A statiscal method to evaluate TX antenna performance using WSPR

Failing to find a proper calibration for VOACAP to use it as reliable reference to evaluate antenna or site performance (*), I decided to try something else. The method I imagined consists in switching 2 antennas every 10 minutes, 1 second before minutes 00, 10, 20, 30, 40 and 50. Selecting a 33% TX'ing rate, in WSPR you have the guarantee to have a minimum of 1 transmission per 10 minute period (most of the time 2). Swiching is done using a little software available for free on the web and a simple interface to the LPT-port using a NPN transistor to drive a coaxial relay (similar to the popular CW interface used for K1EA's contest sofware "CT").
Following are a few examples, based on limited number of 30m spots, which however seem to validate the method. The antennas used are 1) a 5/8 ground-mounted full size vertical with about 30 burried radials with a simulated gain of 0.5dBi @16 degrees 2) An inverted V dipole, with appex at 10m oriented for AZ 130/310, with gain of approx 5 dBi @50degrees.
The first graph show the spots received by VK6DI on April 23. The distance is 14000 kms, the azimut from here, is about 100degrees (so about 30 degrees from the dipole main lobe). For this circuit, and at the time of the spots VOACAP indicates a vertical angle between 3 and 10 degrees (function of the time). All the spots have been corrected, to show a normalized TX-power of 1W (I deducted the corresponding 6dB's, when 4W was used). 4 spots received by VK6DI between 2140 and 00:48 were removed, as they were received only when I was TX'ing on the vertical (leaving those spots would have biaised the comparison). The superiority of the vertical is quite obvious, and peaks at 6dB just before 20z, with an average of 4dB across the window. This is quite normal, considering the gain of the dipole at angles below 10 degrees.
The second plot is from the spots received by K1JT. The distance is 6300 kms and the azimut from here is 296 degrees. VOACAP indicates a radiation angle around 10 degrees for the considered window. The difference peaks to 7dB in favor of the vertical, with an average 3dB accross the window.The 3rd plot is based OY3JE's spots. Distance is 2000kms and the AZ 340 degrees. Here the average received signal is about the same on the 2 TX antennas, but with higher peaks on the dipole.
Reducing further the distance gives a slight advantage to the dipole. Above are GM3AKF spots, at a distance of 1400 Kms and AZ of 338 degrees.
Finaly are the spots received by EA1FAQ at a distance of 1000kms and a bearing of 245 degrees. Here the vertical takes again the advantage, likely due to the directivity of the dipole.
As mentionned earlier, the above plots are based on a limited number of spots, the method being statiscal, more spots, more accurate results... The results may also be biased by the interaction of the 2 antennas only located 15meters away. Nevertheless, the results seem quite coherent with the antenna simulation, and show that WSPR and the network of stations that is growing every day is a fantastic tool, not only for propagation evaluation, but also for antenna and site evaluation (ie using 2 stations a few kms away from each others using the same antenna). The same method could also be used to evaluate RX antenna perfomance.
(*) important differences can be seen function of the selected circuit, so evaluating an antenna or a site with sufficient accuracy, even over a period of 3 weeks was a kind of utopy...