clumsycat: I'm really referring to the general principle that academic research involving human subjects needs some sort of ethics oversight. I do realize that the specific mechanisms and terminology vary in different countries, and I agree I should have used more general language. – Nate Eldredge Apr 24 at 16:50
  • 1
    Many journals have an ethics-oversight requirement as well. So even if your university/country doesn't require it, you may not be able to publish without IRB-equivalent approval. – JeffE Apr 24 at 21:06
5

It is common to use words like "fictitious", "fake data", and even a footnote explicitly saying "these are not actual values" or "values for illustrative purposes only". It is even more common to just pick values intentionally that are obviously fake, so that no reasonable person will think they might be real, such as "John Smith|Gotham|01:11|1999|Jan|Accounting|12345" - etc.

I will however note that there is a way to handle this specific case that is even more high-level, which is just not to do this at all, because most stakeholders couldn't care less, much less even know what a float vector is or how they are suppose to interpret it. Telling people details they don't need makes many people think it must be important so they try to latch on to any random interpretation they can think of - because why are you even telling them if it doesn't matter? So if they are not a very specific type of user, just leave this out entirely.

If they are in the tiny, tiny minority of stakeholders that care about this incredibly low-level amount of detail, then it is reasonable to assume they will recognize obviously fictitious data even if you don't say it, and if you really are worried you can just give an asterisk and say its fictitious - but that's already excessive. If you need to demonstrate for some sort of officious bureaucratic sect, declare in the footnote that the data is fictitious for illustrative purposes, persuant GDPR section blah subsection blah blah blah. Give the lawyers something to smile about, if that's your audience, but you shouldn't let it take over the rest of the presentation of the work.

  • Thank you. In this case the reason for giving examples is that this note should serve as documentation for less-technical users. – Clumsy cat Apr 24 at 16:44
1

How about " real data, but the names / labels have been changed"

  • No part of the published record will come from a real record. It's probably overkill, but don't want to risk it exposing something it shouldn't. the example could be like; "Joe Bloggs|Carmarthen|06:00|Negative|1995|May|Finance|271946" and the only relation that would have to real data is the fields contain the same type of data. – Clumsy cat Apr 24 at 14:53

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.