Sensitivity and criterion ■ The performance is a process of two independent properties: a)Sensitivity (discriminability) § How the observer perceives stimuli § One property of the sensory process § Measure of how close the noise and the signal are b)Criterion (response bias) § How the observer choose to respond § Susceptible to motivation, decision making strategy, etc. § Measure of what is considered noise and what signal
Sensitivity and criterion Low sensitivity High sensitivity Unbiased criterion Biased
Relating criterion and sensitivity ■ A receiver operating characteristic curve (ROC) represent the trade off between False Alarms (FA) and Hits (H). ■ The area under the curve (Au. C) reflects the sensitivity (d’) ■ Perfect classifier - Au. C: 100% ■ Random classifier – Au. C: 50% ■ The closer the graph is to the top and lefthand borders, the higher the sensitivity is. ■ The closer the graph is down and right-hand borders, the lower the sensitivity is. 0, 16; 0, 84 0, 75 p(H) ■ Better classifier – Au. C: 75% 0, 84; 1 1, 00 0, 50 0, 25 0, 00 0; 0, 16 0, 00 0, 25 0, 50 p(FA) 0, 75 1, 00
Increasing sensitivity, increases the slope of ROC
Πηγές ■ A brief introduction of Signal Detection Theory, Dr. Zhuanghua Shi ■ www. datasciencecentral. com ■ www. statisticshowto. com