Mr. Ray wrote:
So while Polar may also be full of smart people, the ones doing the research are likely not the ones coding the simple algorithms, or they may be under tight staffing, schedule, and budget constraints, and in programming, it's easy to get error management wrong by an accumulation of decisions that the calculations are close enough.
Anyone that has worked with accelerometer data knew that the footpad idea was a lost cause from the getgo. I would bet that their algorithm ended up being to just count steps, and you "calibrate" it by essentially telling it how long your stride is. As bad as that is, it is probably still more accurate than trying to determine accumulated distance by integrating the acceleration signal you're getting, regardless of how fancy you get when you do that. The research people you mention probably told them that, and the marketing people decided to go ahead with it anyway.
The Garmin situation is different because there are some pretty simple, standard techniques you can use, such as implementing a real-time filter, that will almost certainly give a better estimate of the distance traveled than just connecting-the-dots.