This is due to a randomized test is always a straight line. Concave upwards is therefore useless.
Above the line PD=PFA
The slope in a given point equals the threshold to achieve power and size of that point $$
\frac{\partial P_{D}}{\partial P_{FA}} = \gamma
## Intuitive Understanding
![[Pasted image 20240221123433.png]]
We want a distribution like the yellow one. It is a better detector.
This is due to the hypothesizes are getting shifted away from each other, *making it easier to distinguish.*