I thought I might wade into this a bit, not because I have any particular opinion of GPS cyclocomputers but because these issues are fairly well understood outside of the realm of cycling. I happen to be a biologist who uses GPS technology to remotely upload location of large mammals in forest (not anywhere near the ideal situation I, and others typically face on a bicycle).
What you must recall is the the jittering of locations observed on replicated tracks is due to sampling and the inherent measurement error present in recording data. We all make these error within acceptable limits when reading off a tape measure for instance. The US military used to add extra error, what was called selective availability...that all ended about a decade ago. The error that remains does exist and regardless of how minor it is must be accounted for. Depending on your needs this measurement error may be unacceptable (as we can see from some posts here). However there is absolutely no way around it. Some units may be better then others (or even phone vs cyclocomputers) but all will suffer....it is a inherent statistical property of sampling.
I have attached a paper I worked on some time ago where we were interested in the extent of this error. It was quite simple really, set out a collar in a known location and record thousands of locations. From this we could plot the X and Y errors and describe their statistical distribution. Unfortunately this may be difficult with cyclocomputers/phones which turn off when not moving.
Plotting 4 loops of a circuit will be completely insufficient to tease out the measurement error of the computer. Once you have a 100+ tracks with locations being obtained at the SAME ground location it would be as simple as taking the average X and Y locations and you would arrive as close to truth as possible. This is because the distributions are symmetric. This is the same reason that while being 14 feet off each location could theoretically put you in a different state within hours the errors do tend to even them selves out making the summarized ride statistics completely correct. The feeling of the data betting off is made worse by the fact that on the different laps the locations are probably not taken at the exact same true location and the computer connects the two data points with a straight line making the segments look different.
So if your ride data has issues or wonky segments remember that drawing from these distribution thousands of times over the course of a long ride will inevitably lead to a statistical outlier...its a statistical inevitability of sampling data
There are further papers in the wildlife literature that follow up on these effects and the influence they have....most are not that relatable here, save this, the questions you ask of the data can not be finer grained then the accuracy of your data.
Obviously this does "fix" the problems people may feel their units have but hopefully it may throw some light on the issues.
Sorry for the long winded post!!http://www.ualberta.ca/%7Evisscher/LINK ... er2005.pdf
" onclick="window.open(this.href);return false;