We analyze shear wave polarizations from local earthquakes recorded by the Anza network in southern California, using an automated method which provides unbiased and quantitative measurements of the polarization and the duration of linear motion following the shear wave arrival (the linearity interval). Initial shear wave particle motions are strongly aligned at four stations, a feature that is not predicted by focal mechanisms. The particle motion alignment is most likely caused by shear wave splitting due to anisotropy beneath these stations, a result supported by the clear shear wave splitting seen in a borehole recording near one of the Anza stations. These results are consistent with an earlier analysis of these data by Peacock et al. <1988>. However, our analysis does not support claims by Crampin et al. <1990> that shear wave splitting delay times at station KNW exhibit temporal variations which can be correlated with the occurrence of the North Palm Springs earthquake (ML=5.6) of July 8, 1986. Automatically determined linearity intervals scatter widely from 0.02 to 0.15 s and exhibit no clear temporal trends. We find a correlation between earthquake moment and the linearity interval, possibly a result of longer effective source time functions for the larger events. The inability to identify a distinct slow quasi-shear wave pulse for the vast majority of these events indicates that scattering strongly affects the particle motion, even in the very early shear wave coda. Analysis of earthquake clusters with similar waveforms recorded at KNW shows that seismic Green functions are stable throughout the observational period and that most linearity variation is due to source and ray path differences between events. If shear wave splitting is causing the observed delay times between horizontal components, the waveform stability for events in these clusters restricts any temporal changes in shear wave splitting delay times to less than 5--10%. ¿ American Geophysical Union 1990