Ask a hundred people the same question.
The answers spread out.That is normal.
Badly tuned simulated respondents bunch together.
The average can still be right.
But two options drift together when they shouldn’t.
We compare the simulated spread with the human spread.
1.00: as much disagreement as among real people.
Before correction: 0.20 to 0.76, depending on the market.
After correction: 0.96 to 1.07 on attitudes.
Measured against public surveys.Not yet validated on new questions.
Space: pause · arrow keys: previous / next scene · click the bar: jump to a scene