Ordinal Regression Quiz

Answers and explanations appear as you take the quiz.

1. What property of ratings does ordinal regression preserve?
   • Their order without requiring equal gaps
   • Their exact numeric distance
   • Their independence from predictors
   • Their equal category frequencies
2. In a latent-threshold interpretation, how are observed rating categories produced?
   • It fits unrelated category models and discards category order
   • Thresholds partition a latent tendency into ordered ranges
   • It treats coded outcomes as equally spaced measurements
   • It assigns each category an independent numeric regression slope
3. What does proportional odds commonly assume about predictor effects?
   • All thresholds are identical
   • Every category has equal probability
   • The same slope applies across cumulative category splits
   • Predictors have no effect on rankings
4. Why can ordinary regression on codes 1, 2, 3 and 4 be inappropriate for satisfaction levels?
   • It cannot include predictor variables
   • It treats category codes as nominal labels and ignores their order
   • It estimates cumulative probabilities but never uses the category codes as distances
   • It assumes numeric gaps between codes and can predict values outside the category set
5. A predictor shifts latent satisfaction upward. What may happen to a middle-category probability?
   • Every category probability must rise by the same amount
   • A middle-category probability may rise or fall as mass crosses its thresholds
   • Only the lowest and highest categories can change probability
   • The category probability becomes equal to the latent score
6. Why should category labels be encoded in their intended order?
   • To force all predictor coefficients to be positive
   • To make all thresholds equally spaced
   • To ensure software uses the intended semantic category order
   • To guarantee the proportional-odds assumption holds
7. Which model restriction should be assessed in a common ordered-logit specification?
   • Whether cut points are equally spaced
   • Whether categories occur equally often
   • Whether each threshold has a different slope by definition
   • Whether a common predictor effect across cumulative splits is plausible
8. A wrong prediction one level away has a different cost from one three levels away. Which evaluation choice is appropriate?
   • Use an order-aware metric or explicit costs that distinguish near from distant mistakes
   • Use ordinary accuracy because it records how many levels apart each error is
   • Use squared error on category codes without checking whether equal spacing is meaningful
   • Ignore validation because category order already defines model quality