|
Size: 423
Comment:
|
Size: 1191
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 3: | Line 3: |
| At its simplest, SignalDetectionTheory or SDT is a model for decision between two hypotheses based on the value of a measurement, ''x''. | At its simplest, SignalDetectionTheory or SDT is a model for the situation of a decision maker choosing between two hypotheses based on the value of a measurement, ''x''. |
| Line 7: | Line 7: |
| It is the job of the '''Observer''' to decide whether it was 'Signal' or 'Noise' that produced 'x'. | It is the job of the '''Observer''' to decide whether it was 'Signal' or 'Noise' that produced ''x''. The assumption that larger values of ''x'' are more typical under ''f,,1,,'' than under ''f,,0,,'' leads to the use of the magnitude of ''x'' as a criterion (''e.g.'' Choose ''H,,1,,'' when ''x>c'', otherwise chose ''H,,0,,''). The performance of this criterion is given by the Hit Rate (''P(x>c|f,,1,,)'') and the False Alarm Rate (''P(x>c|f,,0,,)''). These two quantities are also known in '''Neyman-Pearson-land''' as '''Power''' & '''Size''', or as '''Sensitivity''' and '''1-Specificity''' (the complement of '''Specificity'''). When ''1-Specificity'' is plotted against ''Sensitivity'' as a function of the criterion ''c'' the resulting curve is known as the [:Glossary#ROC: ROC or Receiver Operating Charactistic]. |
Signal Detection Theory
At its simplest, SignalDetectionTheory or SDT is a model for the situation of a decision maker choosing between two hypotheses based on the value of a measurement, x.
Under H1, x comes from the Signal distribution f1 and under H0, x comes from the Noise distribution f0.
It is the job of the Observer to decide whether it was 'Signal' or 'Noise' that produced x.
The assumption that larger values of x are more typical under f1 than under f0 leads to the use of the magnitude of x as a criterion (e.g. Choose H1 when x>c, otherwise chose H0).
The performance of this criterion is given by the Hit Rate (P(x>c|f1)) and the False Alarm Rate (P(x>c|f0)). These two quantities are also known in Neyman-Pearson-land as Power & Size, or as Sensitivity and 1-Specificity (the complement of Specificity).
When 1-Specificity is plotted against Sensitivity as a function of the criterion c the resulting curve is known as the [:Glossary#ROC: ROC or Receiver Operating Charactistic].