Horizontal Accuracy test
With a horizontal test I mean not comparing different modules against each other, but same modules against each other.
First change a battery. A DW-6630B was almost dead and we don't want it dead after one month. I forgot to reset, so it looked dead, but in fact the battery is complete full and fresh.
The GW-300, used to set all watches to atomic time.
Then I need them grouped in similar modules.
DW-069's and DW-6900. I consider these one of the coolest basic models, but they seem not very accurate.
DW-6600, also not particularly known about their accuracy.
Three different groups of DW-004/DW-6900E like models.
I forgot one G-2300, but now there are two basic blacks.
Masters of G: Gaussman, Frogman (GW-200) and Raysman. Very strange the similar models share two different modules
DW-003TL's normal and negative displays
New Mudman-s, I also included the GW-9000 with the auto receive turned off...
My Beloved DW-003's
And when all set to atomic clock, all put in a big box for 30 days.
All watches were set to Atomic time (using a GW-300) on November 23, 2006. A Month is defined as 30 days, so end of test will be on December 22*.
For a real good test you need a pretty large load of the same modules at the time, so specially the tests with a few modules are of course very doubtful. I believe I forgot to set the Gold Defender Frog. I should double check that one.
Is a big σ a problem? No of course not. Casio says often a model is + or - 15 or 30 seconds of per month (I guess it's their 2σ point), so a σ around 7.5 is acceptable for most models, for the 30 seconds of modules 15 is acceptable.
The watches were kept in a relative warm room (my studio with quiet some warm equipment), approx 23ºC). A watch worn on wrist is warmer, so this might effect the quartz crystal.
As I expected the DW-6600, DW-6900 and DW-5600 don't come out to good. The biggest contradiction is found in the DW-003TL section. It looks like the 1661 and the 1686 module differ more than only a 90º twist of the polarization.
So does this test actually have worth....
Actually I don't think so, it's just fun playing with the numbers. I only had a crappy calculator around, so I had to calculate most σ(n-1)'s with just a pen and paper. Casio admits there is a pretty large allowance of the accuracy, but on the other hand. 30 sec's off per month is an accuracy of 30/2626560*100 = 0.0011%!
NOTE: Updated Sunday December 24th. Readers noticed a miscalculation. A 30.4 day month doesn't have 2635200 but 2626560 seconds. I accidentally multiplied by 30.5. All numbers are recalculated on a proper calculator. There appeared to be a miscalculation in the 1597 section. The mean was no 16, but 20, which resulted in a significantly smaller σ(n-1).