CMatrixLab
by Martin Grossman martin.grossman@gmail.com
Your flu test is positive. What’s the chance you really have flu? That’s precision—and it also depends on how common flu is among the people tested.
Compare 10% and 50% sick using the same test. Precision changes, even though the portion of sick people correctly testing positive (TPR) and the portion of people without flu getting a false alarm (FPR) stay the same.
0% · no one has flu
100% · everyone has flu
TP · Correct detection
FP · False alarm
FN · Missed
TN · Correctly ignored
Full marker = 10 people; smaller = partial.
Test +
Test −
Actual flu
500 cases
500 cases
TP · Flu detected
0
FN · Flu missed
0
Actual no flu
500 cases
500 cases
FP · False alarm
0
TN · No flu, test −
0
Among positive tests:
—
Green:
0
have flu
Red:
0
false alarms
1,000 people (averages). Sick portion changes the counts and Precision; TPR and FPR stay fixed at the same test cutoff and separation. · 📋 Cheatsheet ↗ · 📖 Metrics Reference ↗