PDog - put the binders on... - GPS gets an updated position every 'n' seconds, then performs a delta, time / distance to derive speed. The accuracy of the calculation is also based on 0. 6 arc seconds of resolution too.
Speed is therefore the average of the diference in locations over time. If you travel through a curve, the shortest distance between two points is a straight line, although you followed a longer path to reach that point, and the reported speed will be slower than the actual speed.
Delta V speed calculation is referenced to a two dimensional spatial relationship in most consumer and aviation GPS units, because the required calculation to ascertain a three dimensional position take too long when compared to the delta V time desired for updating the indicated speed. Since this is the case, vertical speed is not taken into account.
Although the delta V error is small, it is there. Fly with good LORAN coverage (2 dimensional, horizontal TDOA), and good GPS (3 dimensional ) coverage, compare the speed reported between the units ( they rarely agree ), then descend at say 1200 feet per minute (yee haa!) but keep your airspeed the same!!, If you can do this your GPS and LORAN will both indicate a loss of about about 11 -12 knots in ground speed (offset for wind vector), but you know you are traveling at the same velocity as before. The LORAN will never catch up, the GPS might, but it takes several seconds, this is that 2D over 3D error. (DGPS Aviation units with VSI do not count here, they do use three dimensions, but in two different calculations, altitude with GPS is mathmatically tough)
Heisenbergs uncertainty principle also comes into effect, the only thing you can ever really be sure of is that you can never really be sure about anything. GPS processors are all single threaded, yet triangulation requires three seperate calculations, the first result is always wrong before the 3rd result is calculated, and therefore you can only know where you were but never where you are.
Generally, Altitude error is specified to be 1. 5 x Horizontal error specification. This means that the user of standard consumer GPS receivers should consider +/-23meters (75ft) with a DOP of 1 for 95% confidence. Altitude error is always considerably worse than the horizontal (position error). Much of this is a matter of geometry. With just four satellites, the "optimum" configuration for best overall accuracy is having the four SVs at 40 to 55 degrees above the horizon and one in each general direction N, E, W, and S. (Note: You will get a very BAD DOP if the SVs are at the exact same elevation. Luckily, this is a rare occurrence. )
The similar "best" arrangement for vertical position is with one SV overhead and the others at the horizon and 120 degrees in azimuth apart. Obviously, this arrangement is very poor from a signal standpoint. As a result of this geometry the calculated solution for altitude is not as accurate as it is for horizontal position. Almost any calibrated altimeter will be more stable at reading altitude than a GPS.
GPS altitude measures the users' distance from the center of the orbits. These measurements are referenced to geodetic altitude or ellipsoidal altitude in some GPS equipment. Garmin and most equipment manufacturers utilize a mathematical model, not a real calculation, in the GPS software which roughly approximates the geodetic model of the earth and reference altitude to this model. As with any model, there will be errors as the earth is not a simple mathematical shape to represent, the orbits are round but vary due to other bodies, the moon, sun, solar wind etc, and the earth is not round, it is an ellipsoid, and there are many models of the ellipsoid. The geodetic model of the earth can have much error at any specific point and Second, you have the GPS error itself to add in. As a result of this combined error, Vertical changes are not accurately represented in velocity calculations and are ignored, or stated as seperate vertical speed, at a lower determination of accuracy.
Also, your receiver clock is only periodicaly offset to match the satellitte time, atmospheric refraction and difraction causes un-solvable error without DGPS.
By the way, I'll bet many of you didn't know that GPS will not give speed over 1000 knots, and will not indicate over 18000 meters and 515 meters per second velocity, either or, but not both, or over 4 G's of acceleration, to keep Osama from using a garmin to guide missiles, by law.
Talk about off the topic, So how accurate can you be, and then, who really cares
Attention: TDR Forum Junkies 
If I can avoid speeding tickets, and keep from flying into jumbo jets thats accurate enough. BTW, I use GPS for . 1 meter survey work with way more confidence than anyone ever had with a chain and sticks. 
