I have egregiously sloppy (possibly falsified) data that I need to correct. How should I go about this?

Question

Several years ago, I did a summer project as an undergrad with a PI not at my undergrad institution (I'm in grad school now). Now, the project is finished and he wants to publish it with data that I've gathered.

The problem is, after checking my data, that the spreadsheets that I used to calculate the data are very inaccurate, to say the least. I can't remember exactly what I did, but there do seem to be some egregious inaccuracies (as in, values from the wrong dataset) in my spreadsheet compared to the raw data from the instrument either that I was colossally bad at copying data over correctly (all of the data in my spreadsheet do appear in the raw data someplace, but not necessarily in the right location, the right dataset, etc.), or worse, I'm terrified that I may have done it deliberately at the time. (How would someone be able to tell?) I did analysis with the faulty data back then and produced graphs.

Obviously, I will have to supply the correct data analysis before it gets published. But is it ok to explain it as "I majorly screwed up the analysis and made a lot of data entry errors, here's the corrected version"? I'm hoping that my PI will just use my corrected data and analysis and just focus on that instead trying to pore over my wrong spreadsheets to individually compare all the changes, many of which are embarrassing at best and look suspicious at worst. I am not under any suspicion of wrongdoing thus far.

I redid my data analysis with correct data entry and the results look messier on most of the things I messed up, though they do not affect the underlying trends. Not sure what this means - how likely is it for honest error to have the effect of making data look cleaner than it actually is? Also, my god, I was awful - mishmashing data from one trial with another, copying data from the wrong wavelengths and/or from the wrong trial, wrong normalization numbers, etc. etc. etc.

If you were a professor and you were approached by a student who says that made egregious errors on past data analysis and the corrected data looks not nearly as good, what would you think? Would you think they were honestly trying to fix a mistake, or trying to save face and cover for past misconduct?

Answer 1

I infer from the question that you still have the raw data. In that case, I think you need to re-do all the steps in which errors could be introduced (such as data entry, spread-sheet calculations, analysis of trends). Then see what effect the corrections have on the conclusions of your paper, and rewrite as necessary. I realize that this sounds like a lot of work, essentially repeating most of what you had done earlier, but since you apparently did it wrong the first time, you now need to do it right.

Meanwhile, you should tell the PI about the problem and your plans for correcting it. If I correctly understand the situation, the raw data are the only reliable information currently available. You should make sure he knows that, so that he (and others working with him) won't build further research on a foundation of wrong results.

Answer 2

Everyone will have their own reaction, of course, but the more important thing is that you point out the errors and the need to correct them. Going forward with bad data is the worst outcome.

Work up the best data you can and tell the prof that the older data was seriously flawed.

I think most people will accept your honesty, even if it means some delay. And most will forgive at least some of it due to your lack of experience at the time.

Hopefully the original data is still available making corrections possible.

But, you also need to be honest with yourself about how this all happened. I can think of a few very bad scenarios that I hope weren't contributing factors. There are actual medical conditions that might have such an effect. You don't want a recurrence.

Answer 3

I'll take your questions one at a time.

I'm terrified that I may have done it deliberately at the time. (How would someone be able to tell?)

I'm not sure if by "someone" you mean you, or the person you send this to. I can definitely imagine looking back at my own work from undergrad and doubting the integrity of it if it looked fishy. If I were trying to figure out whether I had done it intentionally, I would look for patterns in the errors (e.g. maybe treatment group is only ever misrecorded for data points that pattern against the tested hypothesis), and see if they jog your memory. I suppose that's probably what I'd do if I were trying to figure out whether someone else had done something intentional, but I seriously doubt most people would bother.

But is it ok to explain it as "I majorly fucked up the analysis and made a lot of data entry errors, here's the corrected version"?

Absolutely. You were an undergrad and it's the prof's responsibility to check your work and to ensure you know what you're doing. Also, the unfortunate reality is that lots of profs intentionally teach students in their labs to mishandle data to "improve" results, so that's a potential explanation here too. Because of that, I'd be more likely to say something like "Looking back at the data now, I'm finding a lot of errors, so many that I don't think correcting them is feasible, so unfortunately I believe that this data would have to be reanalysed from scratch to proceed with the project." That is, tread lightly if you can't recall the origin of the discrepancies, but make your own position clear.

Also, it is absolutely not your responsibility to provide corrected data if you don't want to continue with the project. I assume as an undergrad you were either paid or volunteering in exchange for the training/experience, and I assume neither of those is true of your relationship with this PI/lab now. I recommend asking yourself whether you honestly want to continue with this project before putting any more work into this.

I redid my data analysis with correct data entry and the results look messier on most of the things I messed up, though they do not affect the underlying trends. Not sure what this means - how likely is it for honest error to have the effect of making data look cleaner than it actually is?

More likely than not. Unfortunately, we're a lot more likely to double-check steps in our work if we don't like the look of the data, no matter how thorough/neutral we might intend to be.

If you were a professor and you were approached by a student who says that made egregious errors on past data analysis and the corrected data looks not nearly as good, what would you think? Would you think they were honestly trying to fix a mistake, or trying to save face and cover for past misconduct?

Depends on the student and the professor. Personally I trust the students that I work with or I wouldn't be working with them, so I would believe them. And I promise you, every researcher has encountered egregious errors in their past work at one point or another, often at this very stage of preparing/doublechecking for publication.