This document contains instructions, some answers, comments, tips and tricks, and further explanations for the exercises.
Define: POS = “LIAR”; this is a choice, it’s also possible to define “LIAR” as a negative.
Based on performing the test a couple of times, a so called Confusion Matrix can be created.
| ACTUAL | ||||
|---|---|---|---|---|
| POSITIVE | NEGATIVE | TOTAL | ||
| PREDICTED | POSITIVE | TP | FP | PRED.POS |
| NEGATIVE | FN | TN | PRED.NeG | |
| TOTAL | POS | NEG | N (sample size) |
In the exercise it is not clear what is meant by ‘the probability of
a false positive’; it could at least mention three probabilites (which
three?).
However there is a definition for the False Positive Rate FPR, this is
the rate of the Actual Negatives predicted (or classified) as Positives.
Or as a formula: \[ FPR =
\frac{FP}{NEG}\]
So if FPR = .08 it means that using the lie detector many many times, on
average the lie detector classifies 8% of non-liars as liars.
Remark
In the literature and on the internet lots of examples of Confusion
Matrices (CF’s) can be found. Be aware that there is no uniform
structure for a CF. Sometimes the predicted values are in the rows and
the actual values in the columns, as in the CF above. Sometimes it is
the other way around. Sometimes the first mentioned categories are the
Positives, and sometimes the Negatives are mentioned first.
For instance, a Confusion Matrix may look like this:
| PREDICTED | ||||
|---|---|---|---|---|
| POSITIVE | NEGATIVE | TOTAL | ||
| ACTUAL | POSITIVE | TP | FN | PRED.POS |
| NEGATIVE | FP | TN | PRED.NeG | |
| TOTAL | POS | NEG | N (sample size) |
EXERCISE
Fill in TP, TN, FP, FN, POS, NEG, PRED.POS en PRED.NEG in the correct
cell in the CF below.
| PREDICTED | TOTAL | |||
|---|---|---|---|---|
| NEGATIVE | POSITIVE | TOTAL | ||
| ACTUAL | NEGATIVE | |||
| POSITIVE | ||||
| TOTAL | N (sample size) |
Read the text carefully and you see that a Positive is defined as a woman who has breast cancer.
Definition False Negative Rate, the rate of Actual Positives
classified as Negatives.
\[ FNR = \frac{FN}{POS}\] FNR = 0.10
means that on the long run of all the women who have breast cancer, 10%
is classified as not having breast cancer.