Yes I clearly missed a point there. I blame lack of sleep ... :-)

So, if the true track is a straight line then yes increased data samples will increase the distance recorded. If the random errors alternate to either side of the true track then the increase could be quite large. The example shown on sportypal makes sense: in that example the point error was 75% of the distance between samples resulting in an increase in distance recorded of 25% over the true distance. Increasing the distance between samples compared to the point error reduces the increase in distance. Ok I get it.

However, a couple of points.

Firstly, the error can only be positive. The distance recorded cannot be less than the true track.

Secondly, the software which is calculating the distances could well be smarter than the example given. It could be comparing multiple points and self-correcting.

Finally, does anyone know what the actual error point rate is on the various GPS chips ?

My SGS2 using either Strava or ViewRanger seems to produce tracks that fit very closely to the roads/tracks and it's certainly accurate enough for my purposes. Especially if it's giving me an extra mile in a hundred ... :-D

EDIT:

Of course the balancing issue is that if you take too few point samples then the GPS will 'cut the corners' and record less distance than the true track. This cannot increase the distance recorded though.

So increasing the sample rate can increase the distance recorded, but reducing the sample rate can reduce the distance recorded.

Finding the right sample rate to capture enough points but not introduce error will depend on knowing the point error in the GPS software I guess. With a point error of 0.3 metres (purely taken from the sportypal example) then taking data points every 0.7 metres limits the track increase to 8.5% (again taking the sportypal example, I think the maths is actually a little more complex, but it's illustrative).