Mark Neal reported some nice data in the thread Another GPS test.
I have just analysed 26 runners' GPS results for a half marathon.
The results show a statistically nice looking normal distribution shape, averaging 2.2% long with a standard deviation of 0.8%, some 25 to 50 times worse than the Calibrated Bike of the course measurer.
These runners' GPS results look about twice as bad as Mark Neal's smaller group - but it is a different course and a different sample of runners.
I have written up an account for public consumption here. Comments are welcome before I link it into the website where the GPS wearers were publishing their complaints.
Original Post