How often you sample GPS position is a classic "fractal" geometry problem.
The example is usually a coastline.
Take a decent sized island, say Australia.
Use a 10km ruler to measure the length of the coast, then use a 1km ruler, a 100m ruler, a 1m ruler.
Each time you you will get a larger and larger distance.
I use an application called "MyTracks" on my tablet when I go walking.
The default setting is to take a GPS "fix" every 5m.
Because I go for a walk along the same forest path pretty much every day, I've tried 1,2,3,4,5,10,100 metre intervals.
The difference in total distance covered is amazing!
Under the settings menu there is a GPS accuracy option, the default is 200m, an "excellent GPS" - 10m, a "poor GPS" - 5000m.
The other thing I've noticed when riding my bike into Echunga, is up to a point the faster you are moving, the better the accuracy becomes.
When I've been in the passenger seat of a car, over about 30km/h, the track can be as much as 10-20m off to either side of the road!
Turning on the cell tower option brings that down to around 3-5m!
Still it's better than LORAN or that other thing that used to be around.
When I've had my Arduino on the dash of the car displaying the output from an EM-406 and an MPL115A1 barometer the speed and altitude are around
the same as my tablet.
Speed in particular from a GPS seems pretty good and tends to be close to the cars speedometer and odometer.
From the experiments I've been doing it would seem that GPS by itself is ok, but when you get data from another sensor things get better.
As always the more data you have to work with the more accurate position, altitude and speed get.