As I understand it the government imposed random (SA = selective availability) was turned off in 2000. Now the errors come in from atmospheric effects on signal travel time, number of satellites, signal loss due to obstructions, etc.
I Geocache in heavy terrain so the accuracy of my Garmin gets tested a lot. Using GPS, GLONAAS, and WAAS my typical repeatable accuracy is 8 ft in fairly thick cover, 16 ft in horrible cover, and 3 to 4 ft in open area assuming I let the GPS waypoint average for at least 5 minutes regardless of "%complete" it states.
'Survey units running RTK (like Trimble) where base unit waypoint averages all day and sends the mobile unit time correction signals easily get down to less than an inch.
'For reference when working with a surveyor friend of mine running a RTK setup we have found Google Earth maps to be 20 ft or more off when referencing to an obvious physical point (tree, fence post, etc). For that reason I never use Google Earth as my final method of determining coordinates if the measurement is critical.