I regularly think in terms of decibans, where 10 decibans = 90% confidence, 20 = 99%, 30 = 99.9%, 40 = 99.99%, etc. From that perspective, there’s as much difference between 99% and 99.9% as there is between 99.9% and 99.99%. Put another way, if 99% is the cut-off, then anything more than 20 decibans of evidence is ignored, even if the answerer has at least 40 decibans.
I would be quite surprised if any of us could justifiably estimate any of those questions to an accuracy better than the nearest 1%/epsilon offset.
I regularly think in terms of decibans, where 10 decibans = 90% confidence, 20 = 99%, 30 = 99.9%, 40 = 99.99%, etc. From that perspective, there’s as much difference between 99% and 99.9% as there is between 99.9% and 99.99%. Put another way, if 99% is the cut-off, then anything more than 20 decibans of evidence is ignored, even if the answerer has at least 40 decibans.
And yet every year lots and lots of respondents answer such questions with 0 or 100.