- The length of the time series processed. Uncertainties decrease with increasing time series length, there is therfore a benefit in including as much data as possible to increase precision.
- The noise model assumed for calculating uncertainties. We know now that noise in GPS data is best fit by a model that combines white (random) and colored noise. The most recent results indicate that the colored noise correspond to "flicker noise" (spectral index of 1). Colored noise should be accounted for when estimating uncertainties.
- The geophysical models used in the GPS data processing. Most groups use similar models. Current developments that will allow the modeling of atmospheric and hydrological loading should improve precision in the near future.
- The strategy used to define a reference frame and the propagation of reference frame errors into the final velocities.
- The representation of uncertainties also depends on the level of confidence chosen -- this choice is an arbitrary decision that depends of the "risk" that the interpreter is ready to take. The same uncertainties shown at a 2-standard deviation confidence level (equivalent to a 5% risk level) will look much larger than at a 1-standard deviation confidence level (equivalent to a 61% risk level in 2-dimensions).