Source effects on high-frequency regional seismic discriminant measurements involving P/S-type ratios (Pn/Lg, Pg/Lg) are isolated for earthquakes in southern California and southern Nevada. The scatter in earthquake population discriminant measurements for fixed propagation paths is statistically analyzed to assess the spatial and frequency-dependent contributions from source radiation pattern and source depth variations along with variable near-source scattering. This is useful because procedures for reducing the variance caused by propagation effects must be constrained by knowledge of the intrinsic scatter resulting from independent source effects, for which corrections are not available without a priori knowledge of the source type and depth. Broadband waveforms for earthquakes near the Nevada Test Site (NTS) and in southern California recorded at stations of the Livermore NTS Network (LNN), the Caltech TERRAscope Network, and the Berkeley Digital Seismic Network (BDSN) are used to obtain Pg/Lg and Pn/Lg measurements and to assess source depth and mechanism contributions to measurement variance. Pn/Lg and Pg/Lg spectral amplitude ratios for several frequency bands between 0.5 and 10 Hz from vertical velocity waveforms are computed for a total of 18 tightly located clusters of earthquakes. For each cluster, we examine the correlations between observed Pn/Lg and Pg/Lg ratios and estimated source depths or predictions based on previously determined source mechanisms. Focal mechanism predictions do not match the observations well, likely due to strong near-source scattering and directivity effects that overwhelm the point-source radiation pattern. However, the data do show significant scatter even for sources only a few kilometers apart. Source depth produces some coherent trends, which can be emphasized by averaging results from multiple stations. These trends vary from region to region, presumably due to specific local crustal layering, but most of the variance is not deterministically predictable and appears to arise from strong variations in near-source scattering. Variance estimates for logarithmic amplitude ratios normalized by mean amplitude ratios vary from 0.01 to 0.05 for different phase pairs. Frequency-dependent variations of the mean, variance, minimum, and maximum amplitude ratio parameters show strong distance dependence out to about 300 km and then appear to become independent of event-station distance. These variance parameters appear to be suitable for use in spatial interpolations of observed P/S-type ratios for a specific unidentified event path, as needed for nuclear test monitoring.