In analyzing noise, the number of photosites (pixels) is entirely irrelevant; the important thing is signal to noise ration (SNR) per unit area. Therefore sizing images so they match resolution is a valid idea.
However, mathematically, the right place to do this is with the NEF; doing it with the JPG is only an approximation. This is due to the nature of demosaicking, tone curves, etc.
Furthermore, comparing JPGs from two different cameras of two different scenes with difference firmware settings, lighting, etc. is problematic.
Regards, Bill P.S. - BTW, not all noise is random; but that's a different subject!