If someone learns to catch a ball we don’t assume they know Newton’s equations of motion, and the relevant variables to apply it to the ball, and we certainly don’t expect to get those out of our ball catcher via a questionnaire.
In the real world, many mathematical models involving people do have to get parameterised by questionaires. How do we make that work? My favourite answer is “make someone else do it”. Other answers include hierarchical models.
- a short diatribe about ugly regression outputs with Australians Climate change attitudes as a case study