Although horizontal slip directions determined for earthquakes along transform faults and shallow subduction thrust faults are commonly assumed to parallel the direction of motion between two plates, systematic biases of earthquake slip vectors relative to the local long-term average plate direction can be introduced by a variety of causes, including deformation of forearcs above subduction zones, unmodeled lateral velocity heterogeneities near earthquake sources, and unrecognized changes in plate velocities within the averaging interval of the kinematic model. Given that earthquake slip vectors comprise nearly two thirds of the data used to derive the NUVEL-1 model for present-day global plate velocities, it is important to determine the extent to which potential systematic biases of earthquake slip vectors could degrade estimates of plate velocities given by NUVEL-1. Here, two alternative models for global plate velocities are derived and compared to NUVEL-1. The first model, NUVEL-SZ, is derived from a data set that omits the 240 subduction zone slip vectors in the NUVEL-1 data set but includes all remaining 882 kinematic data. The velocities predicted by NUVEL-SZ and NUVEL-1 differ little, with maximum differences of only 1 mm yr-1 and 2-degrees along all plate boundaries except those surrounding the Caribbean plate. The second model NUVEL-G is derived from a data set that omits all 724 earthquake slip vectors but includes the remaining 277 spreading rates and 121 transform fault azimuths. Velocities predicted by NUVEL-G and NUVEL-1 also differ little, with maximum differences of 2 mm yr-1 and 4-degrees except for the Caribbean plate boundaries. These results indicate that even in the unlikely event that all earthquake slip vectors in the NUVEL-1 data set are unreliable recorders of long-term plate directions, the slip vectors do not significantly degrade the model. It thus appears that the decision to use or not to use earthquake slip vectors to derive a global plate motion model has little practical effect on the ability to derive an accurate description of present-day plate velocities. Little emphasis is placed on the changes in estimates of Caribbean plate velocities that occur upon exclusion of earthquake slip vectors; nearly all of the data from Caribbean plate boundaries are suspect and it is unlikely that the NUVEL-SZ or NUVEL-G models represent a significantly improved description of Caribbean plate velocities. Comparison of 677 slip vectors from transform faults to directions predicted by NUVEL-G shows statistically insignificant differences along 12 of 15 spreading centers. The good agreement between the nearly instantaneous estimates of slip directions provided by slip vectors and the longer-term average directions given by NUVEL-G suggest that in general, transform fault slip vectors parallel the longer-term average directions along transform faults. Interestingly, a statistically significant difference between Slip vectors from right-slipping and left-slipping transform faults is noted for nearly all spreading centers, with slip vectors along right-slipping and left-slipping faults rotated clockwise and counterclockwise, respectively, from the predicted direction. The cause of this bias is unknown, but may be related to biases introduced by unmodeled lateral heterogeneities in the mantle near transform faults.
