Accuracy, precision, repeatability

[+user13371]

I see an awful lot of questions about how accurate a given GPS is, or if one model is better than another. I read a lot of marketing hype and personal anecdotes, but never a definitive answer based on large samplings of empirical data. Is there such a thing?

 

A simple test anyone can do occurs to me - and it requires almost NO understanding of GPS technology, math, circle of probable error, satellites, chipsets, or any other very technical blather.

 

1) Take your GPS to a known location. Set a waypoint. Go home.

2) Come back to the same spot another day at a different time. See how far away the GPS thinks you are.

 

Repeat step 2 with one location for several days at different times.

Repeat steps 1-2 with a few diverse locations.

 

Have two different GPS models? Repeat all of the above with each one, and you'll know which is better.

 

Tech-heads may now proceed to explain everything wrong with this method...

Edited September 18, 2008 by lee_rimar

[+NOV8TR]

Your approach is fine, but a bit over simplified. I have done exactly what you suggested, and yes, I have compared the results with a different GPSr (with a completely different chipset and software). The results were remarkably similar! I found only about 10 to 20 feet of error between units, and repeatability from one reading to the next. What I have noticed is a much larger variation in varying environments (trees, buildings, and the north side of hills). My WAAS unit (Garmin Etrex Vista HCx with V2.3 software and V2.3 firmware using the MTK chipset) is less accurate when the south WAAS bird is blocked. And my Non-WAAS unit (Tomtom ONE XL with navcore v7.1 software and the SIRF 3 chipset) is less accurate in deep tree cover and in between buildings (urban canyons). I happen to live in a pretty good area for GPSr reception, and I expect to have different results in areas that do not have good WAAS line of sight. After all is said and done, I generally consider either one of my GPSr's to be within 50 feet if they can see at least 8 satellites (assuming there is also a decent spread on their locations), and as good as 10 feet when they can see 12 satellites.

One other thing I did was lay down tracks (or breadcrumbs if you prefer), several days in a row and then reversed the track for several days (yes I have loaded aftermarket software on my Tomtom to do this, as it is not an "out of the box" feature for the Tomtom). This forced the use of different satellite constellations and varying line of sight conditions. Again I found both units to be pretty good as far as the track overlaying on previous days and directions. I can't say that I saw errors greater than 50 feet regardless of conditions or make/model GPSr.

Edited September 18, 2008 by 1XL-on-XR650L

[+StarBrand]

The spread/configuration of the sats overhead and signal strength vary considerably over time and affect the accuracy. The thickness of the atmosphere, sun spots, and many other factors can sublty alter the signal at any given time. The general average accuracy of our handheld units is around plus or minus 20-25 feet. I would expect to get a roughly circular area around 20 foot wide with about 50% - 70% of my samples if I did a few thousand at the same reading on my GPS.

 

Doesn't seem to much answewr anything about accuracy. Just repeatablity.