Scoring Rules & Payoffs
The below information about scoring rules and payoffs was provided before the study and was always available during the study.
Scoring Rules:
Interval Response
- Description: Lower scores are better. This scoring rule rewards tighter intervals that contain the true value, and includes a penalty if the provided interval does not contain the true value.
- Equation: $${score} = ({upper} - {lower}) + 40 \times ({lower} - {true\ answer}) \times 1[{true\ answer < {lower}}] $$ $$+ 40 \times ({true\ answer} - {upper}) \times 1[{true\ answer > {upper}}]$$
Binary Response
- Description: Lower scores are better. Scores on binary responses range from 0 to 2—the minimum score is obtained by assigning 1 to yes_guess if yes is the actual answer and 1 to no_guess if no is the actual answer.
- Equation:
$${score} = ({yes\ guess} - {yes\ actual})^{2} + ({no\ guess} - {no\ actual})^{2}$$
- ${yes\ actual} = 1$ if actual answer is yes and 0 otherwise
- ${no\ actual} = 1$ if actual answer is no and 0 otherwise
Score → Payoff ($):
Once you have answered all questions in a “block,” i.e., a set of four questions about a dataset, we normalize your scores so they are all on the same scale, and translate the normalized scores into a payoff. This process is how we translate scores → money. your scores are normalized and translated to a payoff.
First, we calculate a normalized score for each question:
- ${score}_{normalized} = \frac{X - score}{X}$, where $X$ is a constant based on the dataset & question (see bottom of doc)
Then, we calculate a normalized score for the block:
- ${score}_{normalized\ for\ block} = \sum{{score}_{normalized}}$, for each question in the block
Finally, we translate the normalized score into a dollar amount by defining a payoff function:
- ${payoff} = {min}({max}(2.5 \times {score}_{normalized\ for\ block}, 0), 10)$
$X$ constants by dataset (ACS, DIA, STU) and question:
Census (ACS)
Census survey dataset
Question | $X$ constant |
---|---|
(race, age) | 4043.29 |
(race, number of children) | 217.69 |
(age, marital) | 56.80 |
(marital, income) | 2.00 |
Diabetes (DIA)
Diabetes dataset
Question | $X$ constant |
---|---|
(admission type, days in hospital) | 113.40 |
(age, number of medications) | 283.30 |
(race, days in hospital) | 186.90 |
(admission type, age) | 2.00 |
Student (STU)
Student evaluation dataset
Question | $X$ constant |
---|---|
(instructor, instructor's knowledge) | 3703.00 |
(enjoyed class, instructor) | 185.60 |
(instructor's knowledge, difficulty) | 56.80 |
(attendance, difficulty) | 2.00 |