Enter experimental and observational data, and the reward (or cost) of acting on each counterfactual type. The calculator combines both data sources to bound the proportion of each type, using the Tian–Pearl bounds as presented in Mueller & Pearl (2022), and returns the expected benefit of handing out the flyer (acting) rather than not.
| P(ticket) | ||
|---|---|---|
| with flyer | P(y | do(x)) | |
| without flyer | P(y | do(x')) |
Chance of getting a ticket when the flyer is randomly handed out or withheld.
| P(ticket) | ||
|---|---|---|
| with flyer | P(y | x) | |
| without flyer | P(y | x') |
Chance of getting a ticket among people who chose to take the flyer, and among those who didn't. Half of the people are assumed to take the flyer, P(x) = 0.5, as in the experiments.
| complier | ticket only with flyer | |
|---|---|---|
| always-goer | ticket either way | |
| never-goer | never a ticket | |
| defier | ticket only without flyer |
The defaults are the experiment's payoffs: +$10 for a returning customer, −$20 for a discouraged customer, −$0.50 per flyer. That gives complier 9.5, always-goer −0.5, never-goer −0.5, defier −20.5.
Each row shows 10 people. Click a slot to set how many of them got a ticket. Partial tickets show fractional values.