Sensitivity
—
of those in need, found
Specificity
—
of OK people, left alone
Precision (PPV)
—
of those treated, in need
Return on investment
—
saved per $1 spent
Total benefit
—
fiscal only — see note
Receiver Operating Characteristic (ROC)
The curve is fixed by your targeting quality. Your policy choice picks one point on it.
Decision outcomes at this choice
Expected headcounts. Green = correct, amber = the two kinds of error.
Has condition
OK
Total
Treated
True positive
—
False positive
—
treated
—
Not treated
False negative
—
True negative
—
left
—
Total
in need
—
OK
—
population
—
Return on investment
Saved per dollar spent, as you treat further down the list. In the real world this often humps rather than simply falling — the highest-risk can be harder to turn around (though they may carry a larger liability), so the very top isn’t always the best value.
Total benefit
Liability averted minus all treatment cost. The peak is the optimal investment point — treat past it and each extra dollar returns less than $1. Fiscal only (see note).
—
Full breakdown
Every figure for the current scenario — the spreadsheet, made live.
| The world | |
| Population | — |
| In genuine need | — |
| OK (no need) | — |
| Targeting | |
| Discrimination (AUC) | — |
| Treated (% of population) | — |
| Number treated | — |
| Sensitivity (Se) | — |
| 1 − Specificity | — |
| Specificity (Sp) | — |
| Precision (PPV) | — |
| Neg. predictive value (NPV) | — |
| Economics | |
| Cost per treatment | — |
| Total treatment cost | — |
| Effectiveness | — |
| Successful treatments | — |
| Forward liability / case | — |
| Total liability at stake | — |
| Liability averted (saving) | — |
| Remaining liability | — |
| Bottom line | |
| Total benefit | — |
| Return on investment | — |
Teaching scenarios
Load a worked example, read the lesson, then move the sliders to explore for yourself.
Pick a scenario above to load its parameters and read the lesson.