Even at the study design level they couldn't edit the questions, because they've got time series from other studies going back to the year dot. We've got problems with some of ours because language implications are changing, and so people aren't answering them as if they mean what they meant thirty years ago, and eventually they're going to have to be fixed but it can't be done by individual research teams. It isn't clear who it can be done by.

I guess what you could do, in a study as big as ours, is run two versions of the question on half the sample each, and then use the difference to standardise the time-series. But we're not doing anything like that, and most studies are much smaller.