de-embedding vs response calibration

Hello,

I have a doubt about de-embedding and response calibration.
I want to model a transistor on-wafer. To do this, I will calibrate on my probe plane using calibration substrate.
After that, I have do de-embed pads and lines to only "see" my transistor. This can be done measuring open and short response and make a model of pads and line. But, it would not be easier, more automatic, to make a response calibration (find a new correction matrix with short and open) to directly measure in the transistor plane?
Like this: http://ena.support.keysight.com/e507...ction_test.htm
I always saw the first way (deembedding), why?

Regards,
wig