I wonder if you’d get better probability values if you used AJAX slider controls for a continuous value between 0 and 1. Less chance of anchoring percentages on multiples of 10 and 5.
In a survey does an increase in rounding errors in estimators a problem? As long as there’s no bias in how they get rounded we should be fine. If there is such a bias I’m curious what it is and what causes it.
That would make it effectively impossible to distinguish between 1 and 0.001, or 99 and 99.999. To get around that we’d need to work with something like log(odds ratio), but then there isn’t any natural choice of endpoints, people’s intuition for what goes where will generally be poor, etc.
I wonder if you’d get better probability values if you used AJAX slider controls for a continuous value between 0 and 1. Less chance of anchoring percentages on multiples of 10 and 5.
In a survey does an increase in rounding errors in estimators a problem? As long as there’s no bias in how they get rounded we should be fine. If there is such a bias I’m curious what it is and what causes it.
I suspect it would have a strong bias towards obvious fractions and obvious multiples. That isn’t a directional bias, but it’s an anti-precise bias.
That would make it effectively impossible to distinguish between 1 and 0.001, or 99 and 99.999. To get around that we’d need to work with something like log(odds ratio), but then there isn’t any natural choice of endpoints, people’s intuition for what goes where will generally be poor, etc.