I started a postdoc about a year ago. At the time of my hire I was told another postdoc was preparing the large dataset I would eventually use. In the meantime, I was to run some sample models and develop a method on some small test data.

Ten months later when I finally received the data, it was a mess of a spreadsheet (extremely long column names, no column names, typos, no documentation) and I also began finding errors in it immediately. For example some merged in values had been multiplied by 2 or 2.5 or 2.25. Nobody could explain what happened so I spent nearly 2 months redoing those values.

It gets worse. I started checking the other columns and realized some "no data" indicators had been averaged with measured values. This produced large negative values which should have been a red flag for that data. There were other similar sloppy errors in the data. I've tried to be extremely polite with the other postdoc but I'm not getting a lot of help from this person.

I've spent nearly 3 months cleaning up the mess and I continue to find problems. My budget is nearly up (I have about 3 months left) and I'm being asked to start writing and prepare a draft in the next three months. I worry there are more problems with the data that I'm not seeing because the person who prepared it did not have high attention to detail. It was also prepared by hand in Excel and not well documented or not documented at all. I don't want to put my name on something that uses data that is potentially full of errors, especially since I'm early in my career. I've voiced these concerns somewhat indirectly to my main supervisor and he is nothing but reassuring and even defensive of the data. Another colleague, not the one in charge of the project, agrees with me.

What should I do?


You have come across an important lesson in research: if the person or organization providing your data is not responsible somehow for that data being correct, it won't be. Collecting and preprocessing new data is an iterative process with lots of restarts as you discover that common techniques need to be adapted to the particular dataset. If someone just hands you data they didn't have to test or validate anywhere, what you get is their first failed result. If they touch the data to do anything to it at all, they almost certainly did so incorrectly; if there is a sign, they probably created sign errors; if they did any sorting, they probably did not sort the labels and values consistently, etc.

As for how to proceed, there's another unfortunate lesson here for anyone in a mentored research position: you should generally do whatever your supervisor wants. I don't mean behave unethically if they push you to. But to change your focus to doing the best with what you have rather than letting the perfect be the enemy of the "at-least-you-did-something". Honestly your supervisor probably has a better grasp of the big picture. I.e. getting a result out to show the funding agency you were productive, getting a notch on your CV, and such outcomes, are the critical ones. While a wonderful research result that drives the science world vertically is probably out of reach already. You can however adjust how you present the data in your paper, e.g., focus on the methods not the results, or put some mixed validation results and note not all of them worked properly, caution that your data may be flawed as one possible cause, and so forth.


While the advice of A Simple Algorithm is sound, I'd add that you can, in fact, let your supervisor know that the data you got "isn't ready for prime time." Be prepared to show why and what you see as issues with it. You will, I hope, then get advice that you should probably follow even if you don't agree assuming you want to preserve the relationships.

It isn't necessary to blindly go along, nor is it necessary to scream and rant. Just point out the flaws and why they will negatively impact the work. It is possible that the supervisor has some influence to get the data improved that you don't have directly. Of course, one option is that you get the improvement dumped on you, which isn't fair or ideal, but it will result in better work in the end.


I don't want to put my name on something that uses data that is potentially full of errors

You are right. Do not publish if you do not think the data is correct.

Next time address the data quality at the start of the project.

  • 3
    Q: how to tell my supervisor I won't use the data? A: next time do it at the start of the project. Probably good advice, but not an answer that helps OP. – user9646 Aug 4 '18 at 13:33

Your Answer

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