[Richard Williams]

Do you have the original electronic or paper records? Ideally you would go back to them. There may be a coding error in either id or gender.

Gender and self-identified gender can also change across time. Coding errors are probably much more likely, but real change can happen. I'd be curious to know if there is any research on how often this happens and what survey researchers do about it.

[DavidHarmsen]

I haven't been able to find coding errors, do you recommend a special method to check for that? I also deleted and re-entered their gender just to test and see that the issue remains.

[Richard Williams]

The best method is to go back to the original records, e.g. the survey questionnaires. If that isn't possible, I suppose you could check to see if 9 times they were female and once they were male. Maybe you will want to code gender as missing if there are conflicts that can't be resolved.

[DavidHarmsen]

I checked the original records and I checked if either 1 of those 2 persons had a gender change in 1 of their records, nothing seems to be wrong. So I better code the gender of those 2 persons as missing but how do I do that? Replace the gender with a space or a . or something else.

[Richard Williams]

Are you saying that, in the original records, they did not report different genders across time? If so, it sounds like the electronic records got miscoded and should be corrected.

Or, are you saying that in the original records, they actually did report different genders at different times?

[DavidHarmsen]

they did not report different genders across time

[Richard Williams]

In that case it sounds like the electronic records were miscoded. Correct the errors. Be sure to carefully document the changes you make, and why.