G4JNT reported some measurements of WSPR reported SNR vs input signal at (Talbot 2010).
His experiment connected a WSPR modulated RF source directly to an SDR receiver, and he recorded WSPR's receive SNR reports vs input attenuation and configured SDR receiver bandwidth. The direct connection means the test is not subject to normal radio path effects like fading.
The table above is derived from Talbot's, his information about the RF source (-30dBm) and attenuator settings are converted to receiver input power (dBm).
Above is the same data charted. A linear line fit to the 300Hz data is also included, it is a very good fit. The issue that Talbot raised is that the reported SNR is quite dependent on receiver bandwidth.
The chart above shows the sensitivity to bandwidth here at -120dBm input power.
The table above is a calculation of receiver noise figure implied by the first table. The mean of the measurements with 300 and 500Hz bandwidth is 27.2dB, and the cells between 26 and 28dB are highlighted in green. The receiver noise figure should be largely independent of the receiver bandwidth setting.
It is apparent that the results derived from reported SNR show that bandwidths below 200Hz return incorrect SNR, and the right hand column suggests significant non-linearity at SNR=2dB, a hint of some non-linearity of the measurement system (here with only one signal).
Talbot, A. Feb 2010. WSPR reported S/N measurements. http://www.g4jnt.com/WSPR_SNR_test.pdf (accessed 21 May 2017)